Faster and more robust multi-person analysis (#85)

* tests synchro

* draft

* further draft

* affinity ok

* proposals okay, need to incorporate in Pose2Sim+tests

* will transfer sorting across frames in triangulation in next commit

* Lasts tests need to be done but seems to work pretty well

* should all work smoothly

* update readme

* last checks

* fixed linting issues

* getting tired of being forgetful

Pose2Sim/Pose2Sim.py

@@ -287,9 +287,6 @@ def synchronization(config=None):
         start = time.time()
         currentDateAndTime = datetime.now()
         project_dir = os.path.realpath(config_dict.get('project').get('project_dir'))
-        seq_name = os.path.basename(project_dir)
-        frame_range = config_dict.get('project').get('frame_range')
-        frames = ["all frames" if frame_range == [] else f"frames {frame_range[0]} to {frame_range[1]}"][0]
 
         logging.info("\n\n---------------------------------------------------------------------")
         logging.info("Camera synchronization")

Pose2Sim/S00_Demo_Session/Config.toml

@@ -18,9 +18,8 @@
 
 
 [project]
-# multi_person = false # true for trials with multiple participants. If false, only the main person in scene is analyzed (and it run much faster). 
+multi_person = false # If false, only the main person in scene is analyzed. 
-nb_persons_to_detect = 2 # checked only if multi_person is selected
+frame_rate = 60 # fps
-frame_rate = 120 # fps
 frame_range = [] # For example [10,300], or [] for all frames
 ## N.B.: If you want a time range instead, use frame_range = time_range * frame_rate
 ## For example if you want to analyze from 0.1 to 2 seconds with a 60 fps frame rate, 
@@ -31,6 +30,26 @@ exclude_from_batch = [] # List of trials to be excluded from batch analysis, ['<
 # Take heart, calibration is not that complicated once you get the hang of it!
 
 
+[pose]
+pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
+pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
+						      #With mediapipe: BLAZEPOSE. 
+						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
+						      #With deeplabcut: CUSTOM. See example at the end of the file.
+# What follows has not been implemented yet
+overwrite_pose = false
+openpose_path = '' # only checked if OpenPose is used
+
+
+[synchronization]
+display_corr = true # true or false (lowercase)
+reset_sync = true # Recalculate synchronization even if already done
+# id_kpt = [10] # keypoint ID, to be found in skeleton.py. Example RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
+# weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
+sync_frame_range = [] # For example [0,150], or [] for all frames (default)
+      # limit synchronization search (to the beginning or to the end of the capture for example)
+
+
 [calibration]
 calibration_type = 'convert' # 'convert' or 'calculate'
 
@@ -90,30 +109,17 @@ calibration_type = 'convert' # 'convert' or 'calculate'
          # Coming soon!
 
 
-[pose]
-pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
-pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
-						      #With mediapipe: BLAZEPOSE. 
-						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
-						      #With deeplabcut: CUSTOM. See example at the end of the file.
-# What follows has not been implemented yet
-overwrite_pose = false
-openpose_path = '' # only checked if OpenPose is used
-
-
-[synchronization]
-# COMING SOON! 
-reset_sync = true # Recalculate synchronization even if already done
-speed_kind = 'y' # 'y' showed best performance.
-id_kpt = [10] # number from keypoint name in skeleton.py. RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
-
-
 [personAssociation]
-tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+   likelihood_threshold_association = 0.3
-# and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+
-reproj_error_threshold_association = 20 # px
+   [personAssociation.single_person]
-likelihood_threshold_association = 0.3
+   reproj_error_threshold_association = 20 # px
+   tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+               # and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+   
+   [personAssociation.multi_person]
+   reconstruction_error_threshold = 0.1 # 0.1 = 10 cm
+   min_affinity = 0.2 # affinity below which a correspondence is ignored
 
 
 [triangulation]

Pose2Sim/S00_Demo_Session/S00_P00_SingleParticipant/Config.toml

@@ -16,84 +16,22 @@
 # [filtering.butterworth] and set cut_off_frequency = 10, etc.
 
 
-
 # [project]
-# multi_person = false # true for trials with multiple participants. If false, only the main person in scene is analyzed (and it run much faster). 
+# multi_person = true # If false, only the main person in scene is analyzed. 
-# nb_persons_to_detect = 2 # checked only if multi_person is selected
+# frame_rate = 60 # fps
-# frame_rate = 60 # FPS
 # frame_range = [] # For example [10,300], or [] for all frames
 ## N.B.: If you want a time range instead, use frame_range = time_range * frame_rate
 ## For example if you want to analyze from 0.1 to 2 seconds with a 60 fps frame rate, 
 ## frame_range = [0.1, 2.0]*frame_rate = [6, 120]
 
 # exclude_from_batch = [] # List of trials to be excluded from batch analysis, ['<participant_dir/trial_dir>', 'etc'].
-# e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
+## e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
-
-
 ## Take heart, calibration is not that complicated once you get the hang of it!
-# [calibration]
-# calibration_type = 'convert' # 'convert' or 'calculate'
-
-   # [calibration.convert]
-   # convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', or 'biocv'
-      # [calibration.convert.qualisys]
-      # binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
-      # [calibration.convert.optitrack]  # See readme for instructions
-      # [calibration.convert.vicon]      # No parameter needed
-      # [calibration.convert.opencap]    # No parameter needed
-      # [calibration.convert.easymocap]  # No parameter needed
-      # [calibration.convert.biocv]      # No parameter needed
-      # [calibration.convert.anipose]    # No parameter needed
-      # [calibration.convert.freemocap]  # No parameter needed
-  
-
-   # [calibration.calculate] 
-      ## Camera properties, theoretically need to be calculated only once in a camera lifetime
-      # [calibration.calculate.intrinsics]
-      # overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
-      # show_detection_intrinsics = true # true or false (lowercase)
-      # intrinsics_extension = 'jpg' # any video or image extension
-      # extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
-      # intrinsics_corners_nb = [4,7] 
-      # intrinsics_square_size = 60 # mm
-
-      ## Camera placements, need to be done before every session
-      # [calibration.calculate.extrinsics]
-      # extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
-      ## 'board' should be large enough to be detected when laid on the floor. Not recommended.
-      ## 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
-      ## 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
-
-      # moving_cameras = false # Not implemented yet
-      # calculate_extrinsics = true # true or false (lowercase) 
-
-         # [calibration.calculate.extrinsics.board]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         # extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
-         # extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
-
-         # [calibration.calculate.extrinsics.scene]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         ## list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
-         ## in m -> unlike for intrinsics, NOT in mm!
-         # object_coords_3d =   [[-2.0,  0.3,  0.0], 
-         #                      [-2.0 , 0.0,  0.0], 
-         #                      [-2.0, 0.0,  0.05], 
-         #                      [-2.0, -0.3 ,  0.0], 
-         #                      [0.0,  0.3,  0.0], 
-         #                      [0.0, 0.0,  0.0], 
-         #                      [0.0, 0.0,  0.05], 
-         #                      [0.0, -0.3,  0.0]]
-        
-         # [calibration.calculate.extrinsics.keypoints]
-         ## Coming soon!
 
 
 # [pose]
 # pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
-# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII,
+# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
 # 						      #With mediapipe: BLAZEPOSE. 
 # 						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
 # 						      #With deeplabcut: CUSTOM. See example at the end of the file.
@@ -103,33 +41,93 @@
 
 
 # [synchronization]
-## COMING SOON! 
+# display_corr = true # true or false (lowercase)
 # reset_sync = true # Recalculate synchronization even if already done
-# frames = [2850,3490] # Frames to use for synchronization, should point to a moment with fast motion.
+# # id_kpt = [10] # keypoint ID, to be found in skeleton.py. Example RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-# cut_off_frequency = 10 # cut-off frequency for a 4th order low-pass Butterworth filter
+# # weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
-## Vertical speeds (on X, Y, or Z axis, or 2D speeds)
+# sync_frame_range = [] # For example [0,150], or [] for all frames (default)
-# speed_kind = 'y' # 'x', 'y', 'z', or '2D'
+#       # limit synchronization search (to the beginning or to the end of the capture for example)
-# vmax = 20 # px/s
+
-# cam1_nb = 4
+
-# cam2_nb = 3
+# [calibration]
-# id_kpt = [9,10] # Pour plus tard aller chercher numéro depuis keypoint name dans skeleton.py. 'RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
+# calibration_type = 'convert' # 'convert' or 'calculate'
-# weights_kpt = [1,1] # Pris en compte uniquement si on a plusieurs keypoints
+
+#    [calibration.convert]
+#    convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', 'biocv', 'anipose', or 'freemocap'
+#       [calibration.convert.qualisys]
+#       binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
+#       [calibration.convert.optitrack]  # See readme for instructions
+#       [calibration.convert.vicon]      # No parameter needed
+#       [calibration.convert.opencap]    # No parameter needed
+#       [calibration.convert.easymocap]  # No parameter needed
+#       [calibration.convert.biocv]      # No parameter needed
+#       [calibration.convert.anipose]    # No parameter needed
+#       [calibration.convert.freemocap]  # No parameter needed
+  
+
+#    [calibration.calculate] 
+#       # Camera properties, theoretically need to be calculated only once in a camera lifetime
+#       [calibration.calculate.intrinsics]
+#       overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
+#       show_detection_intrinsics = true # true or false (lowercase)
+#       intrinsics_extension = 'jpg' # any video or image extension
+#       extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
+#       intrinsics_corners_nb = [4,7] 
+#       intrinsics_square_size = 60 # mm
+
+#       # Camera placements, need to be done before every session
+#       [calibration.calculate.extrinsics]
+#       calculate_extrinsics = true # true or false (lowercase) 
+#       extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
+#       # 'board' should be large enough to be detected when laid on the floor. Not recommended.
+#       # 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
+#       # 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
+#       moving_cameras = false # Not implemented yet
+
+#          [calibration.calculate.extrinsics.board]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
+#          extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
+
+#          [calibration.calculate.extrinsics.scene]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          # list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
+#          # in m -> unlike for intrinsics, NOT in mm!
+#          object_coords_3d =   [[-2.0,  0.3,  0.0], 
+#                               [-2.0 , 0.0,  0.0], 
+#                               [-2.0, 0.0,  0.05], 
+#                               [-2.0, -0.3 ,  0.0], 
+#                               [0.0,  0.3,  0.0], 
+#                               [0.0, 0.0,  0.0], 
+#                               [0.0, 0.0,  0.05], 
+#                               [0.0, -0.3,  0.0]]
+        
+#          [calibration.calculate.extrinsics.keypoints]
+#          # Coming soon!
 
 
 # [personAssociation]
-# # tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#    likelihood_threshold_association = 0.3
-## and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+
+#    [personAssociation.single_person]
 #    reproj_error_threshold_association = 20 # px
-# likelihood_threshold_association = 0.05
+#    tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#                # and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+   
+#    [personAssociation.multi_person]
+#    reconstruction_error_threshold = 0.1 # 0.1 = 10 cm
+#    min_affinity = 0.2 # affinity below which a correspondence is ignored
 
 
 # [triangulation]
 # reorder_trc = false # only checked if multi_person analysis
 # reproj_error_threshold_triangulation = 15 # px
-# likelihood_threshold_triangulation= 0.05
+# likelihood_threshold_triangulation= 0.3
 # min_cameras_for_triangulation = 2
 # interpolation = 'cubic' #linear, slinear, quadratic, cubic, or none
-## 'none' if you don't want to interpolate missing points
+#                         # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
@@ -141,26 +139,26 @@
 # type = 'butterworth' # butterworth, kalman, gaussian, LOESS, median, butterworth_on_speed
 # display_figures = false # true or false (lowercase)
 
-   # [filtering.butterworth]
+#    [filtering.butterworth]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 6 # Hz
+#    cut_off_frequency = 6 # Hz
-   # [filtering.kalman]
+#    [filtering.kalman]
-   ## How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
+#    # How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
-   # trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
+#    trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
-   # smooth = true # should be true, unless you need real-time filtering
+#    smooth = true # should be true, unless you need real-time filtering
-   # [filtering.butterworth_on_speed]
+#    [filtering.butterworth_on_speed]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 10 # Hz
+#    cut_off_frequency = 10 # Hz
-   # [filtering.gaussian]
+#    [filtering.gaussian]
-   # sigma_kernel = 2 #px
+#    sigma_kernel = 2 #px
-   # [filtering.LOESS]
+#    [filtering.LOESS]
-   # nb_values_used = 30 # = fraction of data used * nb frames
+#    nb_values_used = 30 # = fraction of data used * nb frames
-   # [filtering.median]
+#    [filtering.median]
-   # kernel_size = 9
+#    kernel_size = 9
 
 
 # [markerAugmentation] 
-# ## Only works on BODY_25 and BODY_25B models
+## Only works on BODY_25 and BODY_25B models
 # participant_height = 1.72 # m # float if single person, list of float if multi-person (same order as the Static trials)
 # participant_mass = 70.0 # kg
 
@@ -173,6 +171,7 @@
 # opensim_bin_path = 'C:\OpenSim 4.4\bin'
 
 
+
 ## CUSTOM skeleton, if you trained your own DeepLabCut model for example. 
 ## Make sure the node ids correspond to the column numbers of the 2D pose file, starting from zero.
 ## 
@@ -188,65 +187,65 @@
 # name = "CHip"
 # id = "None"
 #   [[pose.CUSTOM.children]]
-#   id = 12
 #   name = "RHip"
+#   id = 12
 #      [[pose.CUSTOM.children.children]]
-#      id = 14
 #      name = "RKnee"
+#      id = 14
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 16
 #         name = "RAnkle"
+#         id = 16
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 22
 #            name = "RBigToe"
+#            id = 22
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 23
 #               name = "RSmallToe"
+#               id = 23
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 24
 #            name = "RHeel"
+#            id = 24
 #   [[pose.CUSTOM.children]]
-#   id = 11
 #   name = "LHip"
+#   id = 11
 #      [[pose.CUSTOM.children.children]]
-#      id = 13
 #      name = "LKnee"
+#      id = 13
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 15
 #         name = "LAnkle"
+#         id = 15
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 19
 #            name = "LBigToe"
+#            id = 19
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 20
 #               name = "LSmallToe"
+#               id = 20
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 21
 #            name = "LHeel"
+#            id = 21
 #   [[pose.CUSTOM.children]]
-#   id = 17
 #   name = "Neck"
+#   id = 17
 #      [[pose.CUSTOM.children.children]]
-#      id = 18
 #      name = "Head"
+#      id = 18
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 0
 #         name = "Nose"
+#         id = 0
 #      [[pose.CUSTOM.children.children]]
-#      id = 6
 #      name = "RShoulder"
+#      id = 6
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 8
 #         name = "RElbow"
+#         id = 8
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 10
 #            name = "RWrist"
+#            id = 10
 #      [[pose.CUSTOM.children.children]]
-#      id = 5
 #      name = "LShoulder"
+#      id = 5
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 7
 #         name = "LElbow"
+#         id = 7
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 9
 #            name = "LWrist"
+#            id = 9

Pose2Sim/S00_Demo_Session/S00_P00_SingleParticipant/S00_P00_T00_StaticTrial/Config.toml

@@ -16,84 +16,22 @@
 # [filtering.butterworth] and set cut_off_frequency = 10, etc.
 
 
-
 # [project]
-# multi_person = false # true for trials with multiple participants. If false, only the main person in scene is analyzed (and it run much faster). 
+# multi_person = true # If false, only the main person in scene is analyzed. 
-# nb_persons_to_detect = 2 # checked only if multi_person is selected
+# frame_rate = 60 # fps
-# frame_rate = 60 # FPS
 # frame_range = [] # For example [10,300], or [] for all frames
 ## N.B.: If you want a time range instead, use frame_range = time_range * frame_rate
 ## For example if you want to analyze from 0.1 to 2 seconds with a 60 fps frame rate, 
 ## frame_range = [0.1, 2.0]*frame_rate = [6, 120]
 
 # exclude_from_batch = [] # List of trials to be excluded from batch analysis, ['<participant_dir/trial_dir>', 'etc'].
-# e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
+## e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
-
-
 ## Take heart, calibration is not that complicated once you get the hang of it!
-# [calibration]
-# calibration_type = 'convert' # 'convert' or 'calculate'
-
-   # [calibration.convert]
-   # convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', or 'biocv'
-      # [calibration.convert.qualisys]
-      # binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
-      # [calibration.convert.optitrack]  # See readme for instructions
-      # [calibration.convert.vicon]      # No parameter needed
-      # [calibration.convert.opencap]    # No parameter needed
-      # [calibration.convert.easymocap]  # No parameter needed
-      # [calibration.convert.biocv]      # No parameter needed
-      # [calibration.convert.anipose]    # No parameter needed
-      # [calibration.convert.freemocap]  # No parameter needed
-  
-
-   # [calibration.calculate] 
-      ## Camera properties, theoretically need to be calculated only once in a camera lifetime
-      # [calibration.calculate.intrinsics]
-      # overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
-      # show_detection_intrinsics = true # true or false (lowercase)
-      # intrinsics_extension = 'jpg' # any video or image extension
-      # extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
-      # intrinsics_corners_nb = [4,7] 
-      # intrinsics_square_size = 60 # mm
-
-      ## Camera placements, need to be done before every session
-      # [calibration.calculate.extrinsics]
-      # extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
-      ## 'board' should be large enough to be detected when laid on the floor. Not recommended.
-      ## 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
-      ## 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
-
-      # moving_cameras = false # Not implemented yet
-      # calculate_extrinsics = true # true or false (lowercase) 
-
-         # [calibration.calculate.extrinsics.board]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         # extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
-         # extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
-
-         # [calibration.calculate.extrinsics.scene]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         ## list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
-         ## in m -> unlike for intrinsics, NOT in mm!
-         # object_coords_3d =   [[-2.0,  0.3,  0.0], 
-         #                      [-2.0 , 0.0,  0.0], 
-         #                      [-2.0, 0.0,  0.05], 
-         #                      [-2.0, -0.3 ,  0.0], 
-         #                      [0.0,  0.3,  0.0], 
-         #                      [0.0, 0.0,  0.0], 
-         #                      [0.0, 0.0,  0.05], 
-         #                      [0.0, -0.3,  0.0]]
-        
-         # [calibration.calculate.extrinsics.keypoints]
-         ## Coming soon!
 
 
 # [pose]
 # pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
-# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII,
+# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
 # 						      #With mediapipe: BLAZEPOSE. 
 # 						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
 # 						      #With deeplabcut: CUSTOM. See example at the end of the file.
@@ -103,33 +41,93 @@
 
 
 # [synchronization]
-## COMING SOON! 
+# display_corr = true # true or false (lowercase)
 # reset_sync = true # Recalculate synchronization even if already done
-# frames = [2850,3490] # Frames to use for synchronization, should point to a moment with fast motion.
+# # id_kpt = [10] # keypoint ID, to be found in skeleton.py. Example RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-# cut_off_frequency = 10 # cut-off frequency for a 4th order low-pass Butterworth filter
+# # weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
-## Vertical speeds (on X, Y, or Z axis, or 2D speeds)
+# sync_frame_range = [] # For example [0,150], or [] for all frames (default)
-# speed_kind = 'y' # 'x', 'y', 'z', or '2D'
+#       # limit synchronization search (to the beginning or to the end of the capture for example)
-# vmax = 20 # px/s
+
-# cam1_nb = 4
+
-# cam2_nb = 3
+# [calibration]
-# id_kpt = [9,10] # Pour plus tard aller chercher numéro depuis keypoint name dans skeleton.py. 'RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
+# calibration_type = 'convert' # 'convert' or 'calculate'
-# weights_kpt = [1,1] # Pris en compte uniquement si on a plusieurs keypoints
+
+#    [calibration.convert]
+#    convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', 'biocv', 'anipose', or 'freemocap'
+#       [calibration.convert.qualisys]
+#       binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
+#       [calibration.convert.optitrack]  # See readme for instructions
+#       [calibration.convert.vicon]      # No parameter needed
+#       [calibration.convert.opencap]    # No parameter needed
+#       [calibration.convert.easymocap]  # No parameter needed
+#       [calibration.convert.biocv]      # No parameter needed
+#       [calibration.convert.anipose]    # No parameter needed
+#       [calibration.convert.freemocap]  # No parameter needed
+  
+
+#    [calibration.calculate] 
+#       # Camera properties, theoretically need to be calculated only once in a camera lifetime
+#       [calibration.calculate.intrinsics]
+#       overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
+#       show_detection_intrinsics = true # true or false (lowercase)
+#       intrinsics_extension = 'jpg' # any video or image extension
+#       extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
+#       intrinsics_corners_nb = [4,7] 
+#       intrinsics_square_size = 60 # mm
+
+#       # Camera placements, need to be done before every session
+#       [calibration.calculate.extrinsics]
+#       calculate_extrinsics = true # true or false (lowercase) 
+#       extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
+#       # 'board' should be large enough to be detected when laid on the floor. Not recommended.
+#       # 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
+#       # 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
+#       moving_cameras = false # Not implemented yet
+
+#          [calibration.calculate.extrinsics.board]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
+#          extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
+
+#          [calibration.calculate.extrinsics.scene]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          # list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
+#          # in m -> unlike for intrinsics, NOT in mm!
+#          object_coords_3d =   [[-2.0,  0.3,  0.0], 
+#                               [-2.0 , 0.0,  0.0], 
+#                               [-2.0, 0.0,  0.05], 
+#                               [-2.0, -0.3 ,  0.0], 
+#                               [0.0,  0.3,  0.0], 
+#                               [0.0, 0.0,  0.0], 
+#                               [0.0, 0.0,  0.05], 
+#                               [0.0, -0.3,  0.0]]
+        
+#          [calibration.calculate.extrinsics.keypoints]
+#          # Coming soon!
 
 
 # [personAssociation]
-# # tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#    likelihood_threshold_association = 0.3
-## and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+
+#    [personAssociation.single_person]
 #    reproj_error_threshold_association = 20 # px
-# likelihood_threshold_association = 0.05
+#    tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#                # and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+   
+#    [personAssociation.multi_person]
+#    reconstruction_error_threshold = 0.1 # 0.1 = 10 cm
+#    min_affinity = 0.2 # affinity below which a correspondence is ignored
 
 
 # [triangulation]
 # reorder_trc = false # only checked if multi_person analysis
 # reproj_error_threshold_triangulation = 15 # px
-# likelihood_threshold_triangulation= 0.05
+# likelihood_threshold_triangulation= 0.3
 # min_cameras_for_triangulation = 2
 # interpolation = 'cubic' #linear, slinear, quadratic, cubic, or none
-## 'none' if you don't want to interpolate missing points
+#                         # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
@@ -141,26 +139,26 @@
 # type = 'butterworth' # butterworth, kalman, gaussian, LOESS, median, butterworth_on_speed
 # display_figures = false # true or false (lowercase)
 
-   # [filtering.butterworth]
+#    [filtering.butterworth]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 6 # Hz
+#    cut_off_frequency = 6 # Hz
-   # [filtering.kalman]
+#    [filtering.kalman]
-   ## How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
+#    # How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
-   # trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
+#    trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
-   # smooth = true # should be true, unless you need real-time filtering
+#    smooth = true # should be true, unless you need real-time filtering
-   # [filtering.butterworth_on_speed]
+#    [filtering.butterworth_on_speed]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 10 # Hz
+#    cut_off_frequency = 10 # Hz
-   # [filtering.gaussian]
+#    [filtering.gaussian]
-   # sigma_kernel = 2 #px
+#    sigma_kernel = 2 #px
-   # [filtering.LOESS]
+#    [filtering.LOESS]
-   # nb_values_used = 30 # = fraction of data used * nb frames
+#    nb_values_used = 30 # = fraction of data used * nb frames
-   # [filtering.median]
+#    [filtering.median]
-   # kernel_size = 9
+#    kernel_size = 9
 
 
 # [markerAugmentation] 
-# ## Only works on BODY_25 and BODY_25B models
+## Only works on BODY_25 and BODY_25B models
 # participant_height = 1.72 # m # float if single person, list of float if multi-person (same order as the Static trials)
 # participant_mass = 70.0 # kg
 
@@ -173,6 +171,7 @@
 # opensim_bin_path = 'C:\OpenSim 4.4\bin'
 
 
+
 ## CUSTOM skeleton, if you trained your own DeepLabCut model for example. 
 ## Make sure the node ids correspond to the column numbers of the 2D pose file, starting from zero.
 ## 
@@ -188,65 +187,65 @@
 # name = "CHip"
 # id = "None"
 #   [[pose.CUSTOM.children]]
-#   id = 12
 #   name = "RHip"
+#   id = 12
 #      [[pose.CUSTOM.children.children]]
-#      id = 14
 #      name = "RKnee"
+#      id = 14
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 16
 #         name = "RAnkle"
+#         id = 16
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 22
 #            name = "RBigToe"
+#            id = 22
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 23
 #               name = "RSmallToe"
+#               id = 23
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 24
 #            name = "RHeel"
+#            id = 24
 #   [[pose.CUSTOM.children]]
-#   id = 11
 #   name = "LHip"
+#   id = 11
 #      [[pose.CUSTOM.children.children]]
-#      id = 13
 #      name = "LKnee"
+#      id = 13
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 15
 #         name = "LAnkle"
+#         id = 15
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 19
 #            name = "LBigToe"
+#            id = 19
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 20
 #               name = "LSmallToe"
+#               id = 20
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 21
 #            name = "LHeel"
+#            id = 21
 #   [[pose.CUSTOM.children]]
-#   id = 17
 #   name = "Neck"
+#   id = 17
 #      [[pose.CUSTOM.children.children]]
-#      id = 18
 #      name = "Head"
+#      id = 18
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 0
 #         name = "Nose"
+#         id = 0
 #      [[pose.CUSTOM.children.children]]
-#      id = 6
 #      name = "RShoulder"
+#      id = 6
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 8
 #         name = "RElbow"
+#         id = 8
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 10
 #            name = "RWrist"
+#            id = 10
 #      [[pose.CUSTOM.children.children]]
-#      id = 5
 #      name = "LShoulder"
+#      id = 5
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 7
 #         name = "LElbow"
+#         id = 7
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 9
 #            name = "LWrist"
+#            id = 9

Pose2Sim/S00_Demo_Session/S00_P00_SingleParticipant/S00_P00_T01_BalancingTrial/Config.toml

@@ -16,84 +16,22 @@
 # [filtering.butterworth] and set cut_off_frequency = 10, etc.
 
 
-
 # [project]
-# multi_person = false # true for trials with multiple participants. If false, only the main person in scene is analyzed (and it run much faster). 
+# multi_person = true # If false, only the main person in scene is analyzed. 
-# nb_persons_to_detect = 2 # checked only if multi_person is selected
+# frame_rate = 60 # fps
-# frame_rate = 60 # FPS
 # frame_range = [] # For example [10,300], or [] for all frames
 ## N.B.: If you want a time range instead, use frame_range = time_range * frame_rate
 ## For example if you want to analyze from 0.1 to 2 seconds with a 60 fps frame rate, 
 ## frame_range = [0.1, 2.0]*frame_rate = [6, 120]
 
 # exclude_from_batch = [] # List of trials to be excluded from batch analysis, ['<participant_dir/trial_dir>', 'etc'].
-# e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
+## e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
-
-
 ## Take heart, calibration is not that complicated once you get the hang of it!
-# [calibration]
-# calibration_type = 'convert' # 'convert' or 'calculate'
-
-   # [calibration.convert]
-   # convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', or 'biocv'
-      # [calibration.convert.qualisys]
-      # binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
-      # [calibration.convert.optitrack]  # See readme for instructions
-      # [calibration.convert.vicon]      # No parameter needed
-      # [calibration.convert.opencap]    # No parameter needed
-      # [calibration.convert.easymocap]  # No parameter needed
-      # [calibration.convert.biocv]      # No parameter needed
-      # [calibration.convert.anipose]    # No parameter needed
-      # [calibration.convert.freemocap]  # No parameter needed
-  
-
-   # [calibration.calculate] 
-      ## Camera properties, theoretically need to be calculated only once in a camera lifetime
-      # [calibration.calculate.intrinsics]
-      # overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
-      # show_detection_intrinsics = true # true or false (lowercase)
-      # intrinsics_extension = 'jpg' # any video or image extension
-      # extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
-      # intrinsics_corners_nb = [4,7] 
-      # intrinsics_square_size = 60 # mm
-
-      ## Camera placements, need to be done before every session
-      # [calibration.calculate.extrinsics]
-      # extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
-      ## 'board' should be large enough to be detected when laid on the floor. Not recommended.
-      ## 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
-      ## 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
-
-      # moving_cameras = false # Not implemented yet
-      # calculate_extrinsics = true # true or false (lowercase) 
-
-         # [calibration.calculate.extrinsics.board]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         # extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
-         # extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
-
-         # [calibration.calculate.extrinsics.scene]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         ## list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
-         ## in m -> unlike for intrinsics, NOT in mm!
-         # object_coords_3d =   [[-2.0,  0.3,  0.0], 
-         #                      [-2.0 , 0.0,  0.0], 
-         #                      [-2.0, 0.0,  0.05], 
-         #                      [-2.0, -0.3 ,  0.0], 
-         #                      [0.0,  0.3,  0.0], 
-         #                      [0.0, 0.0,  0.0], 
-         #                      [0.0, 0.0,  0.05], 
-         #                      [0.0, -0.3,  0.0]]
-        
-         # [calibration.calculate.extrinsics.keypoints]
-         ## Coming soon!
 
 
 # [pose]
 # pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
-# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII,
+# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
 # 						      #With mediapipe: BLAZEPOSE. 
 # 						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
 # 						      #With deeplabcut: CUSTOM. See example at the end of the file.
@@ -103,33 +41,93 @@
 
 
 # [synchronization]
-## COMING SOON! 
+# display_corr = true # true or false (lowercase)
 # reset_sync = true # Recalculate synchronization even if already done
-# frames = [2850,3490] # Frames to use for synchronization, should point to a moment with fast motion.
+# # id_kpt = [10] # keypoint ID, to be found in skeleton.py. Example RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-# cut_off_frequency = 10 # cut-off frequency for a 4th order low-pass Butterworth filter
+# # weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
-## Vertical speeds (on X, Y, or Z axis, or 2D speeds)
+# sync_frame_range = [] # For example [0,150], or [] for all frames (default)
-# speed_kind = 'y' # 'x', 'y', 'z', or '2D'
+#       # limit synchronization search (to the beginning or to the end of the capture for example)
-# vmax = 20 # px/s
+
-# cam1_nb = 4
+
-# cam2_nb = 3
+# [calibration]
-# id_kpt = [9,10] # Pour plus tard aller chercher numéro depuis keypoint name dans skeleton.py. 'RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
+# calibration_type = 'convert' # 'convert' or 'calculate'
-# weights_kpt = [1,1] # Pris en compte uniquement si on a plusieurs keypoints
+
+#    [calibration.convert]
+#    convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', 'biocv', 'anipose', or 'freemocap'
+#       [calibration.convert.qualisys]
+#       binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
+#       [calibration.convert.optitrack]  # See readme for instructions
+#       [calibration.convert.vicon]      # No parameter needed
+#       [calibration.convert.opencap]    # No parameter needed
+#       [calibration.convert.easymocap]  # No parameter needed
+#       [calibration.convert.biocv]      # No parameter needed
+#       [calibration.convert.anipose]    # No parameter needed
+#       [calibration.convert.freemocap]  # No parameter needed
+  
+
+#    [calibration.calculate] 
+#       # Camera properties, theoretically need to be calculated only once in a camera lifetime
+#       [calibration.calculate.intrinsics]
+#       overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
+#       show_detection_intrinsics = true # true or false (lowercase)
+#       intrinsics_extension = 'jpg' # any video or image extension
+#       extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
+#       intrinsics_corners_nb = [4,7] 
+#       intrinsics_square_size = 60 # mm
+
+#       # Camera placements, need to be done before every session
+#       [calibration.calculate.extrinsics]
+#       calculate_extrinsics = true # true or false (lowercase) 
+#       extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
+#       # 'board' should be large enough to be detected when laid on the floor. Not recommended.
+#       # 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
+#       # 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
+#       moving_cameras = false # Not implemented yet
+
+#          [calibration.calculate.extrinsics.board]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
+#          extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
+
+#          [calibration.calculate.extrinsics.scene]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          # list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
+#          # in m -> unlike for intrinsics, NOT in mm!
+#          object_coords_3d =   [[-2.0,  0.3,  0.0], 
+#                               [-2.0 , 0.0,  0.0], 
+#                               [-2.0, 0.0,  0.05], 
+#                               [-2.0, -0.3 ,  0.0], 
+#                               [0.0,  0.3,  0.0], 
+#                               [0.0, 0.0,  0.0], 
+#                               [0.0, 0.0,  0.05], 
+#                               [0.0, -0.3,  0.0]]
+        
+#          [calibration.calculate.extrinsics.keypoints]
+#          # Coming soon!
 
 
 # [personAssociation]
-# # tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#    likelihood_threshold_association = 0.3
-## and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+
+#    [personAssociation.single_person]
 #    reproj_error_threshold_association = 20 # px
-# likelihood_threshold_association = 0.05
+#    tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#                # and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+   
+#    [personAssociation.multi_person]
+#    reconstruction_error_threshold = 0.1 # 0.1 = 10 cm
+#    min_affinity = 0.2 # affinity below which a correspondence is ignored
 
 
 # [triangulation]
 # reorder_trc = false # only checked if multi_person analysis
 # reproj_error_threshold_triangulation = 15 # px
-# likelihood_threshold_triangulation= 0.05
+# likelihood_threshold_triangulation= 0.3
 # min_cameras_for_triangulation = 2
 # interpolation = 'cubic' #linear, slinear, quadratic, cubic, or none
-## 'none' if you don't want to interpolate missing points
+#                         # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
@@ -139,28 +137,28 @@
 
 [filtering]
 # type = 'butterworth' # butterworth, kalman, gaussian, LOESS, median, butterworth_on_speed
-display_figures = true # true or false (lowercase)
+display_figures = false # true or false (lowercase)
 
-   # [filtering.butterworth]
+#    [filtering.butterworth]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 6 # Hz
+#    cut_off_frequency = 6 # Hz
-   # [filtering.kalman]
+#    [filtering.kalman]
-   ## How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
+#    # How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
-   # trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
+#    trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
-   # smooth = true # should be true, unless you need real-time filtering
+#    smooth = true # should be true, unless you need real-time filtering
-   # [filtering.butterworth_on_speed]
+#    [filtering.butterworth_on_speed]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 10 # Hz
+#    cut_off_frequency = 10 # Hz
-   # [filtering.gaussian]
+#    [filtering.gaussian]
-   # sigma_kernel = 2 #px
+#    sigma_kernel = 2 #px
-   # [filtering.LOESS]
+#    [filtering.LOESS]
-   # nb_values_used = 30 # = fraction of data used * nb frames
+#    nb_values_used = 30 # = fraction of data used * nb frames
-   # [filtering.median]
+#    [filtering.median]
-   # kernel_size = 9
+#    kernel_size = 9
 
 
 # [markerAugmentation] 
-# ## Only works on BODY_25 and BODY_25B models
+## Only works on BODY_25 and BODY_25B models
 # participant_height = 1.72 # m # float if single person, list of float if multi-person (same order as the Static trials)
 # participant_mass = 70.0 # kg
 
@@ -173,6 +171,7 @@ display_figures = true # true or false (lowercase)
 # opensim_bin_path = 'C:\OpenSim 4.4\bin'
 
 
+
 ## CUSTOM skeleton, if you trained your own DeepLabCut model for example. 
 ## Make sure the node ids correspond to the column numbers of the 2D pose file, starting from zero.
 ## 
@@ -188,65 +187,65 @@ display_figures = true # true or false (lowercase)
 # name = "CHip"
 # id = "None"
 #   [[pose.CUSTOM.children]]
-#   id = 12
 #   name = "RHip"
+#   id = 12
 #      [[pose.CUSTOM.children.children]]
-#      id = 14
 #      name = "RKnee"
+#      id = 14
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 16
 #         name = "RAnkle"
+#         id = 16
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 22
 #            name = "RBigToe"
+#            id = 22
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 23
 #               name = "RSmallToe"
+#               id = 23
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 24
 #            name = "RHeel"
+#            id = 24
 #   [[pose.CUSTOM.children]]
-#   id = 11
 #   name = "LHip"
+#   id = 11
 #      [[pose.CUSTOM.children.children]]
-#      id = 13
 #      name = "LKnee"
+#      id = 13
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 15
 #         name = "LAnkle"
+#         id = 15
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 19
 #            name = "LBigToe"
+#            id = 19
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 20
 #               name = "LSmallToe"
+#               id = 20
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 21
 #            name = "LHeel"
+#            id = 21
 #   [[pose.CUSTOM.children]]
-#   id = 17
 #   name = "Neck"
+#   id = 17
 #      [[pose.CUSTOM.children.children]]
-#      id = 18
 #      name = "Head"
+#      id = 18
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 0
 #         name = "Nose"
+#         id = 0
 #      [[pose.CUSTOM.children.children]]
-#      id = 6
 #      name = "RShoulder"
+#      id = 6
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 8
 #         name = "RElbow"
+#         id = 8
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 10
 #            name = "RWrist"
+#            id = 10
 #      [[pose.CUSTOM.children.children]]
-#      id = 5
 #      name = "LShoulder"
+#      id = 5
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 7
 #         name = "LElbow"
+#         id = 7
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 9
 #            name = "LWrist"
+#            id = 9

Pose2Sim/S00_Demo_Session/S00_P01_MultiParticipants/Config.toml

@@ -16,84 +16,22 @@
 # [filtering.butterworth] and set cut_off_frequency = 10, etc.
 
 
-
 # [project]
-# multi_person = false # true for trials with multiple participants. If false, only the main person in scene is analyzed (and it run much faster). 
+# multi_person = true # If false, only the main person in scene is analyzed. 
-# nb_persons_to_detect = 2 # checked only if multi_person is selected
+# frame_rate = 60 # fps
-# frame_rate = 60 # FPS
 # frame_range = [] # For example [10,300], or [] for all frames
 ## N.B.: If you want a time range instead, use frame_range = time_range * frame_rate
 ## For example if you want to analyze from 0.1 to 2 seconds with a 60 fps frame rate, 
 ## frame_range = [0.1, 2.0]*frame_rate = [6, 120]
 
 # exclude_from_batch = [] # List of trials to be excluded from batch analysis, ['<participant_dir/trial_dir>', 'etc'].
-# e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
+## e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
 ## Take heart, calibration is not that complicated once you get the hang of it!
 
 
-# [calibration]
-# calibration_type = 'convert' # 'convert' or 'calculate'
-
-   # [calibration.convert]
-   # convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', or 'biocv'
-      # [calibration.convert.qualisys]
-      # binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
-      # [calibration.convert.optitrack]  # See readme for instructions
-      # [calibration.convert.vicon]      # No parameter needed
-      # [calibration.convert.opencap]    # No parameter needed
-      # [calibration.convert.easymocap]  # No parameter needed
-      # [calibration.convert.biocv]      # No parameter needed
-      # [calibration.convert.anipose]    # No parameter needed
-      # [calibration.convert.freemocap]  # No parameter needed
-  
-
-   # [calibration.calculate] 
-      ## Camera properties, theoretically need to be calculated only once in a camera lifetime
-      # [calibration.calculate.intrinsics]
-      # overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
-      # show_detection_intrinsics = true # true or false (lowercase)
-      # intrinsics_extension = 'jpg' # any video or image extension
-      # extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
-      # intrinsics_corners_nb = [4,7] 
-      # intrinsics_square_size = 60 # mm
-
-      ## Camera placements, need to be done before every session
-      # [calibration.calculate.extrinsics]
-      # extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
-      ## 'board' should be large enough to be detected when laid on the floor. Not recommended.
-      ## 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
-      ## 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
-
-      # moving_cameras = false # Not implemented yet
-      # calculate_extrinsics = true # true or false (lowercase) 
-
-         # [calibration.calculate.extrinsics.board]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         # extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
-         # extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
-
-         # [calibration.calculate.extrinsics.scene]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         ## list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
-         ## in m -> unlike for intrinsics, NOT in mm!
-         # object_coords_3d =   [[-2.0,  0.3,  0.0], 
-         #                      [-2.0 , 0.0,  0.0], 
-         #                      [-2.0, 0.0,  0.05], 
-         #                      [-2.0, -0.3 ,  0.0], 
-         #                      [0.0,  0.3,  0.0], 
-         #                      [0.0, 0.0,  0.0], 
-         #                      [0.0, 0.0,  0.05], 
-         #                      [0.0, -0.3,  0.0]]
-        
-         # [calibration.calculate.extrinsics.keypoints]
-         ## Coming soon!
-
-
 # [pose]
 # pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
-# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII,
+# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
 # 						      #With mediapipe: BLAZEPOSE. 
 # 						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
 # 						      #With deeplabcut: CUSTOM. See example at the end of the file.
@@ -103,33 +41,93 @@
 
 
 # [synchronization]
-## COMING SOON! 
+# display_corr = true # true or false (lowercase)
 # reset_sync = true # Recalculate synchronization even if already done
-# frames = [2850,3490] # Frames to use for synchronization, should point to a moment with fast motion.
+# # id_kpt = [10] # keypoint ID, to be found in skeleton.py. Example RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-# cut_off_frequency = 10 # cut-off frequency for a 4th order low-pass Butterworth filter
+# # weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
-## Vertical speeds (on X, Y, or Z axis, or 2D speeds)
+# sync_frame_range = [] # For example [0,150], or [] for all frames (default)
-# speed_kind = 'y' # 'x', 'y', 'z', or '2D'
+#       # limit synchronization search (to the beginning or to the end of the capture for example)
-# vmax = 20 # px/s
+
-# cam1_nb = 4
+
-# cam2_nb = 3
+# [calibration]
-# id_kpt = [9,10] # Pour plus tard aller chercher numéro depuis keypoint name dans skeleton.py. 'RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
+# calibration_type = 'convert' # 'convert' or 'calculate'
-# weights_kpt = [1,1] # Pris en compte uniquement si on a plusieurs keypoints
+
+#    [calibration.convert]
+#    convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', 'biocv', 'anipose', or 'freemocap'
+#       [calibration.convert.qualisys]
+#       binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
+#       [calibration.convert.optitrack]  # See readme for instructions
+#       [calibration.convert.vicon]      # No parameter needed
+#       [calibration.convert.opencap]    # No parameter needed
+#       [calibration.convert.easymocap]  # No parameter needed
+#       [calibration.convert.biocv]      # No parameter needed
+#       [calibration.convert.anipose]    # No parameter needed
+#       [calibration.convert.freemocap]  # No parameter needed
+  
+
+#    [calibration.calculate] 
+#       # Camera properties, theoretically need to be calculated only once in a camera lifetime
+#       [calibration.calculate.intrinsics]
+#       overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
+#       show_detection_intrinsics = true # true or false (lowercase)
+#       intrinsics_extension = 'jpg' # any video or image extension
+#       extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
+#       intrinsics_corners_nb = [4,7] 
+#       intrinsics_square_size = 60 # mm
+
+#       # Camera placements, need to be done before every session
+#       [calibration.calculate.extrinsics]
+#       calculate_extrinsics = true # true or false (lowercase) 
+#       extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
+#       # 'board' should be large enough to be detected when laid on the floor. Not recommended.
+#       # 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
+#       # 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
+#       moving_cameras = false # Not implemented yet
+
+#          [calibration.calculate.extrinsics.board]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
+#          extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
+
+#          [calibration.calculate.extrinsics.scene]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          # list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
+#          # in m -> unlike for intrinsics, NOT in mm!
+#          object_coords_3d =   [[-2.0,  0.3,  0.0], 
+#                               [-2.0 , 0.0,  0.0], 
+#                               [-2.0, 0.0,  0.05], 
+#                               [-2.0, -0.3 ,  0.0], 
+#                               [0.0,  0.3,  0.0], 
+#                               [0.0, 0.0,  0.0], 
+#                               [0.0, 0.0,  0.05], 
+#                               [0.0, -0.3,  0.0]]
+        
+#          [calibration.calculate.extrinsics.keypoints]
+#          # Coming soon!
 
 
 # [personAssociation]
-# # tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#    likelihood_threshold_association = 0.3
-## and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+
+#    [personAssociation.single_person]
 #    reproj_error_threshold_association = 20 # px
-# likelihood_threshold_association = 0.05
+#    tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#                # and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+   
+#    [personAssociation.multi_person]
+#    reconstruction_error_threshold = 0.1 # 0.1 = 10 cm
+#    min_affinity = 0.2 # affinity below which a correspondence is ignored
 
 
 # [triangulation]
 # reorder_trc = false # only checked if multi_person analysis
 # reproj_error_threshold_triangulation = 15 # px
-# likelihood_threshold_triangulation= 0.05
+# likelihood_threshold_triangulation= 0.3
 # min_cameras_for_triangulation = 2
 # interpolation = 'cubic' #linear, slinear, quadratic, cubic, or none
-## 'none' if you don't want to interpolate missing points
+#                         # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
@@ -141,28 +139,28 @@
 # type = 'butterworth' # butterworth, kalman, gaussian, LOESS, median, butterworth_on_speed
 # display_figures = false # true or false (lowercase)
 
-   # [filtering.butterworth]
+#    [filtering.butterworth]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 6 # Hz
+#    cut_off_frequency = 6 # Hz
-   # [filtering.kalman]
+#    [filtering.kalman]
-   ## How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
+#    # How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
-   # trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
+#    trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
-   # smooth = true # should be true, unless you need real-time filtering
+#    smooth = true # should be true, unless you need real-time filtering
-   # [filtering.butterworth_on_speed]
+#    [filtering.butterworth_on_speed]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 10 # Hz
+#    cut_off_frequency = 10 # Hz
-   # [filtering.gaussian]
+#    [filtering.gaussian]
-   # sigma_kernel = 2 #px
+#    sigma_kernel = 2 #px
-   # [filtering.LOESS]
+#    [filtering.LOESS]
-   # nb_values_used = 30 # = fraction of data used * nb frames
+#    nb_values_used = 30 # = fraction of data used * nb frames
-   # [filtering.median]
+#    [filtering.median]
-   # kernel_size = 9
+#    kernel_size = 9
 
 
 # [markerAugmentation] 
-# ## Only works on BODY_25 and BODY_25B models
+## Only works on BODY_25 and BODY_25B models
-# participant_height = [1.21, 1.72] # m # float if single person, list of float if multi-person (same order as the Static trials)
+# participant_height = 1.72 # m # float if single person, list of float if multi-person (same order as the Static trials)
-# participant_mass = [25.0, 70.0] # kg
+# participant_mass = 70.0 # kg
 
 
 # [opensim]
@@ -173,6 +171,7 @@
 # opensim_bin_path = 'C:\OpenSim 4.4\bin'
 
 
+
 ## CUSTOM skeleton, if you trained your own DeepLabCut model for example. 
 ## Make sure the node ids correspond to the column numbers of the 2D pose file, starting from zero.
 ## 
@@ -188,65 +187,65 @@
 # name = "CHip"
 # id = "None"
 #   [[pose.CUSTOM.children]]
-#   id = 12
 #   name = "RHip"
+#   id = 12
 #      [[pose.CUSTOM.children.children]]
-#      id = 14
 #      name = "RKnee"
+#      id = 14
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 16
 #         name = "RAnkle"
+#         id = 16
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 22
 #            name = "RBigToe"
+#            id = 22
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 23
 #               name = "RSmallToe"
+#               id = 23
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 24
 #            name = "RHeel"
+#            id = 24
 #   [[pose.CUSTOM.children]]
-#   id = 11
 #   name = "LHip"
+#   id = 11
 #      [[pose.CUSTOM.children.children]]
-#      id = 13
 #      name = "LKnee"
+#      id = 13
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 15
 #         name = "LAnkle"
+#         id = 15
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 19
 #            name = "LBigToe"
+#            id = 19
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 20
 #               name = "LSmallToe"
+#               id = 20
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 21
 #            name = "LHeel"
+#            id = 21
 #   [[pose.CUSTOM.children]]
-#   id = 17
 #   name = "Neck"
+#   id = 17
 #      [[pose.CUSTOM.children.children]]
-#      id = 18
 #      name = "Head"
+#      id = 18
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 0
 #         name = "Nose"
+#         id = 0
 #      [[pose.CUSTOM.children.children]]
-#      id = 6
 #      name = "RShoulder"
+#      id = 6
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 8
 #         name = "RElbow"
+#         id = 8
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 10
 #            name = "RWrist"
+#            id = 10
 #      [[pose.CUSTOM.children.children]]
-#      id = 5
 #      name = "LShoulder"
+#      id = 5
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 7
 #         name = "LElbow"
+#         id = 7
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 9
 #            name = "LWrist"
+#            id = 9

Pose2Sim/S00_Demo_Session/S00_P01_MultiParticipants/S00_P01_T00_StaticTrialParticipant1/Config.toml

@@ -16,84 +16,22 @@
 # [filtering.butterworth] and set cut_off_frequency = 10, etc.
 
 
-
 # [project]
-# multi_person = false # true for trials with multiple participants. If false, only the main person in scene is analyzed (and it run much faster). 
+# multi_person = true # If false, only the main person in scene is analyzed. 
-# nb_persons_to_detect = 2 # checked only if multi_person is selected
+# frame_rate = 60 # fps
-# frame_rate = 60 # FPS
 # frame_range = [] # For example [10,300], or [] for all frames
 ## N.B.: If you want a time range instead, use frame_range = time_range * frame_rate
 ## For example if you want to analyze from 0.1 to 2 seconds with a 60 fps frame rate, 
 ## frame_range = [0.1, 2.0]*frame_rate = [6, 120]
 
 # exclude_from_batch = [] # List of trials to be excluded from batch analysis, ['<participant_dir/trial_dir>', 'etc'].
-# e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
+## e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
-
-
 ## Take heart, calibration is not that complicated once you get the hang of it!
-# [calibration]
-# calibration_type = 'convert' # 'convert' or 'calculate'
-
-   # [calibration.convert]
-   # convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', or 'biocv'
-      # [calibration.convert.qualisys]
-      # binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
-      # [calibration.convert.optitrack]  # See readme for instructions
-      # [calibration.convert.vicon]      # No parameter needed
-      # [calibration.convert.opencap]    # No parameter needed
-      # [calibration.convert.easymocap]  # No parameter needed
-      # [calibration.convert.biocv]      # No parameter needed
-      # [calibration.convert.anipose]    # No parameter needed
-      # [calibration.convert.freemocap]  # No parameter needed
-  
-
-   # [calibration.calculate] 
-      ## Camera properties, theoretically need to be calculated only once in a camera lifetime
-      # [calibration.calculate.intrinsics]
-      # overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
-      # show_detection_intrinsics = true # true or false (lowercase)
-      # intrinsics_extension = 'jpg' # any video or image extension
-      # extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
-      # intrinsics_corners_nb = [4,7] 
-      # intrinsics_square_size = 60 # mm
-
-      ## Camera placements, need to be done before every session
-      # [calibration.calculate.extrinsics]
-      # extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
-      ## 'board' should be large enough to be detected when laid on the floor. Not recommended.
-      ## 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
-      ## 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
-
-      # moving_cameras = false # Not implemented yet
-      # calculate_extrinsics = true # true or false (lowercase) 
-
-         # [calibration.calculate.extrinsics.board]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         # extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
-         # extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
-
-         # [calibration.calculate.extrinsics.scene]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         ## list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
-         ## in m -> unlike for intrinsics, NOT in mm!
-         # object_coords_3d =   [[-2.0,  0.3,  0.0], 
-         #                      [-2.0 , 0.0,  0.0], 
-         #                      [-2.0, 0.0,  0.05], 
-         #                      [-2.0, -0.3 ,  0.0], 
-         #                      [0.0,  0.3,  0.0], 
-         #                      [0.0, 0.0,  0.0], 
-         #                      [0.0, 0.0,  0.05], 
-         #                      [0.0, -0.3,  0.0]]
-        
-         # [calibration.calculate.extrinsics.keypoints]
-         ## Coming soon!
 
 
 # [pose]
 # pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
-# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII,
+# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
 # 						      #With mediapipe: BLAZEPOSE. 
 # 						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
 # 						      #With deeplabcut: CUSTOM. See example at the end of the file.
@@ -103,33 +41,93 @@
 
 
 # [synchronization]
-## COMING SOON! 
+# display_corr = true # true or false (lowercase)
 # reset_sync = true # Recalculate synchronization even if already done
-# frames = [2850,3490] # Frames to use for synchronization, should point to a moment with fast motion.
+# # id_kpt = [10] # keypoint ID, to be found in skeleton.py. Example RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-# cut_off_frequency = 10 # cut-off frequency for a 4th order low-pass Butterworth filter
+# # weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
-## Vertical speeds (on X, Y, or Z axis, or 2D speeds)
+# sync_frame_range = [] # For example [0,150], or [] for all frames (default)
-# speed_kind = 'y' # 'x', 'y', 'z', or '2D'
+#       # limit synchronization search (to the beginning or to the end of the capture for example)
-# vmax = 20 # px/s
+
-# cam1_nb = 4
+
-# cam2_nb = 3
+# [calibration]
-# id_kpt = [9,10] # Pour plus tard aller chercher numéro depuis keypoint name dans skeleton.py. 'RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
+# calibration_type = 'convert' # 'convert' or 'calculate'
-# weights_kpt = [1,1] # Pris en compte uniquement si on a plusieurs keypoints
+
+#    [calibration.convert]
+#    convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', 'biocv', 'anipose', or 'freemocap'
+#       [calibration.convert.qualisys]
+#       binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
+#       [calibration.convert.optitrack]  # See readme for instructions
+#       [calibration.convert.vicon]      # No parameter needed
+#       [calibration.convert.opencap]    # No parameter needed
+#       [calibration.convert.easymocap]  # No parameter needed
+#       [calibration.convert.biocv]      # No parameter needed
+#       [calibration.convert.anipose]    # No parameter needed
+#       [calibration.convert.freemocap]  # No parameter needed
+  
+
+#    [calibration.calculate] 
+#       # Camera properties, theoretically need to be calculated only once in a camera lifetime
+#       [calibration.calculate.intrinsics]
+#       overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
+#       show_detection_intrinsics = true # true or false (lowercase)
+#       intrinsics_extension = 'jpg' # any video or image extension
+#       extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
+#       intrinsics_corners_nb = [4,7] 
+#       intrinsics_square_size = 60 # mm
+
+#       # Camera placements, need to be done before every session
+#       [calibration.calculate.extrinsics]
+#       calculate_extrinsics = true # true or false (lowercase) 
+#       extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
+#       # 'board' should be large enough to be detected when laid on the floor. Not recommended.
+#       # 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
+#       # 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
+#       moving_cameras = false # Not implemented yet
+
+#          [calibration.calculate.extrinsics.board]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
+#          extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
+
+#          [calibration.calculate.extrinsics.scene]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          # list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
+#          # in m -> unlike for intrinsics, NOT in mm!
+#          object_coords_3d =   [[-2.0,  0.3,  0.0], 
+#                               [-2.0 , 0.0,  0.0], 
+#                               [-2.0, 0.0,  0.05], 
+#                               [-2.0, -0.3 ,  0.0], 
+#                               [0.0,  0.3,  0.0], 
+#                               [0.0, 0.0,  0.0], 
+#                               [0.0, 0.0,  0.05], 
+#                               [0.0, -0.3,  0.0]]
+        
+#          [calibration.calculate.extrinsics.keypoints]
+#          # Coming soon!
 
 
 # [personAssociation]
-# # tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#    likelihood_threshold_association = 0.3
-## and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+
+#    [personAssociation.single_person]
 #    reproj_error_threshold_association = 20 # px
-# likelihood_threshold_association = 0.05
+#    tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#                # and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+   
+#    [personAssociation.multi_person]
+#    reconstruction_error_threshold = 0.1 # 0.1 = 10 cm
+#    min_affinity = 0.2 # affinity below which a correspondence is ignored
 
 
 # [triangulation]
 # reorder_trc = false # only checked if multi_person analysis
 # reproj_error_threshold_triangulation = 15 # px
-# likelihood_threshold_triangulation= 0.05
+# likelihood_threshold_triangulation= 0.3
 # min_cameras_for_triangulation = 2
 # interpolation = 'cubic' #linear, slinear, quadratic, cubic, or none
-## 'none' if you don't want to interpolate missing points
+#                         # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
@@ -141,26 +139,26 @@
 # type = 'butterworth' # butterworth, kalman, gaussian, LOESS, median, butterworth_on_speed
 # display_figures = false # true or false (lowercase)
 
-   # [filtering.butterworth]
+#    [filtering.butterworth]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 6 # Hz
+#    cut_off_frequency = 6 # Hz
-   # [filtering.kalman]
+#    [filtering.kalman]
-   ## How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
+#    # How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
-   # trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
+#    trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
-   # smooth = true # should be true, unless you need real-time filtering
+#    smooth = true # should be true, unless you need real-time filtering
-   # [filtering.butterworth_on_speed]
+#    [filtering.butterworth_on_speed]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 10 # Hz
+#    cut_off_frequency = 10 # Hz
-   # [filtering.gaussian]
+#    [filtering.gaussian]
-   # sigma_kernel = 2 #px
+#    sigma_kernel = 2 #px
-   # [filtering.LOESS]
+#    [filtering.LOESS]
-   # nb_values_used = 30 # = fraction of data used * nb frames
+#    nb_values_used = 30 # = fraction of data used * nb frames
-   # [filtering.median]
+#    [filtering.median]
-   # kernel_size = 9
+#    kernel_size = 9
 
 
 [markerAugmentation] 
-# ## Only works on BODY_25 and BODY_25B models
+## Only works on BODY_25 and BODY_25B models
 participant_height = 1.21 # m # float if single person, list of float if multi-person (same order as the Static trials)
 participant_mass = 25.0 # kg
 
@@ -173,6 +171,7 @@ participant_mass = 25.0 # kg
 # opensim_bin_path = 'C:\OpenSim 4.4\bin'
 
 
+
 ## CUSTOM skeleton, if you trained your own DeepLabCut model for example. 
 ## Make sure the node ids correspond to the column numbers of the 2D pose file, starting from zero.
 ## 
@@ -188,65 +187,65 @@ participant_mass = 25.0 # kg
 # name = "CHip"
 # id = "None"
 #   [[pose.CUSTOM.children]]
-#   id = 12
 #   name = "RHip"
+#   id = 12
 #      [[pose.CUSTOM.children.children]]
-#      id = 14
 #      name = "RKnee"
+#      id = 14
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 16
 #         name = "RAnkle"
+#         id = 16
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 22
 #            name = "RBigToe"
+#            id = 22
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 23
 #               name = "RSmallToe"
+#               id = 23
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 24
 #            name = "RHeel"
+#            id = 24
 #   [[pose.CUSTOM.children]]
-#   id = 11
 #   name = "LHip"
+#   id = 11
 #      [[pose.CUSTOM.children.children]]
-#      id = 13
 #      name = "LKnee"
+#      id = 13
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 15
 #         name = "LAnkle"
+#         id = 15
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 19
 #            name = "LBigToe"
+#            id = 19
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 20
 #               name = "LSmallToe"
+#               id = 20
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 21
 #            name = "LHeel"
+#            id = 21
 #   [[pose.CUSTOM.children]]
-#   id = 17
 #   name = "Neck"
+#   id = 17
 #      [[pose.CUSTOM.children.children]]
-#      id = 18
 #      name = "Head"
+#      id = 18
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 0
 #         name = "Nose"
+#         id = 0
 #      [[pose.CUSTOM.children.children]]
-#      id = 6
 #      name = "RShoulder"
+#      id = 6
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 8
 #         name = "RElbow"
+#         id = 8
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 10
 #            name = "RWrist"
+#            id = 10
 #      [[pose.CUSTOM.children.children]]
-#      id = 5
 #      name = "LShoulder"
+#      id = 5
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 7
 #         name = "LElbow"
+#         id = 7
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 9
 #            name = "LWrist"
+#            id = 9

Pose2Sim/S00_Demo_Session/S00_P01_MultiParticipants/S00_P01_T01_StaticTrialParticipant2/Config.toml

@@ -16,84 +16,22 @@
 # [filtering.butterworth] and set cut_off_frequency = 10, etc.
 
 
-
 # [project]
-# multi_person = false # true for trials with multiple participants. If false, only the main person in scene is analyzed (and it run much faster). 
+# multi_person = true # If false, only the main person in scene is analyzed. 
-# nb_persons_to_detect = 2 # checked only if multi_person is selected
+# frame_rate = 60 # fps
-# frame_rate = 60 # FPS
 # frame_range = [] # For example [10,300], or [] for all frames
 ## N.B.: If you want a time range instead, use frame_range = time_range * frame_rate
 ## For example if you want to analyze from 0.1 to 2 seconds with a 60 fps frame rate, 
 ## frame_range = [0.1, 2.0]*frame_rate = [6, 120]
 
 # exclude_from_batch = [] # List of trials to be excluded from batch analysis, ['<participant_dir/trial_dir>', 'etc'].
-# e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
+## e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
-
-
 ## Take heart, calibration is not that complicated once you get the hang of it!
-# [calibration]
-# calibration_type = 'convert' # 'convert' or 'calculate'
-
-   # [calibration.convert]
-   # convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', or 'biocv'
-      # [calibration.convert.qualisys]
-      # binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
-      # [calibration.convert.optitrack]  # See readme for instructions
-      # [calibration.convert.vicon]      # No parameter needed
-      # [calibration.convert.opencap]    # No parameter needed
-      # [calibration.convert.easymocap]  # No parameter needed
-      # [calibration.convert.biocv]      # No parameter needed
-      # [calibration.convert.anipose]    # No parameter needed
-      # [calibration.convert.freemocap]  # No parameter needed
-  
-
-   # [calibration.calculate] 
-      ## Camera properties, theoretically need to be calculated only once in a camera lifetime
-      # [calibration.calculate.intrinsics]
-      # overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
-      # show_detection_intrinsics = true # true or false (lowercase)
-      # intrinsics_extension = 'jpg' # any video or image extension
-      # extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
-      # intrinsics_corners_nb = [4,7] 
-      # intrinsics_square_size = 60 # mm
-
-      ## Camera placements, need to be done before every session
-      # [calibration.calculate.extrinsics]
-      # extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
-      ## 'board' should be large enough to be detected when laid on the floor. Not recommended.
-      ## 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
-      ## 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
-
-      # moving_cameras = false # Not implemented yet
-      # calculate_extrinsics = true # true or false (lowercase) 
-
-         # [calibration.calculate.extrinsics.board]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         # extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
-         # extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
-
-         # [calibration.calculate.extrinsics.scene]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         ## list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
-         ## in m -> unlike for intrinsics, NOT in mm!
-         # object_coords_3d =   [[-2.0,  0.3,  0.0], 
-         #                      [-2.0 , 0.0,  0.0], 
-         #                      [-2.0, 0.0,  0.05], 
-         #                      [-2.0, -0.3 ,  0.0], 
-         #                      [0.0,  0.3,  0.0], 
-         #                      [0.0, 0.0,  0.0], 
-         #                      [0.0, 0.0,  0.05], 
-         #                      [0.0, -0.3,  0.0]]
-        
-         # [calibration.calculate.extrinsics.keypoints]
-         ## Coming soon!
 
 
 # [pose]
 # pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
-# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII,
+# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
 # 						      #With mediapipe: BLAZEPOSE. 
 # 						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
 # 						      #With deeplabcut: CUSTOM. See example at the end of the file.
@@ -103,33 +41,93 @@
 
 
 # [synchronization]
-## COMING SOON! 
+# display_corr = true # true or false (lowercase)
 # reset_sync = true # Recalculate synchronization even if already done
-# frames = [2850,3490] # Frames to use for synchronization, should point to a moment with fast motion.
+# # id_kpt = [10] # keypoint ID, to be found in skeleton.py. Example RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-# cut_off_frequency = 10 # cut-off frequency for a 4th order low-pass Butterworth filter
+# # weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
-## Vertical speeds (on X, Y, or Z axis, or 2D speeds)
+# sync_frame_range = [] # For example [0,150], or [] for all frames (default)
-# speed_kind = 'y' # 'x', 'y', 'z', or '2D'
+#       # limit synchronization search (to the beginning or to the end of the capture for example)
-# vmax = 20 # px/s
+
-# cam1_nb = 4
+
-# cam2_nb = 3
+# [calibration]
-# id_kpt = [9,10] # Pour plus tard aller chercher numéro depuis keypoint name dans skeleton.py. 'RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
+# calibration_type = 'convert' # 'convert' or 'calculate'
-# weights_kpt = [1,1] # Pris en compte uniquement si on a plusieurs keypoints
+
+#    [calibration.convert]
+#    convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', 'biocv', 'anipose', or 'freemocap'
+#       [calibration.convert.qualisys]
+#       binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
+#       [calibration.convert.optitrack]  # See readme for instructions
+#       [calibration.convert.vicon]      # No parameter needed
+#       [calibration.convert.opencap]    # No parameter needed
+#       [calibration.convert.easymocap]  # No parameter needed
+#       [calibration.convert.biocv]      # No parameter needed
+#       [calibration.convert.anipose]    # No parameter needed
+#       [calibration.convert.freemocap]  # No parameter needed
+  
+
+#    [calibration.calculate] 
+#       # Camera properties, theoretically need to be calculated only once in a camera lifetime
+#       [calibration.calculate.intrinsics]
+#       overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
+#       show_detection_intrinsics = true # true or false (lowercase)
+#       intrinsics_extension = 'jpg' # any video or image extension
+#       extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
+#       intrinsics_corners_nb = [4,7] 
+#       intrinsics_square_size = 60 # mm
+
+#       # Camera placements, need to be done before every session
+#       [calibration.calculate.extrinsics]
+#       calculate_extrinsics = true # true or false (lowercase) 
+#       extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
+#       # 'board' should be large enough to be detected when laid on the floor. Not recommended.
+#       # 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
+#       # 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
+#       moving_cameras = false # Not implemented yet
+
+#          [calibration.calculate.extrinsics.board]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
+#          extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
+
+#          [calibration.calculate.extrinsics.scene]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          # list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
+#          # in m -> unlike for intrinsics, NOT in mm!
+#          object_coords_3d =   [[-2.0,  0.3,  0.0], 
+#                               [-2.0 , 0.0,  0.0], 
+#                               [-2.0, 0.0,  0.05], 
+#                               [-2.0, -0.3 ,  0.0], 
+#                               [0.0,  0.3,  0.0], 
+#                               [0.0, 0.0,  0.0], 
+#                               [0.0, 0.0,  0.05], 
+#                               [0.0, -0.3,  0.0]]
+        
+#          [calibration.calculate.extrinsics.keypoints]
+#          # Coming soon!
 
 
 # [personAssociation]
-# # tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#    likelihood_threshold_association = 0.3
-## and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+
+#    [personAssociation.single_person]
 #    reproj_error_threshold_association = 20 # px
-# likelihood_threshold_association = 0.05
+#    tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#                # and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+   
+#    [personAssociation.multi_person]
+#    reconstruction_error_threshold = 0.1 # 0.1 = 10 cm
+#    min_affinity = 0.2 # affinity below which a correspondence is ignored
 
 
 # [triangulation]
 # reorder_trc = false # only checked if multi_person analysis
 # reproj_error_threshold_triangulation = 15 # px
-# likelihood_threshold_triangulation= 0.05
+# likelihood_threshold_triangulation= 0.3
 # min_cameras_for_triangulation = 2
 # interpolation = 'cubic' #linear, slinear, quadratic, cubic, or none
-## 'none' if you don't want to interpolate missing points
+#                         # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
@@ -141,26 +139,26 @@
 # type = 'butterworth' # butterworth, kalman, gaussian, LOESS, median, butterworth_on_speed
 # display_figures = false # true or false (lowercase)
 
-   # [filtering.butterworth]
+#    [filtering.butterworth]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 6 # Hz
+#    cut_off_frequency = 6 # Hz
-   # [filtering.kalman]
+#    [filtering.kalman]
-   ## How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
+#    # How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
-   # trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
+#    trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
-   # smooth = true # should be true, unless you need real-time filtering
+#    smooth = true # should be true, unless you need real-time filtering
-   # [filtering.butterworth_on_speed]
+#    [filtering.butterworth_on_speed]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 10 # Hz
+#    cut_off_frequency = 10 # Hz
-   # [filtering.gaussian]
+#    [filtering.gaussian]
-   # sigma_kernel = 2 #px
+#    sigma_kernel = 2 #px
-   # [filtering.LOESS]
+#    [filtering.LOESS]
-   # nb_values_used = 30 # = fraction of data used * nb frames
+#    nb_values_used = 30 # = fraction of data used * nb frames
-   # [filtering.median]
+#    [filtering.median]
-   # kernel_size = 9
+#    kernel_size = 9
 
 
 [markerAugmentation] 
-# ## Only works on BODY_25 and BODY_25B models
+## Only works on BODY_25 and BODY_25B models
 participant_height = 1.72 # m # float if single person, list of float if multi-person (same order as the Static trials)
 participant_mass = 70.0 # kg
 
@@ -173,6 +171,7 @@ participant_mass = 70.0 # kg
 # opensim_bin_path = 'C:\OpenSim 4.4\bin'
 
 
+
 ## CUSTOM skeleton, if you trained your own DeepLabCut model for example. 
 ## Make sure the node ids correspond to the column numbers of the 2D pose file, starting from zero.
 ## 
@@ -188,65 +187,65 @@ participant_mass = 70.0 # kg
 # name = "CHip"
 # id = "None"
 #   [[pose.CUSTOM.children]]
-#   id = 12
 #   name = "RHip"
+#   id = 12
 #      [[pose.CUSTOM.children.children]]
-#      id = 14
 #      name = "RKnee"
+#      id = 14
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 16
 #         name = "RAnkle"
+#         id = 16
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 22
 #            name = "RBigToe"
+#            id = 22
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 23
 #               name = "RSmallToe"
+#               id = 23
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 24
 #            name = "RHeel"
+#            id = 24
 #   [[pose.CUSTOM.children]]
-#   id = 11
 #   name = "LHip"
+#   id = 11
 #      [[pose.CUSTOM.children.children]]
-#      id = 13
 #      name = "LKnee"
+#      id = 13
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 15
 #         name = "LAnkle"
+#         id = 15
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 19
 #            name = "LBigToe"
+#            id = 19
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 20
 #               name = "LSmallToe"
+#               id = 20
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 21
 #            name = "LHeel"
+#            id = 21
 #   [[pose.CUSTOM.children]]
-#   id = 17
 #   name = "Neck"
+#   id = 17
 #      [[pose.CUSTOM.children.children]]
-#      id = 18
 #      name = "Head"
+#      id = 18
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 0
 #         name = "Nose"
+#         id = 0
 #      [[pose.CUSTOM.children.children]]
-#      id = 6
 #      name = "RShoulder"
+#      id = 6
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 8
 #         name = "RElbow"
+#         id = 8
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 10
 #            name = "RWrist"
+#            id = 10
 #      [[pose.CUSTOM.children.children]]
-#      id = 5
 #      name = "LShoulder"
+#      id = 5
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 7
 #         name = "LElbow"
+#         id = 7
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 9
 #            name = "LWrist"
+#            id = 9

Pose2Sim/S00_Demo_Session/S00_P01_MultiParticipants/S00_P01_T02_Participants1-2/Config.toml

@@ -16,84 +16,22 @@
 # [filtering.butterworth] and set cut_off_frequency = 10, etc.
 
 
-
 [project]
-multi_person = true # true for trials with multiple participants. If false, only the main person in scene is analyzed (and it run much faster). 
+multi_person = true # If false, only the main person in scene is analyzed. 
-nb_persons_to_detect = 2 # checked only if multi_person is selected
+# frame_rate = 60 # fps
-# frame_rate = 60 # FPS
 # frame_range = [] # For example [10,300], or [] for all frames
 ## N.B.: If you want a time range instead, use frame_range = time_range * frame_rate
 ## For example if you want to analyze from 0.1 to 2 seconds with a 60 fps frame rate, 
 ## frame_range = [0.1, 2.0]*frame_rate = [6, 120]
 
 # exclude_from_batch = [] # List of trials to be excluded from batch analysis, ['<participant_dir/trial_dir>', 'etc'].
-# e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
+## e.g. ['S00_P00_Participant/S00_P00_T00_StaticTrial', 'S00_P00_Participant/S00_P00_T01_BalancingTrial']
-
-
 ## Take heart, calibration is not that complicated once you get the hang of it!
-# [calibration]
-# calibration_type = 'convert' # 'convert' or 'calculate'
-
-   # [calibration.convert]
-   # convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', or 'biocv'
-      # [calibration.convert.qualisys]
-      # binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
-      # [calibration.convert.optitrack]  # See readme for instructions
-      # [calibration.convert.vicon]      # No parameter needed
-      # [calibration.convert.opencap]    # No parameter needed
-      # [calibration.convert.easymocap]  # No parameter needed
-      # [calibration.convert.biocv]      # No parameter needed
-      # [calibration.convert.anipose]    # No parameter needed
-      # [calibration.convert.freemocap]  # No parameter needed
-  
-
-   # [calibration.calculate] 
-      ## Camera properties, theoretically need to be calculated only once in a camera lifetime
-      # [calibration.calculate.intrinsics]
-      # overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
-      # show_detection_intrinsics = true # true or false (lowercase)
-      # intrinsics_extension = 'jpg' # any video or image extension
-      # extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
-      # intrinsics_corners_nb = [4,7] 
-      # intrinsics_square_size = 60 # mm
-
-      ## Camera placements, need to be done before every session
-      # [calibration.calculate.extrinsics]
-      # extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
-      ## 'board' should be large enough to be detected when laid on the floor. Not recommended.
-      ## 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
-      ## 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
-
-      # moving_cameras = false # Not implemented yet
-      # calculate_extrinsics = true # true or false (lowercase) 
-
-         # [calibration.calculate.extrinsics.board]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         # extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
-         # extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
-
-         # [calibration.calculate.extrinsics.scene]
-         # show_reprojection_error = true # true or false (lowercase)
-         # extrinsics_extension = 'png' # any video or image extension
-         ## list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
-         ## in m -> unlike for intrinsics, NOT in mm!
-         # object_coords_3d =   [[-2.0,  0.3,  0.0], 
-         #                      [-2.0 , 0.0,  0.0], 
-         #                      [-2.0, 0.0,  0.05], 
-         #                      [-2.0, -0.3 ,  0.0], 
-         #                      [0.0,  0.3,  0.0], 
-         #                      [0.0, 0.0,  0.0], 
-         #                      [0.0, 0.0,  0.05], 
-         #                      [0.0, -0.3,  0.0]]
-        
-         # [calibration.calculate.extrinsics.keypoints]
-         ## Coming soon!
 
 
 # [pose]
 # pose_framework = 'openpose' # 'openpose', 'mediapipe', 'alphapose', 'deeplabcut' 
-# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII,
+# pose_model = 'BODY_25B' #With openpose: BODY_25B, BODY_25, BODY_135, COCO, MPII
 # 						      #With mediapipe: BLAZEPOSE. 
 # 						      #With alphapose: HALPE_26, HALPE_68, HALPE_136, COCO_133. 
 # 						      #With deeplabcut: CUSTOM. See example at the end of the file.
@@ -103,33 +41,93 @@ nb_persons_to_detect = 2 # checked only if multi_person is selected
 
 
 # [synchronization]
-## COMING SOON! 
+# display_corr = true # true or false (lowercase)
 # reset_sync = true # Recalculate synchronization even if already done
-# frames = [2850,3490] # Frames to use for synchronization, should point to a moment with fast motion.
+# # id_kpt = [10] # keypoint ID, to be found in skeleton.py. Example RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-# cut_off_frequency = 10 # cut-off frequency for a 4th order low-pass Butterworth filter
+# # weights_kpt = [1] # Only taken into account if you have several keypoints (Currently only one keypoint is supported).
-## Vertical speeds (on X, Y, or Z axis, or 2D speeds)
+# sync_frame_range = [] # For example [0,150], or [] for all frames (default)
-# speed_kind = 'y' # 'x', 'y', 'z', or '2D'
+#       # limit synchronization search (to the beginning or to the end of the capture for example)
-# vmax = 20 # px/s
+
-# cam1_nb = 4
+
-# cam2_nb = 3
+# [calibration]
-# id_kpt = [9,10] # Pour plus tard aller chercher numéro depuis keypoint name dans skeleton.py. 'RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
+# calibration_type = 'convert' # 'convert' or 'calculate'
-# weights_kpt = [1,1] # Pris en compte uniquement si on a plusieurs keypoints
+
+#    [calibration.convert]
+#    convert_from = 'qualisys' # 'qualisys', 'optitrack', vicon', 'opencap', 'easymocap', 'biocv', 'anipose', or 'freemocap'
+#       [calibration.convert.qualisys]
+#       binning_factor = 1 # Usually 1, except when filming in 540p where it usually is 2
+#       [calibration.convert.optitrack]  # See readme for instructions
+#       [calibration.convert.vicon]      # No parameter needed
+#       [calibration.convert.opencap]    # No parameter needed
+#       [calibration.convert.easymocap]  # No parameter needed
+#       [calibration.convert.biocv]      # No parameter needed
+#       [calibration.convert.anipose]    # No parameter needed
+#       [calibration.convert.freemocap]  # No parameter needed
+  
+
+#    [calibration.calculate] 
+#       # Camera properties, theoretically need to be calculated only once in a camera lifetime
+#       [calibration.calculate.intrinsics]
+#       overwrite_intrinsics = false # overwrite (or not) if they have already been calculated?
+#       show_detection_intrinsics = true # true or false (lowercase)
+#       intrinsics_extension = 'jpg' # any video or image extension
+#       extract_every_N_sec = 1 # if video, extract frames every N seconds (can be <1 )
+#       intrinsics_corners_nb = [4,7] 
+#       intrinsics_square_size = 60 # mm
+
+#       # Camera placements, need to be done before every session
+#       [calibration.calculate.extrinsics]
+#       calculate_extrinsics = true # true or false (lowercase) 
+#       extrinsics_method = 'scene' # 'board', 'scene', 'keypoints'
+#       # 'board' should be large enough to be detected when laid on the floor. Not recommended.
+#       # 'scene' involves manually clicking any point of know coordinates on scene. Usually more accurate if points are spread out.
+#       # 'keypoints' uses automatic pose estimation of a person freely walking and waving arms in the scene. Slighlty less accurate, requires synchronized cameras.
+#       moving_cameras = false # Not implemented yet
+
+#          [calibration.calculate.extrinsics.board]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          extrinsics_corners_nb = [4,7] # [H,W] rather than [w,h]
+#          extrinsics_square_size = 60 # mm # [h,w] if square is actually a rectangle
+
+#          [calibration.calculate.extrinsics.scene]
+#          show_reprojection_error = true # true or false (lowercase)
+#          extrinsics_extension = 'png' # any video or image extension
+#          # list of 3D coordinates to be manually labelled on images. Can also be a 2 dimensional plane. 
+#          # in m -> unlike for intrinsics, NOT in mm!
+#          object_coords_3d =   [[-2.0,  0.3,  0.0], 
+#                               [-2.0 , 0.0,  0.0], 
+#                               [-2.0, 0.0,  0.05], 
+#                               [-2.0, -0.3 ,  0.0], 
+#                               [0.0,  0.3,  0.0], 
+#                               [0.0, 0.0,  0.0], 
+#                               [0.0, 0.0,  0.05], 
+#                               [0.0, -0.3,  0.0]]
+        
+#          [calibration.calculate.extrinsics.keypoints]
+#          # Coming soon!
 
 
 # [personAssociation]
-# tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#    likelihood_threshold_association = 0.3
-## and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+
+#    [personAssociation.single_person]
 #    reproj_error_threshold_association = 20 # px
-# likelihood_threshold_association = 0.05
+#    tracked_keypoint = 'Neck' # If the neck is not detected by the pose_model, check skeleton.py 
+#                # and choose a stable point for tracking the person of interest (e.g., 'right_shoulder' with BLAZEPOSE)
+   
+#    [personAssociation.multi_person]
+#    reconstruction_error_threshold = 0.1 # 0.1 = 10 cm
+#    min_affinity = 0.2 # affinity below which a correspondence is ignored
 
 
 [triangulation]
-reorder_trc = true # only checked if multi_person analysis
+reorder_trc = false # only checked if multi_person analysis
 # reproj_error_threshold_triangulation = 15 # px
-# likelihood_threshold_triangulation= 0.05
+# likelihood_threshold_triangulation= 0.3
 # min_cameras_for_triangulation = 2
 # interpolation = 'cubic' #linear, slinear, quadratic, cubic, or none
-## 'none' if you don't want to interpolate missing points
+#                         # 'none' if you don't want to interpolate missing points
 # interp_if_gap_smaller_than = 10 # do not interpolate bigger gaps
 # show_interp_indices = true # true or false (lowercase). For each keypoint, return the frames that need to be interpolated
 # handle_LR_swap = false # Better if few cameras (eg less than 4) with risk of limb swapping (eg camera facing sagittal plane), otherwise slightly less accurate and slower
@@ -141,22 +139,22 @@ reorder_trc = true # only checked if multi_person analysis
 # type = 'butterworth' # butterworth, kalman, gaussian, LOESS, median, butterworth_on_speed
 # display_figures = false # true or false (lowercase)
 
-   # [filtering.butterworth]
+#    [filtering.butterworth]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 6 # Hz
+#    cut_off_frequency = 6 # Hz
-   # [filtering.kalman]
+#    [filtering.kalman]
-   ## How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
+#    # How much more do you trust triangulation results (measurements), than previous data (process assuming constant acceleration)?
-   # trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
+#    trust_ratio = 100 # = measurement_trust/process_trust ~= process_noise/measurement_noise
-   # smooth = true # should be true, unless you need real-time filtering
+#    smooth = true # should be true, unless you need real-time filtering
-   # [filtering.butterworth_on_speed]
+#    [filtering.butterworth_on_speed]
-   # order = 4 
+#    order = 4 
-   # cut_off_frequency = 10 # Hz
+#    cut_off_frequency = 10 # Hz
-   # [filtering.gaussian]
+#    [filtering.gaussian]
-   # sigma_kernel = 2 #px
+#    sigma_kernel = 2 #px
-   # [filtering.LOESS]
+#    [filtering.LOESS]
-   # nb_values_used = 30 # = fraction of data used * nb frames
+#    nb_values_used = 30 # = fraction of data used * nb frames
-   # [filtering.median]
+#    [filtering.median]
-   # kernel_size = 9
+#    kernel_size = 9
 
 
 [markerAugmentation] 
@@ -173,6 +171,7 @@ participant_mass = [25.0, 70.0] # kg
 # opensim_bin_path = 'C:\OpenSim 4.4\bin'
 
 
+
 ## CUSTOM skeleton, if you trained your own DeepLabCut model for example. 
 ## Make sure the node ids correspond to the column numbers of the 2D pose file, starting from zero.
 ## 
@@ -188,65 +187,65 @@ participant_mass = [25.0, 70.0] # kg
 # name = "CHip"
 # id = "None"
 #   [[pose.CUSTOM.children]]
-#   id = 12
 #   name = "RHip"
+#   id = 12
 #      [[pose.CUSTOM.children.children]]
-#      id = 14
 #      name = "RKnee"
+#      id = 14
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 16
 #         name = "RAnkle"
+#         id = 16
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 22
 #            name = "RBigToe"
+#            id = 22
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 23
 #               name = "RSmallToe"
+#               id = 23
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 24
 #            name = "RHeel"
+#            id = 24
 #   [[pose.CUSTOM.children]]
-#   id = 11
 #   name = "LHip"
+#   id = 11
 #      [[pose.CUSTOM.children.children]]
-#      id = 13
 #      name = "LKnee"
+#      id = 13
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 15
 #         name = "LAnkle"
+#         id = 15
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 19
 #            name = "LBigToe"
+#            id = 19
 #               [[pose.CUSTOM.children.children.children.children.children]]
-#               id = 20
 #               name = "LSmallToe"
+#               id = 20
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 21
 #            name = "LHeel"
+#            id = 21
 #   [[pose.CUSTOM.children]]
-#   id = 17
 #   name = "Neck"
+#   id = 17
 #      [[pose.CUSTOM.children.children]]
-#      id = 18
 #      name = "Head"
+#      id = 18
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 0
 #         name = "Nose"
+#         id = 0
 #      [[pose.CUSTOM.children.children]]
-#      id = 6
 #      name = "RShoulder"
+#      id = 6
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 8
 #         name = "RElbow"
+#         id = 8
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 10
 #            name = "RWrist"
+#            id = 10
 #      [[pose.CUSTOM.children.children]]
-#      id = 5
 #      name = "LShoulder"
+#      id = 5
 #         [[pose.CUSTOM.children.children.children]]
-#         id = 7
 #         name = "LElbow"
+#         id = 7
 #            [[pose.CUSTOM.children.children.children.children]]
-#            id = 9
 #            name = "LWrist"
+#            id = 9

Pose2Sim/Utilities/json_display_without_img.py

@@ -15,7 +15,7 @@
     
     Usage: 
     python -m json_display_without_img -j json_folder -W 1920 -H 1080
-    python -m json_display_without_img -j json_folder -o output_img_folder -d True -s True -W 1920 -H 1080 - 30
+    python -m json_display_without_img -j json_folder -o output_img_folder -d True -s True -W 1920 -H 1080 -f 30
     import json_display_without_img; json_display_without_img.json_display_without_img_func(json_folder=r'<json_folder>', image_width=1920, image_height = 1080)
 '''
 
@@ -60,7 +60,7 @@ def json_display_without_img_func(**args):
     
     Usage: 
     json_display_without_img -j json_folder -W 1920 -H 1080
-    json_display_without_img -j json_folder -o output_img_folder -d True -s True -W 1920 -H 1080
+    json_display_without_img -j json_folder -o output_img_folder -d True -s True -W 1920 -H 1080 -f 30
     import json_display_without_img; json_display_without_img.json_display_without_img_func(json_folder=r'<json_folder>', image_width=1920, image_height = 1080)
     '''
 

Pose2Sim/Utilities/synchronize_cams_draft.py

@@ -114,10 +114,10 @@ def interpolate_nans(col, kind):
 
 def plot_time_lagged_cross_corr(camx, camy, ax):
     pearson_r = [camx.corr(camy.shift(lag)) for lag in range(-2*fps, 2*fps)] # lag -2 sec à +2 sec
-    offset = int(np.floor(len(pearson_r)*2)-np.argmax(pearson_r))
+    offset = int(np.floor(len(pearson_r)/2)-np.argmax(pearson_r))
     max_corr = np.max(pearson_r)
     ax.plot(list(range(-2*fps, 2*fps)), pearson_r)
-    ax.axvline(np.ceil(len(pearson_r)*2)-2*fps,color='k',linestyle='--')
+    ax.axvline(np.ceil(len(pearson_r)/2)-2*fps,color='k',linestyle='--')
     ax.axvline(np.argmax(pearson_r)-2*fps,color='r',linestyle='--',label='Peak synchrony')
     plt.annotate(f'Max correlation={np.round(max_corr,2)}', xy=(0.05, 0.9), xycoords='axes fraction')
     ax.set(title=f'Offset = {offset} frames', xlabel='Offset (frames)',ylabel='Pearson r')
@@ -156,7 +156,6 @@ with open(os.path.join(pose_dir, 'coords'), 'wb') as fp:
 #############################
 
 
-
 # Vitesse verticale    
 df_speed = []
 for i in range(len(json_dirs)):
@@ -199,6 +198,41 @@ else:
     raise ValueError('wrong values for id_kpt or weights_kpt')
 
 
+
+# camx = df_speed[1][16]
+# camy = df_speed[2][16]
+# camx = df_speed[1][10]
+# camy = df_speed[2][10]
+# camx = df_speed[1].sum(axis=1)
+# camy = df_speed[2].sum(axis=1)
+# camx.plot()
+# camy.plot()
+# plt.show()
+
+for i in range(25):
+    df_coords[1].iloc[:,i*2+1].plot(label='1')
+    df_coords[2].iloc[:,i*2+1].plot(label='2')
+    plt.title(i)
+    plt.legend()
+    plt.show()
+
+for i in range(25):
+    df_speed[1].iloc[:,i].plot(label='1')
+    df_speed[2].iloc[:,i].plot(label='2')
+    plt.title(i)
+    plt.legend()
+    plt.show()
+
+for i in range(4):
+    abs(df_speed[i]).sum(axis=1).plot(label=i)
+plt.legend()
+plt.show()
+
+df_speed[0].plot()  # --> remove janky points
+plt.show()
+
+
+
 f, ax = plt.subplots(2,1)
 # speed
 camx.plot(ax=ax[0], label = f'cam {cam1_nb}')

Pose2Sim/common.py

@@ -12,6 +12,7 @@ Functions shared between modules, and other utilities
 
 ## INIT
 import toml
+import json
 import numpy as np
 import re
 import cv2
@@ -37,6 +38,58 @@ __status__ = "Development"
 
 
 ## FUNCTIONS
+def common_items_in_list(list1, list2):
+    '''
+    Do two lists have any items in common at the same index?
+    Returns True or False
+    '''
+    
+    for i, j in enumerate(list1):
+        if j == list2[i]:
+            return True
+    return False
+
+
+def bounding_boxes(js_file, margin_percent=0.1, around='extremities'):
+    '''
+    Compute the bounding boxes of the people in the json file.
+    Either around the extremities (with a margin)
+    or around the center of the person (with a margin).
+
+    INPUTS:
+    - js_file: json file
+    - margin_percent: margin around the person
+    - around: 'extremities' or 'center'
+
+    OUTPUT:
+    - bounding_boxes: list of bounding boxes [x_min, y_min, x_max, y_max]
+    '''
+
+    bounding_boxes = []
+    with open(js_file, 'r') as json_f:
+        js = json.load(json_f)
+        for people in range(len(js['people'])):
+            if len(js['people'][people]['pose_keypoints_2d']) < 3: continue
+            else:
+                x = js['people'][people]['pose_keypoints_2d'][0::3]
+                y = js['people'][people]['pose_keypoints_2d'][1::3]
+                x_min, x_max = min(x), max(x)
+                y_min, y_max = min(y), max(y)
+
+                if around == 'extremities':
+                    dx = (x_max - x_min) * margin_percent
+                    dy = (y_max - y_min) * margin_percent
+                    bounding_boxes.append([x_min-dx, y_min-dy, x_max+dx, y_max+dy])
+                
+                elif around == 'center':
+                    x_mean, y_mean = np.mean(x), np.mean(y)
+                    x_size = (x_max - x_min) * (1 + margin_percent)
+                    y_size = (y_max - y_min) * (1 + margin_percent)
+                    bounding_boxes.append([x_mean - x_size/2, y_mean - y_size/2, x_mean + x_size/2, y_mean + y_size/2])
+
+    return bounding_boxes   
+
+
 def retrieve_calib_params(calib_file):
     '''
     Compute projection matrices from toml calibration file.
@@ -48,6 +101,7 @@ def retrieve_calib_params(calib_file):
     - S: (h,w) vectors as list of 2x1 arrays
     - K: intrinsic matrices as list of 3x3 arrays
     - dist: distortion vectors as list of 4x1 arrays
+    - inv_K: inverse intrinsic matrices as list of 3x3 arrays
     - optim_K: intrinsic matrices for undistorting points as list of 3x3 arrays
     - R: rotation rodrigue vectors as list of 3x1 arrays
     - T: translation vectors as list of 3x1 arrays
@@ -55,16 +109,18 @@ def retrieve_calib_params(calib_file):
     
     calib = toml.load(calib_file)
 
-    S, K, dist, optim_K, R, T = [], [], [], [], [], []
+    S, K, dist, optim_K, inv_K, R, R_mat, T = [], [], [], [], [], [], [], []
     for c, cam in enumerate(calib.keys()):
         if cam != 'metadata':
             S.append(np.array(calib[cam]['size']))
             K.append(np.array(calib[cam]['matrix']))
             dist.append(np.array(calib[cam]['distortions']))
             optim_K.append(cv2.getOptimalNewCameraMatrix(K[c], dist[c], [int(s) for s in S[c]], 1, [int(s) for s in S[c]])[0])
+            inv_K.append(np.linalg.inv(K[c]))
             R.append(np.array(calib[cam]['rotation']))
+            R_mat.append(cv2.Rodrigues(R[c])[0])
             T.append(np.array(calib[cam]['translation']))
-    calib_params = {'S': S, 'K': K, 'dist': dist, 'optim_K': optim_K, 'R': R, 'T': T}
+    calib_params = {'S': S, 'K': K, 'dist': dist, 'inv_K': inv_K, 'optim_K': optim_K, 'R': R, 'R_mat': R_mat, 'T': T}
             
     return calib_params
 

Pose2Sim/personAssociation.py

@@ -8,13 +8,17 @@
 ###########################################################################
 
 Openpose detects all people in the field of view. 
-Which is the one of interest?
+- multi_person = false: Which is the one of interest?
+- multi_person = true: How to triangulate the same persons across views?
+                       How to associate them across time frames? Done in the 
+                       triangulation stage.
 
-This module tries all possible triangulations of a chosen anatomical 
+If multi_person = false, this module tries all possible triangulations of a chosen
-point. If "multi_person" mode is not used, it chooses the person for
+anatomical point, and chooses the person for whom the reprojection error is smallest. 
-whom the reprojection error is smallest. Otherwise, it selects all 
+
-persons with a reprojection error smaller than a threshold, and then 
+If multi_person = true, it computes the distance between epipolar lines (camera to 
-associates them across time frames by minimizing the displacement speed.
+keypoint lines) for all persons detected in all views, and selects the best correspondences. 
+The computation of the affinity matrix from the distance is inspired from the EasyMocap approach.
 
 INPUTS: 
 - a calibration file (.toml extension)
@@ -58,97 +62,6 @@ __status__ = "Development"
 
 
 ## FUNCTIONS
-def common_items_in_list(list1, list2):
-    '''
-    Do two lists have any items in common at the same index?
-    Returns True or False
-    '''
-    
-    for i, j in enumerate(list1):
-        if j == list2[i]:
-            return True
-    return False
-    
-
-def min_with_single_indices(L, T):
-    '''
-    Let L be a list (size s) with T associated tuple indices (size s).
-    Select the smallest values of L, considering that 
-    the next smallest value cannot have the same numbers 
-    in the associated tuple as any of the previous ones.
-
-    Example:
-    L = [  20,   27,  51,    33,   43,   23,   37,   24,   4,   68,   84,    3  ]
-    T = list(it.product(range(2),range(3)))
-      = [(0,0),(0,1),(0,2),(0,3),(1,0),(1,1),(1,2),(1,3),(2,0),(2,1),(2,2),(2,3)]
-
-    - 1st smallest value: 3 with tuple (2,3), index 11
-    - 2nd smallest value when excluding indices (2,.) and (.,3), i.e. [(0,0),(0,1),(0,2),X,(1,0),(1,1),(1,2),X,X,X,X,X]:
-    20 with tuple (0,0), index 0
-    - 3rd smallest value when excluding [X,X,X,X,X,(1,1),(1,2),X,X,X,X,X]:
-    23 with tuple (1,1), index 5
-    
-    INPUTS:
-    - L: list (size s)
-    - T: T associated tuple indices (size s)
-
-    OUTPUTS: 
-    - minL: list of smallest values of L, considering constraints on tuple indices
-    - argminL: list of indices of smallest values of L
-    - T_minL: list of tuples associated with smallest values of L
-    '''
-
-    minL = [np.min(L)]
-    argminL = [np.argmin(L)]
-    T_minL = [T[argminL[0]]]
-    
-    mask_tokeep = np.array([True for t in T])
-    i=0
-    while mask_tokeep.any()==True:
-        mask_tokeep = mask_tokeep & np.array([t[0]!=T_minL[i][0] and t[1]!=T_minL[i][1] for t in T])
-        if mask_tokeep.any()==True:
-            indicesL_tokeep = np.where(mask_tokeep)[0]
-            minL += [np.min(np.array(L)[indicesL_tokeep])]
-            argminL += [indicesL_tokeep[np.argmin(np.array(L)[indicesL_tokeep])]]
-            T_minL += (T[argminL[i+1]],)
-            i+=1
-    
-    return minL, argminL, T_minL
-    
-    
-def sort_people(Q_kpt_old, Q_kpt):
-    '''
-    Associate persons across frames
-    Persons' indices are sometimes swapped when changing frame
-    A person is associated to another in the next frame when they are at a small distance
-    
-    INPUTS:
-    - Q_kpt_old: list of arrays of 3D coordinates [X, Y, Z, 1.] for the previous frame
-    - Q_kpt: idem Q_kpt_old, for current frame
-    
-    OUTPUT:
-    - Q_kpt: array with reordered persons
-    - personsIDs_sorted: index of reordered persons
-    '''
-    
-    # Generate possible person correspondences across frames
-    if len(Q_kpt_old) < len(Q_kpt):
-        Q_kpt_old = np.concatenate((Q_kpt_old, [[0., 0., 0., 1.]]*(len(Q_kpt)-len(Q_kpt_old))))
-    personsIDs_comb = sorted(list(it.product(range(len(Q_kpt_old)),range(len(Q_kpt)))))
-    # Compute distance between persons from one frame to another
-    frame_by_frame_dist = []
-    for comb in personsIDs_comb:
-        frame_by_frame_dist += [euclidean_distance(Q_kpt_old[comb[0]][:3],Q_kpt[comb[1]][:3])]
-    # sort correspondences by distance
-    _, index_best_comb, _ = min_with_single_indices(frame_by_frame_dist, personsIDs_comb)
-    index_best_comb.sort()
-    personsIDs_sorted = np.array(personsIDs_comb)[index_best_comb][:,1]
-    # rearrange persons
-    Q_kpt = np.array(Q_kpt)[personsIDs_sorted]
-    
-    return Q_kpt, personsIDs_sorted
-
-
 def persons_combinations(json_files_framef):
     '''
     Find all possible combinations of detected persons' ids. 
@@ -179,10 +92,63 @@ def persons_combinations(json_files_framef):
     return personsIDs_comb
 
 
+def triangulate_comb(comb, coords, P_all, calib_params, config):
+    '''
+    Triangulate 2D points and compute reprojection error for a combination of cameras.
+    INPUTS:
+    - comb: list of ints: combination of persons' ids for each camera
+    - coords: array: x, y, likelihood for each camera
+    - P_all: list of arrays: projection matrices for each camera
+    - calib_params: dict: calibration parameters
+    - config: dictionary from Config.toml file
+    OUTPUTS:
+    - error_comb: float: reprojection error
+    - comb: list of ints: combination of persons' ids for each camera
+    - Q_comb: array: 3D coordinates of the triangulated point
+    ''' 
+
+    undistort_points = config.get('triangulation').get('undistort_points')
+    likelihood_threshold = config.get('personAssociation').get('likelihood_threshold_association')
+
+    # Replace likelihood by 0. if under likelihood_threshold
+    coords[:,2][coords[:,2] < likelihood_threshold] = 0.
+    comb[coords[:,2] == 0.] = np.nan
+
+    # Filter coords and projection_matrices containing nans
+    coords_filt = [coords[i] for i in range(len(comb)) if not np.isnan(comb[i])]
+    projection_matrices_filt = [P_all[i] for i in range(len(comb)) if not np.isnan(comb[i])]
+    if undistort_points:
+        calib_params_R_filt = [calib_params['R'][i] for i in range(len(comb)) if not np.isnan(comb[i])]
+        calib_params_T_filt = [calib_params['T'][i] for i in range(len(comb)) if not np.isnan(comb[i])]
+        calib_params_K_filt = [calib_params['K'][i] for i in range(len(comb)) if not np.isnan(comb[i])]
+        calib_params_dist_filt = [calib_params['dist'][i] for i in range(len(comb)) if not np.isnan(comb[i])]
+
+    # Triangulate 2D points
+    x_files_filt, y_files_filt, likelihood_files_filt = np.array(coords_filt).T
+    Q_comb = weighted_triangulation(projection_matrices_filt, x_files_filt, y_files_filt, likelihood_files_filt)
+
+    # Reprojection
+    if undistort_points:
+        coords_2D_kpt_calc_filt = [cv2.projectPoints(np.array(Q_comb[:-1]), calib_params_R_filt[i], calib_params_T_filt[i], calib_params_K_filt[i], calib_params_dist_filt[i])[0] for i in range(len(Q_comb))]
+        x_calc = [coords_2D_kpt_calc_filt[i][0,0,0] for i in range(len(Q_comb))]
+        y_calc = [coords_2D_kpt_calc_filt[i][0,0,1] for i in range(len(Q_comb))]
+    else:
+        x_calc, y_calc = reprojection(projection_matrices_filt, Q_comb)
+
+    # Reprojection error
+    error_comb_per_cam = []
+    for cam in range(len(x_calc)):
+        q_file = (x_files_filt[cam], y_files_filt[cam])
+        q_calc = (x_calc[cam], y_calc[cam])
+        error_comb_per_cam.append( euclidean_distance(q_file, q_calc) )
+    error_comb = np.mean(error_comb_per_cam)
+
+    return error_comb, comb, Q_comb
+
+
 def best_persons_and_cameras_combination(config, json_files_framef, personsIDs_combinations, projection_matrices, tracked_keypoint_id, calib_params):
     '''
-    - if multi_person: Choose all the combination of cameras that give a reprojection error below a threshold
+    Chooses the right person among the multiple ones found by
-    - else: Chooses the right person among the multiple ones found by
     OpenPose & excludes cameras with wrong 2d-pose estimation.
     
     1. triangulate the tracked keypoint for all possible combinations of people,
@@ -203,9 +169,7 @@ def best_persons_and_cameras_combination(config, json_files_framef, personsIDs_c
     - comb_errors_below_thresh: list of arrays of ints
     '''
     
-    multi_person = config.get('project').get('multi_person')
+    error_threshold_tracking = config.get('personAssociation').get('single_person').get('reproj_error_threshold_association')
-    nb_persons_to_detect = config.get('project').get('nb_persons_to_detect')
-    error_threshold_tracking = config.get('personAssociation').get('reproj_error_threshold_association')
     likelihood_threshold = config.get('personAssociation').get('likelihood_threshold_association')
     min_cameras_for_triangulation = config.get('triangulation').get('min_cameras_for_triangulation')
     undistort_points = config.get('triangulation').get('undistort_points')
@@ -219,29 +183,22 @@ def best_persons_and_cameras_combination(config, json_files_framef, personsIDs_c
     while error_min > error_threshold_tracking and n_cams - nb_cams_off >= min_cameras_for_triangulation:
         # Try all persons combinations
         for combination in personsIDs_combinations:
-            # Get x,y,likelihood values from files
+            #  Get coords from files
-            x_files, y_files,likelihood_files = [], [], []
+            coords = []
             for index_cam, person_nb in enumerate(combination):
-                with open(json_files_framef[index_cam], 'r') as json_f:
-                    js = json.load(json_f)
                 try:
-                        x_files.append( js['people'][int(person_nb)]['pose_keypoints_2d'][tracked_keypoint_id*3] )
+                    js = read_json(json_files_framef[index_cam])
-                        y_files.append( js['people'][int(person_nb)]['pose_keypoints_2d'][tracked_keypoint_id*3+1] )
+                    coords.append(js[int(person_nb)][tracked_keypoint_id*3:tracked_keypoint_id*3+3])
-                        likelihood_files.append( js['people'][int(person_nb)]['pose_keypoints_2d'][tracked_keypoint_id*3+2] )
                 except:
-                        x_files.append(np.nan)
+                    coords.append([np.nan, np.nan, np.nan])
-                        y_files.append(np.nan)
+            coords = np.array(coords)
-                        likelihood_files.append(np.nan)
             
             # undistort points
             if undistort_points:
-                points = np.array(tuple(zip(x_files,y_files))).reshape(-1, 1, 2).astype('float32')
+                points = np.array(coords)[:,None,:2]
                 undistorted_points = [cv2.undistortPoints(points[i], calib_params['K'][i], calib_params['dist'][i], None, calib_params['optim_K'][i]) for i in range(n_cams)]
-                x_files = np.array([[u[i][0][0] for i in range(len(u))] for u in undistorted_points]).squeeze()
+                coords[:,0] = np.array([[u[i][0][0] for i in range(len(u))] for u in undistorted_points]).squeeze()
-                y_files = np.array([[u[i][0][1] for i in range(len(u))] for u in undistorted_points]).squeeze()
+                coords[:,1] = np.array([[u[i][0][1] for i in range(len(u))] for u in undistorted_points]).squeeze()
-            
-            # Replace likelihood by 0. if under likelihood_threshold
-            likelihood_files = [0. if lik < likelihood_threshold else lik for lik in likelihood_files]
 
             # For each persons combination, create subsets with "nb_cams_off" cameras excluded
             id_cams_off = list(it.combinations(range(len(combination)), nb_cams_off))
@@ -250,91 +207,338 @@ def best_persons_and_cameras_combination(config, json_files_framef, personsIDs_c
                 combinations_with_cams_off[i,id] = np.nan
 
             # Try all subsets
-            error_comb = []
+            error_comb_all, comb_all, Q_comb_all = [], [], []
-            Q_comb = []
             for comb in combinations_with_cams_off:
-                # Filter x, y, likelihood, projection_matrices, with subset
+                error_comb, comb, Q_comb = triangulate_comb(comb, coords, projection_matrices, calib_params, config)
-                x_files_filt = [x_files[i] for i in range(len(comb)) if not np.isnan(comb[i])]
+                error_comb_all.append(error_comb)
-                y_files_filt = [y_files[i] for i in range(len(comb)) if not np.isnan(comb[i])]
+                comb_all.append(comb)
-                likelihood_files_filt = [likelihood_files[i] for i in range(len(comb)) if not np.isnan(comb[i])]
+                Q_comb_all.append(Q_comb)
-                projection_matrices_filt = [projection_matrices[i] for i in range(len(comb)) if not np.isnan(comb[i])]
-                if undistort_points:
-                    calib_params_R_filt = [calib_params['R'][i] for i in range(len(comb)) if not np.isnan(comb[i])]
-                    calib_params_T_filt = [calib_params['T'][i] for i in range(len(comb)) if not np.isnan(comb[i])]
-                    calib_params_K_filt = [calib_params['K'][i] for i in range(len(comb)) if not np.isnan(comb[i])]
-                    calib_params_dist_filt = [calib_params['dist'][i] for i in range(len(comb)) if not np.isnan(comb[i])]
 
-                # Triangulate 2D points
+            error_min = np.nanmin(error_comb_all)
-                Q_comb.append(weighted_triangulation(projection_matrices_filt, x_files_filt, y_files_filt, likelihood_files_filt))
+            comb_error_min = [comb_all[np.argmin(error_comb_all)]]
-                
+            Q_kpt = [Q_comb_all[np.argmin(error_comb_all)]]
-                # Reprojection
+            if error_min < error_threshold_tracking:
-                if undistort_points:
-                    coords_2D_kpt_calc_filt = [cv2.projectPoints(np.array(Q_comb[-1][:-1]), calib_params_R_filt[i], calib_params_T_filt[i], calib_params_K_filt[i], calib_params_dist_filt[i])[0] for i in range(n_cams-nb_cams_off)]
-                    x_calc = [coords_2D_kpt_calc_filt[i][0,0,0] for i in range(n_cams-nb_cams_off)]
-                    y_calc = [coords_2D_kpt_calc_filt[i][0,0,1] for i in range(n_cams-nb_cams_off)]
-                else:
-                    x_calc, y_calc = reprojection(projection_matrices_filt, Q_comb[-1])
-                                                
-                # Reprojection error
-                error_comb_per_cam = []
-                for cam in range(len(x_calc)):
-                    q_file = (x_files_filt[cam], y_files_filt[cam])
-                    q_calc = (x_calc[cam], y_calc[cam])
-                    error_comb_per_cam.append( euclidean_distance(q_file, q_calc) )
-                error_comb.append( np.mean(error_comb_per_cam) )
-            
-            if multi_person:
-                errors_below_thresh += [e for e in error_comb if e<error_threshold_tracking]
-                comb_errors_below_thresh += [combinations_with_cams_off[error_comb.index(e)] for e in error_comb if e<error_threshold_tracking]
-                Q_kpt += [Q_comb[error_comb.index(e)] for e in error_comb if e<error_threshold_tracking]
-            else:
-                error_min = np.nanmin(error_comb)
-                errors_below_thresh = [error_min]
-                comb_errors_below_thresh = [combinations_with_cams_off[np.argmin(error_comb)]]
-                Q_kpt = [Q_comb[np.argmin(error_comb)]]
-                if errors_below_thresh[0] < error_threshold_tracking:
-                    break 
-                
-        if multi_person:
-            if len(errors_below_thresh)>0:
-                # sort combinations by error magnitude
-                errors_below_thresh_sorted = sorted(errors_below_thresh)
-                sorted_idx = np.array([errors_below_thresh.index(e) for e in errors_below_thresh_sorted])
-                comb_errors_below_thresh = np.array(comb_errors_below_thresh)[sorted_idx]
-                Q_kpt = np.array(Q_kpt)[sorted_idx]
-                # remove combinations with indices used several times for the same person 
-                comb_errors_below_thresh = [c.tolist() for c in comb_errors_below_thresh]
-                comb = comb_errors_below_thresh.copy()
-                comb_ok = np.array([comb[0]])
-                for i, c1 in enumerate(comb):
-                    idx_ok = np.array([not(common_items_in_list(c1, c2)) for c2 in comb[1:]])
-                    try:
-                        comb = np.array(comb[1:])[idx_ok]
-                        comb_ok = np.concatenate((comb_ok, [comb[0]]))
-                    except:
-                        break
-                sorted_pruned_idx = [i for i, x in enumerate(comb_errors_below_thresh) for c in comb_ok if np.array_equal(x,c,equal_nan=True)]
-                errors_below_thresh = np.array(errors_below_thresh_sorted)[sorted_pruned_idx].tolist()
-                comb_errors_below_thresh = np.array(comb_errors_below_thresh)[sorted_pruned_idx].tolist()
-                Q_kpt = Q_kpt[sorted_pruned_idx].tolist()
-
-            # Remove indices already used for a person
-            personsIDs_combinations = np.array([personsIDs_combinations[i] for i in range(len(personsIDs_combinations))
-                                           if not np.array( 
-                                                [personsIDs_combinations[i,j]==comb[j] for comb in comb_errors_below_thresh for j in range(len(comb))]
-                                                ).any()])
-            if len(errors_below_thresh) >= len(personsIDs_combinations) or len(errors_below_thresh) >= nb_persons_to_detect: 
-                errors_below_thresh = errors_below_thresh[:nb_persons_to_detect]
-                comb_errors_below_thresh = comb_errors_below_thresh[:nb_persons_to_detect]
-                Q_kpt = Q_kpt[:nb_persons_to_detect]
                 break 
 
         nb_cams_off += 1
     
-    return errors_below_thresh, comb_errors_below_thresh, Q_kpt
+    return error_min, comb_error_min, Q_kpt
 
 
-def recap_tracking(config, error, nb_cams_excluded):
+def read_json(js_file):
+    '''
+    Read OpenPose json file
+    '''
+    with open(js_file, 'r') as json_f:
+        js = json.load(json_f)
+        json_data = []
+        for people in range(len(js['people'])):
+            if len(js['people'][people]['pose_keypoints_2d']) < 3: continue
+            else:
+                json_data.append(js['people'][people]['pose_keypoints_2d'])
+    return json_data
+
+
+def compute_rays(json_coord, calib_params, cam_id):
+    '''
+    Plucker coordinates of rays from camera to each joint of a person
+    Plucker coordinates: camera to keypoint line direction (size 3) 
+                         moment: origin ^ line (size 3)
+                         additionally, confidence
+
+    INPUTS:
+    - json_coord: x, y, likelihood for a person seen from a camera (list of 3*joint_nb)
+    - calib_params: calibration parameters from retrieve_calib_params('calib.toml')
+    - cam_id: camera id (int)
+
+    OUTPUT:
+    - plucker: array. nb joints * (6 plucker coordinates + 1 likelihood)
+    '''
+
+    x = json_coord[0::3]
+    y = json_coord[1::3]
+    likelihood = json_coord[2::3]
+    
+    inv_K = calib_params['inv_K'][cam_id]
+    R_mat = calib_params['R_mat'][cam_id]
+    T = calib_params['T'][cam_id]
+
+    cam_center = -R_mat.T @ T
+    plucker = []
+    for i in range(len(x)):
+        q = np.array([x[i], y[i], 1])
+        norm_Q = R_mat.T @ (inv_K @ q -T)
+        
+        line = norm_Q - cam_center
+        norm_line = line/np.linalg.norm(line)
+        moment = np.cross(cam_center, norm_line)
+        plucker.append(np.concatenate([norm_line, moment, [likelihood[i]]]))
+
+    return np.array(plucker)
+
+
+def broadcast_line_to_line_distance(p0, p1):
+    '''
+    Compute the distance between two lines in 3D space.
+
+    see: https://faculty.sites.iastate.edu/jia/files/inline-files/plucker-coordinates.pdf
+    p0 = (l0,m0), p1 = (l1,m1)
+    dist = | (l0,m0) * (l1,m1) | / || l0 x l1 ||
+    (l0,m0) * (l1,m1) = l0 @ m1 + m0 @ l1 (reciprocal product)
+    
+    No need to divide by the norm of the cross product of the directions, since we
+    don't need the actual distance but whether the lines are close to intersecting or not
+    => dist = | (l0,m0) * (l1,m1) |
+
+    INPUTS:
+    - p0: array(nb_persons_detected * 1 * nb_joints * 7 coordinates)
+    - p1: array(1 * nb_persons_detected * nb_joints * 7 coordinates)
+
+    OUTPUT:
+    - dist: distances between the two lines (not normalized). 
+            array(nb_persons_0 * nb_persons_1 * nb_joints)
+    '''
+
+    product = np.sum(p0[..., :3] * p1[..., 3:6], axis=-1) + np.sum(p1[..., :3] * p0[..., 3:6], axis=-1)
+    dist = np.abs(product)
+
+    return dist
+
+
+def compute_affinity(all_json_data_f, calib_params, cum_persons_per_view, reconstruction_error_threshold=0.1):
+    '''
+    Compute the affinity between all the people in the different views.
+
+    The affinity is defined as 1 - distance/max_distance, with distance the
+    distance between epipolar lines in each view (reciprocal product of Plucker 
+    coordinates).
+
+    Another approach would be to project one epipolar line onto the other camera
+    plane and compute the line to point distance, but it is more computationally 
+    intensive (simple dot product vs. projection and distance calculation). 
+    
+    INPUTS:
+    - all_json_data_f: list of json data. For frame f, nb_views*nb_persons*(x,y,likelihood)*nb_joints
+    - calib_params: calibration parameters from retrieve_calib_params('calib.toml')
+    - cum_persons_per_view: cumulative number of persons per view
+    - reconstruction_error_threshold: maximum distance between epipolar lines to consider a match
+
+    OUTPUT:
+    - affinity: affinity matrix between all the people in the different views. 
+                (nb_views*nb_persons_per_view * nb_views*nb_persons_per_view)
+    '''
+
+    # Compute plucker coordinates for all keypoints for each person in each view
+    # pluckers_f: dims=(camera, person, joint, 7 coordinates)
+    pluckers_f = []
+    for cam_id, json_cam  in enumerate(all_json_data_f):
+        pluckers = []
+        for json_coord in json_cam:
+            plucker = compute_rays(json_coord, calib_params, cam_id) # LIMIT TO 15 JOINTS? json_coord[:15*3]
+            pluckers.append(plucker)
+        pluckers = np.array(pluckers)
+        pluckers_f.append(pluckers)
+
+    # Compute affinity matrix
+    distance = np.zeros((cum_persons_per_view[-1], cum_persons_per_view[-1])) + 2*reconstruction_error_threshold
+    for compared_cam0, compared_cam1 in it.combinations(range(len(all_json_data_f)), 2):
+        # skip when no detection for a camera
+        if cum_persons_per_view[compared_cam0] == cum_persons_per_view[compared_cam0+1] \
+            or cum_persons_per_view[compared_cam1] == cum_persons_per_view[compared_cam1 +1]:
+            continue
+
+        # compute distance
+        p0 = pluckers_f[compared_cam0][:,None] # add coordinate on second dimension
+        p1 = pluckers_f[compared_cam1][None,:] # add coordinate on first dimension
+        dist = broadcast_line_to_line_distance(p0, p1)
+        likelihood = np.sqrt(p0[..., -1] * p1[..., -1])
+        mean_weighted_dist = np.sum(dist*likelihood, axis=-1)/(1e-5 + likelihood.sum(axis=-1)) # array(nb_persons_0 * nb_persons_1)
+        
+        # populate distance matrix
+        distance[cum_persons_per_view[compared_cam0]:cum_persons_per_view[compared_cam0+1], \
+                 cum_persons_per_view[compared_cam1]:cum_persons_per_view[compared_cam1+1]] \
+                 = mean_weighted_dist
+        distance[cum_persons_per_view[compared_cam1]:cum_persons_per_view[compared_cam1+1], \
+                 cum_persons_per_view[compared_cam0]:cum_persons_per_view[compared_cam0+1]] \
+                 = mean_weighted_dist.T
+
+    # compute affinity matrix and clamp it to zero when distance > reconstruction_error_threshold
+    distance[distance > reconstruction_error_threshold] = reconstruction_error_threshold
+    affinity = 1 - distance / reconstruction_error_threshold
+
+    return affinity
+
+
+def circular_constraint(cum_persons_per_view):
+    '''
+    A person can be matched only with themselves in the same view, and with any 
+    person from other views
+
+    INPUT:
+    - cum_persons_per_view: cumulative number of persons per view
+
+    OUTPUT:
+    - circ_constraint: circular constraint matrix
+    '''
+
+    circ_constraint = np.identity(cum_persons_per_view[-1])
+    for i in range(len(cum_persons_per_view)-1):
+        circ_constraint[cum_persons_per_view[i]:cum_persons_per_view[i+1], cum_persons_per_view[i+1]:cum_persons_per_view[-1]] = 1
+        circ_constraint[cum_persons_per_view[i+1]:cum_persons_per_view[-1], cum_persons_per_view[i]:cum_persons_per_view[i+1]] = 1
+    
+    return circ_constraint
+
+
+def SVT(matrix, threshold):
+    '''
+    Find a low-rank approximation of the matrix using Singular Value Thresholding.
+
+    INPUTS:
+    - matrix: matrix to decompose
+    - threshold: threshold for singular values
+
+    OUTPUT:
+    - matrix_thresh: low-rank approximation of the matrix
+    '''
+    
+    U, s, Vt = np.linalg.svd(matrix) # decompose matrix
+    s_thresh = np.maximum(s - threshold, 0) # set smallest singular values to zero
+    matrix_thresh = U @ np.diag(s_thresh) @ Vt # recompose matrix
+
+    return matrix_thresh
+
+
+def matchSVT(affinity, cum_persons_per_view, circ_constraint, max_iter = 20, w_rank = 50, tol = 1e-4, w_sparse=0.1):
+    '''
+    Find low-rank approximation of 'affinity' while satisfying the circular constraint.
+
+    INPUTS:
+    - affinity: affinity matrix between all the people in the different views
+    - cum_persons_per_view: cumulative number of persons per view
+    - circ_constraint: circular constraint matrix
+    - max_iter: maximum number of iterations
+    - w_rank: threshold for singular values
+    - tol: tolerance for convergence
+    - w_sparse: regularization parameter
+
+    OUTPUT:
+    - new_aff: low-rank approximation of the affinity matrix
+    '''
+
+    new_aff = affinity.copy()
+    N = new_aff.shape[0]
+    index_diag = np.arange(N)
+    new_aff[index_diag, index_diag] = 0.
+    # new_aff = (new_aff + new_aff.T)/2 # symmetric by construction
+
+    Y = np.zeros_like(new_aff) # Initial deviation matrix / residual ()
+    W = w_sparse - new_aff # Initial sparse matrix / regularization (prevent overfitting)
+    mu = 64 # initial step size
+
+    for iter in range(max_iter):
+        new_aff0 = new_aff.copy()
+        
+        Q = new_aff + Y*1.0/mu
+        Q = SVT(Q,w_rank/mu)
+        new_aff = Q - (W + Y)/mu
+
+        # Project X onto dimGroups
+        for i in range(len(cum_persons_per_view) - 1):
+            ind1, ind2 = cum_persons_per_view[i], cum_persons_per_view[i + 1]
+            new_aff[ind1:ind2, ind1:ind2] = 0
+            
+        # Reset diagonal elements to one and ensure X is within valid range [0, 1]
+        new_aff[index_diag, index_diag] = 1.
+        new_aff[new_aff < 0] = 0
+        new_aff[new_aff > 1] = 1
+        
+        # Enforce circular constraint
+        new_aff = new_aff * circ_constraint
+        new_aff = (new_aff + new_aff.T) / 2 # kept just in case X loses its symmetry during optimization 
+        Y = Y + mu * (new_aff - Q)
+        
+        # Compute convergence criteria: break if new_aff is close enough to Q and no evolution anymore
+        pRes = np.linalg.norm(new_aff - Q) / N # primal residual (diff between new_aff and SVT result)
+        dRes = mu * np.linalg.norm(new_aff - new_aff0) / N # dual residual (diff between new_aff and previous new_aff)
+        if pRes < tol and dRes < tol:
+            break
+        if pRes > 10 * dRes: mu = 2 * mu
+        elif dRes > 10 * pRes: mu = mu / 2
+
+        iter +=1
+
+    return new_aff
+
+
+def person_index_per_cam(affinity, cum_persons_per_view, min_cameras_for_triangulation):
+    '''
+    For each detected person, gives their index for each camera
+
+    INPUTS:
+    - affinity: affinity matrix between all the people in the different views
+    - min_cameras_for_triangulation: exclude proposals if less than N cameras see them
+
+    OUTPUT:
+    - proposals: 2D array: n_persons * n_cams
+    '''
+
+    # index of the max affinity for each group (-1 if no detection)
+    proposals = []
+    for row in range(affinity.shape[0]):
+        proposal_row = []
+        for cam in range(len(cum_persons_per_view)-1):
+            id_persons_per_view = affinity[row, cum_persons_per_view[cam]:cum_persons_per_view[cam+1]]
+            proposal_row += [np.argmax(id_persons_per_view) if (len(id_persons_per_view)>0 and max(id_persons_per_view)>0) else -1]
+        proposals.append(proposal_row)
+    proposals = np.array(proposals, dtype=float)
+
+    # remove duplicates and order
+    proposals, nb_detections = np.unique(proposals, axis=0, return_counts=True)
+    proposals = proposals[np.argsort(nb_detections)[::-1]]
+
+    # remove row if any value is the same in previous rows at same index (nan!=nan so nan ignored)
+    proposals[proposals==-1] = np.nan
+    mask = np.ones(proposals.shape[0], dtype=bool)
+    for i in range(1, len(proposals)):
+        mask[i] = ~np.any(proposals[i] == proposals[:i], axis=0).any()
+    proposals = proposals[mask]
+
+    # remove identifications if less than N cameras see them
+    nb_cams_per_person = [np.count_nonzero(~np.isnan(p)) for p in proposals]
+    proposals = np.array([p for (n,p) in zip(nb_cams_per_person, proposals) if n >= min_cameras_for_triangulation])
+
+    return proposals
+
+
+def rewrite_json_files(json_tracked_files_f, json_files_f, proposals, n_cams):
+    '''
+    Write new json files with correct association of people across cameras.
+
+    INPUTS:
+    - json_tracked_files_f: list of strings: json files to write
+    - json_files_f: list of strings: json files to read
+    - proposals: 2D array: n_persons * n_cams
+    - n_cams: int: number of cameras
+
+    OUTPUT:
+    - json files with correct association of people across cameras
+    '''
+
+    for cam in range(n_cams):
+        with open(json_tracked_files_f[cam], 'w') as json_tracked_f:
+            with open(json_files_f[cam], 'r') as json_f:
+                js = json.load(json_f)
+                js_new = js.copy()
+                js_new['people'] = []
+                for new_comb in proposals:
+                    if not np.isnan(new_comb[cam]):
+                        js_new['people'] += [js['people'][int(new_comb[cam])]]
+                    else:
+                        js_new['people'] += [{}]
+            json_tracked_f.write(json.dumps(js_new))
+
+
+def recap_tracking(config, error=0, nb_cams_excluded=0):
     '''
     Print a message giving statistics on reprojection errors (in pixel and in m)
     as well as the number of cameras that had to be excluded to reach threshold
@@ -352,12 +556,18 @@ def recap_tracking(config, error, nb_cams_excluded):
     # Read config
     project_dir = config.get('project').get('project_dir')
     session_dir = os.path.realpath(os.path.join(project_dir, '..', '..'))
-    tracked_keypoint = config.get('personAssociation').get('tracked_keypoint')
+    multi_person = config.get('project').get('multi_person')
-    error_threshold_tracking = config.get('personAssociation').get('reproj_error_threshold_association')
+    likelihood_threshold_association = config.get('personAssociation').get('likelihood_threshold_association')
+    tracked_keypoint = config.get('personAssociation').get('single_person').get('tracked_keypoint')
+    error_threshold_tracking = config.get('personAssociation').get('single_person').get('reproj_error_threshold_association')
+    reconstruction_error_threshold = config.get('personAssociation').get('multi_person').get('reconstruction_error_threshold')
+    min_affinity = config.get('personAssociation').get('multi_person').get('min_affinity')
     poseTracked_dir = os.path.join(project_dir, 'pose-associated')
     calib_dir = [os.path.join(session_dir, c) for c in os.listdir(session_dir) if 'calib' in c.lower()][0]
     calib_file = glob.glob(os.path.join(calib_dir, '*.toml'))[0] # lastly created calibration file
     
+    if not multi_person:
+        logging.info('\nSingle-person analysis selected.')
         # Error
         mean_error_px = np.around(np.mean(error), decimals=1)
         
@@ -372,7 +582,13 @@ def recap_tracking(config, error, nb_cams_excluded):
 
         # Recap
         logging.info(f'\n--> Mean reprojection error for {tracked_keypoint} point on all frames is {mean_error_px} px, which roughly corresponds to {mean_error_mm} mm. ')
-    logging.info(f'--> In average, {mean_cam_off_count} cameras had to be excluded to reach the demanded {error_threshold_tracking} px error threshold.')
+        logging.info(f'--> In average, {mean_cam_off_count} cameras had to be excluded to reach the demanded {error_threshold_tracking} px error threshold after excluding points with likelihood below {likelihood_threshold_association}.')
+    
+    else:
+        logging.info('\nMulti-person analysis selected.')
+        logging.info(f'\n--> A person was reconstructed if the lines from cameras to their keypoints intersected within {reconstruction_error_threshold} m and if the calculated affinity stayed below {min_affinity} after excluding points with likelihood below {likelihood_threshold_association}.')
+        logging.info(f'--> Beware that people were sorted across cameras, but not across frames. This will be done in the triangulation stage.')
+
     logging.info(f'\nTracked json files are stored in {os.path.realpath(poseTracked_dir)}.')
     
 
@@ -401,7 +617,11 @@ def track_2d_all(config):
     session_dir = os.path.realpath(os.path.join(project_dir, '..', '..'))
     multi_person = config.get('project').get('multi_person')
     pose_model = config.get('pose').get('pose_model')
-    tracked_keypoint = config.get('personAssociation').get('tracked_keypoint')
+    tracked_keypoint = config.get('personAssociation').get('single_person').get('tracked_keypoint')
+    likelihood_threshold = config.get('personAssociation').get('likelihood_threshold_association')
+    min_cameras_for_triangulation = config.get('triangulation').get('min_cameras_for_triangulation')
+    reconstruction_error_threshold = config.get('personAssociation').get('multi_person').get('reconstruction_error_threshold')
+    min_affinity = config.get('personAssociation').get('multi_person').get('min_affinity')
     frame_range = config.get('project').get('frame_range')
     undistort_points = config.get('triangulation').get('undistort_points')
     
@@ -414,12 +634,12 @@ def track_2d_all(config):
     poseTracked_dir = os.path.join(project_dir, 'pose-associated')
 
     if multi_person:
-        logging.info('\nMulti-person analysis selected. Note that you can set this option to false for faster runtime if you only need the main person in the scene.')
+        logging.info('\nMulti-person analysis selected. Note that you can set this option to false if you only need the main person in the scene.')
     else:
         logging.info('\nSingle-person analysis selected.')
 
     # projection matrix from toml calibration file
-    P = computeP(calib_file, undistort=undistort_points)
+    P_all = computeP(calib_file, undistort=undistort_points)
     calib_params = retrieve_calib_params(calib_file)
         
     # selection of tracked keypoint id
@@ -448,15 +668,14 @@ def track_2d_all(config):
     except: pass
     json_tracked_files = [[os.path.join(poseTracked_dir, j_dir, j_file) for j_file in json_files_names[j]] for j, j_dir in enumerate(json_dirs_names)]
     
-    # person's tracking
     f_range = [[min([len(j) for j in json_files])] if frame_range==[] else frame_range][0]
     n_cams = len(json_dirs_names)
     error_min_tot, cameras_off_tot = [], []
     
     # Check that camera number is consistent between calibration file and pose folders
-    if n_cams != len(P):
+    if n_cams != len(P_all):
         raise Exception(f'Error: The number of cameras is not consistent:\
-                    Found {len(P)} cameras in the calibration file,\
+                    Found {len(P_all)} cameras in the calibration file,\
                     and {n_cams} cameras based on the number of pose folders.')
     
     Q_kpt = [np.array([0., 0., 0., 1.])]
@@ -464,35 +683,40 @@ def track_2d_all(config):
         # print(f'\nFrame {f}:')
         json_files_f = [json_files[c][f] for c in range(n_cams)]
         json_tracked_files_f = [json_tracked_files[c][f] for c in range(n_cams)]
+        Q_kpt_old = Q_kpt
 
+        if not multi_person:
             # all possible combinations of persons
             personsIDs_comb = persons_combinations(json_files_f) 
             
             # choose persons of interest and exclude cameras with bad pose estimation
-        Q_kpt_old = Q_kpt
+            error_proposals, proposals, Q_kpt = best_persons_and_cameras_combination(config, json_files_f, personsIDs_comb, P_all, tracked_keypoint_id, calib_params)
-        errors_below_thresh, comb_errors_below_thresh, Q_kpt = best_persons_and_cameras_combination(config, json_files_f, personsIDs_comb, P, tracked_keypoint_id, calib_params)
 
-        # reID persons across frames by checking the distance from one frame to another
+            error_min_tot.append(np.mean(error_proposals))
-        Q_kpt, personsIDs_sorted = sort_people(Q_kpt_old, Q_kpt)
+            cameras_off_count = np.count_nonzero([np.isnan(comb) for comb in proposals]) / len(proposals)
-        errors_below_thresh = np.array(errors_below_thresh)[personsIDs_sorted]
+            cameras_off_tot.append(cameras_off_count)            
-        comb_errors_below_thresh = np.array(comb_errors_below_thresh)[personsIDs_sorted]
+
+        else:
+            # read data
+            all_json_data_f = []
+            for js_file in json_files_f:
+                all_json_data_f.append(read_json(js_file))
+            #TODO: remove people with average likelihood < 0.3, no full torso, less than 12 joints... (cf filter2d in dataset/base.py L498)
+            
+            # obtain proposals after computing affinity between all the people in the different views
+            persons_per_view = [0] + [len(j) for j in all_json_data_f]
+            cum_persons_per_view = np.cumsum(persons_per_view)
+            affinity = compute_affinity(all_json_data_f, calib_params, cum_persons_per_view, reconstruction_error_threshold=reconstruction_error_threshold)
+            circ_constraint = circular_constraint(cum_persons_per_view)
+            affinity = affinity * circ_constraint
+            #TODO: affinity without hand, face, feet (cf ray.py L31)
+            affinity = matchSVT(affinity, cum_persons_per_view, circ_constraint, max_iter = 20, w_rank = 50, tol = 1e-4, w_sparse=0.1)
+            affinity[affinity<min_affinity] = 0
+            proposals = person_index_per_cam(affinity, cum_persons_per_view, min_cameras_for_triangulation)
         
         # rewrite json files with a single or multiple persons of interest
-        error_min_tot.append(np.mean(errors_below_thresh))
+        rewrite_json_files(json_tracked_files_f, json_files_f, proposals, n_cams)
-        cameras_off_count = np.count_nonzero([np.isnan(comb) for comb in comb_errors_below_thresh]) / len(comb_errors_below_thresh)
+
-        cameras_off_tot.append(cameras_off_count)
-        for cam in range(n_cams):
-            with open(json_tracked_files_f[cam], 'w') as json_tracked_f:
-                with open(json_files_f[cam], 'r') as json_f:
-                    js = json.load(json_f)
-                    js_new = js.copy()
-                    js_new['people'] = []
-                    for new_comb in comb_errors_below_thresh:
-                        if not np.isnan(new_comb[cam]):
-                            js_new['people'] += [js['people'][int(new_comb[cam])]]
-                        else:
-                            js_new['people'] += [{}]
-                json_tracked_f.write(json.dumps(js_new))
 
     # recap message
     recap_tracking(config, error_min_tot, cameras_off_tot)

Pose2Sim/synchronize_cams.py

@@ -1,18 +1,10 @@
-import numpy as np
+#!/usr/bin/env python
-import pandas as pd
+# -*- coding: utf-8 -*-
-import matplotlib.pyplot as plt
-from scipy import signal
-from scipy import interpolate
-import json
-import os
-import fnmatch
-import pickle as pk
-import re
 
 
 '''
     #########################################
-    ## Synchronize cameras                 ##
+    ## SYNCHRONIZE CAMERAS                 ##
     #########################################
 
     Steps undergone in this script
@@ -25,26 +17,49 @@ import re
 '''
 
 
-############
+## INIT
-# FUNCTIONS#
+import numpy as np
-############
+import pandas as pd
+import matplotlib.pyplot as plt
+from scipy import signal
+from scipy import interpolate
+import json
+import os
+import fnmatch
+import pickle as pk
+import re
 
-def convert_json2csv(json_dir):
+from Pose2Sim.filtering import loess_filter_1d
-    """
+
+
+## AUTHORSHIP INFORMATION
+__author__ = "HunMin Kim, David Pagnon"
+__copyright__ = "Copyright 2021, Pose2Sim"
+__credits__ = ["David Pagnon"]
+__license__ = "BSD 3-Clause License"
+__version__ = '0.7'
+__maintainer__ = "David Pagnon"
+__email__ = "contact@david-pagnon.com"
+__status__ = "Development"
+
+
+# FUNCTIONS
+def convert_json2pandas(json_dir):
+    '''
     Convert JSON files in a directory to a pandas DataFrame.
 
-    Args:
+    INPUTS:
-        json_dir (str): The directory path containing the JSON files.
+    - json_dir: str. The directory path containing the JSON files.
+
+    OUTPUT:
+    - df_json_coords: dataframe. Extracted coordinates in a pandas dataframe.
+    '''
 
-    Returns:
-        pandas.DataFrame: A DataFrame containing the coordinates extracted from the JSON files.
-    """
     json_files_names = fnmatch.filter(os.listdir(os.path.join(json_dir)), '*.json') # modified ( 'json' to '*.json' )
-    json_files_names.sort(key=lambda name: int(re.search(r'(\d+)_keypoints\.json', name).group(1)))
+    json_files_names.sort(key=lambda name: int(re.search(r'(\d+)\.json', name).group(1)))
     json_files_path = [os.path.join(json_dir, j_f) for j_f in json_files_names]
     json_coords = []
     for i, j_p in enumerate(json_files_path):
-        # if i in range(frames)
         with open(j_p) as j_f:
             try:
                 json_data = json.load(j_f)['people'][0]['pose_keypoints_2d']
@@ -55,34 +70,37 @@ def convert_json2csv(json_dir):
     df_json_coords = pd.DataFrame(json_coords)
     return df_json_coords
 
+
 def drop_col(df, col_nb):
-    """
+    '''
     Drops every nth column from a DataFrame.
 
-    Parameters:
+    INPUTS:
-    df (pandas.DataFrame): The DataFrame from which columns will be dropped.
+    - df: dataframe. The DataFrame from which columns will be dropped.
-    col_nb (int): The column number to drop.
+    - col_nb: int. The column number to drop.
 
-    Returns:
+    OUTPUT:
-    pandas.DataFrame: The DataFrame with dropped columns.
+    - dataframe: DataFrame with dropped columns.
-    """
+    '''
 
     idx_col = list(range(col_nb-1, df.shape[1], col_nb)) 
     df_dropped = df.drop(idx_col, axis=1)
     df_dropped.columns = range(df_dropped.columns.size)
     return df_dropped
 
+
 def speed_vert(df, axis='y'):
-    """
+    '''
     Calculate the vertical speed of a DataFrame along a specified axis.
 
     Parameters:
-    df (DataFrame): The input DataFrame.
+    - df: dataframe. DataFrame of 2D coordinates.
-    axis (str): The axis along which to calculate the speed. Default is 'y'.
+    - axis (str): The axis along which to calculate the speed. Default is 'y'.
+
+    OUTPUT:
+    - DataFrame: The DataFrame containing the vertical speed values.
+    '''
 
-    Returns:
-    DataFrame: The DataFrame containing the vertical speed values.
-    """
     axis_dict = {'x':0, 'y':1, 'z':2}
     df_diff = df.diff()
     df_diff = df_diff.fillna(df_diff.iloc[1]*2)
@@ -91,45 +109,58 @@ def speed_vert(df, axis='y'):
     return df_vert_speed
 
 
-def interpolate_nans(col, kind):
+def speed_2D(df):
+    '''
+    Calculate the 2D speed of a DataFrame.
+
+    INPUTS:
+    - df: dataframe. DataFrame of 2D coordinates.
+
+    OUTPUT:
+    - DataFrame: The DataFrame containing the 2D speed values.
+    '''
+    
+    df_diff = df.diff()
+    df_diff = df_diff.fillna(df_diff.iloc[1]*2)
+    df_2Dspeed = pd.DataFrame([np.sqrt(df_diff.loc[:,2*k]*2 + df_diff.loc[:,2*k+1]*2) for k in range(int(df_diff.shape[1]*2))]).T
+    return df_2Dspeed
+
+
+def interpolate_zeros_nans(col, kind):
     '''
     Interpolate missing points (of value nan)
 
     INPUTS
-    - col pandas column of coordinates
+    - col: pandas column of coordinates
-    - kind 'linear', 'slinear', 'quadratic', 'cubic'. Default 'cubic'
+    - kind: 'linear', 'slinear', 'quadratic', 'cubic'. Default 'cubic'
 
     OUTPUT
-    - col_interp interpolated pandas column
+    - col_interp: interpolated pandas column
     '''
     
-    idx = col.index
+    mask = ~(np.isnan(col) | col.eq(0)) # true where nans or zeros
-    idx_good = np.where(np.isfinite(col))[0] #index of non zeros
+    idx_good = np.where(mask)[0]
-    if len(idx_good) == 10: return col
+    try: 
-    # idx_notgood = np.delete(np.arange(len(col)), idx_good)
+        f_interp = interpolate.interp1d(idx_good, col[idx_good], kind=kind, bounds_error=False)
-
+        col_interp = np.where(mask, col, f_interp(col.index))
-    if not kind: # 'linear', 'slinear', 'quadratic', 'cubic'
+        return col_interp 
-        f_interp = interpolate.interp1d(idx_good, col[idx_good], kind='cubic', bounds_error=False)
+    except:
-    else:
+        print('No good values to interpolate')
-        f_interp = interpolate.interp1d(idx_good, col[idx_good], kind=kind, bounds_error=False) # modified
+        return col
-    col_interp = np.where(np.isfinite(col), col, f_interp(idx)) #replace nans with interpolated values
-    col_interp = np.where(np.isfinite(col_interp), col_interp, np.nanmean(col_interp)) #replace remaining nans
-
-    return col_interp #, idx_notgood
 
 
 def find_highest_wrist_position(df_coords, wrist_index):
-    """
+    '''
     Find the frame with the highest wrist position in a list of coordinate DataFrames.
     Highest wrist position frame use for finding the fastest frame.
     
-    Args:
+    INPUT:
-    df_coords (list): List of coordinate DataFrames.
+    - df_coords (list): List of coordinate DataFrames.
-    wrist_index (int): The index of the wrist in the keypoint list.
+    - wrist_index (int): The index of the wrist in the keypoint list.
     
-    Returns:
+    OUTPUT:
-    list: The index of the frame with the highest wrist position.
+    - list: The index of the frame with the highest wrist position.
-    """
+    '''
 
     start_frames = []
     min_y_coords = []
@@ -149,20 +180,22 @@ def find_highest_wrist_position(df_coords, wrist_index):
 
     return start_frames, min_y_coords
 
+
 def find_motion_end(df_coords, wrist_index, start_frame, lowest_y, fps):
-    """
+    '''
     Find the frame where hands down movement ends.
     Hands down movement is defined as the time when the wrist moves down from the highest position.
 
-    Args:
+    INPUT:
-    df_coord (DataFrame): The coordinate DataFrame of the reference camera.
+    - df_coord (DataFrame): The coordinate DataFrame of the reference camera.
-    wrist_index (int): The index of the wrist in the keypoint list.
+    - wrist_index (int): The index of the wrist in the keypoint list.
-    start_frame (int): The frame where the hands down movement starts.
+    - start_frame (int): The frame where the hands down movement starts.
-    fps (int): The frame rate of the cameras in Hz.
+    - fps (int): The frame rate of the cameras in Hz.
+
+    OUTPUT:
+    - int: The index of the frame where hands down movement ends.
+    '''
 
-    Returns:
-    int: The index of the frame where hands down movement ends.
-    """
     y_col_index = wrist_index * 2 + 1
     wrist_y_values = df_coords.iloc[:, y_col_index].values # wrist y-coordinates
     highest_y_value = lowest_y
@@ -181,20 +214,21 @@ def find_motion_end(df_coords, wrist_index, start_frame, lowest_y, fps):
 
     return time
 
+
 def find_fastest_frame(df_speed_list):
-    """
+    '''
     Find the frame with the highest speed in a list of speed DataFrames.
     Fastest frame should locate in after highest wrist position frame.
     
-    Args:
+    INPUT:
-    df_speed_list (list): List of speed DataFrames.
+    - df_speed_list (list): List of speed DataFrames.
-    df_speed (DataFrame): The speed DataFrame of the reference camera.
+    - df_speed (DataFrame): The speed DataFrame of the reference camera.
-    fps (int): The frame rate of the cameras in Hz.
+    - fps (int): The frame rate of the cameras in Hz.
-    lag_time (float): The time lag in seconds.
+    - lag_time (float): The time lag in seconds.
 
-    Returns:
+    OUTPUT:
-    int: The index of the frame with the highest speed.
+    - int: The index of the frame with the highest speed.
-    """
+    '''
 
     for speed_series in df_speed_list:
         max_speed = speed_series.abs().max()
@@ -205,31 +239,25 @@ def find_fastest_frame(df_speed_list):
     return max_speed_index, max_speed
 
 
-def plot_time_lagged_cross_corr(camx, camy, ax, fps, lag_time, camx_max_speed_index, camy_max_speed_index):
+def plot_time_lagged_cross_corr(camx, camy, ax, fps, lag_time):
-    """
+    '''
     Calculate and plot the max correlation between two cameras with a time lag.
     How it works:
      1. Reference camera is camx and the other is camy. (Reference camera should record last. If not, the offset will be positive.)
      2. The initial shift alppied to camy to match camx is calculated.
      3. Additionally shift camy by max_lag frames to find the max correlation.
     
-    Args:
+    INPUT:
-    camx (pandas.Series): The speed series of the reference camera.
+    - camx: pd.Series. Speed series of the reference camera.
-    camy (pandas.Series): The speed series of the other camera.
+    - camy: pd.Series). Speed series of the other camera.
-    ax (matplotlib.axes.Axes): The axes to plot the correlation.
+    - ax: plt.axis. Plot correlation on second axis.
-    fps (int): The frame rate of the cameras in Hz.
+    - fps: int. Framerate of the cameras in Hz.
-    lag_time (float): The time lag in seconds.
+    - lag_time: float. Time lag in seconds.
-    camx_max_speed_index (int): The index of the frame with the highest speed in camx.
-    camy_max_speed_index (int): The index of the frame with the highest speed in camy.
 
-    Returns:
+    OUTPUT:
-    int: The offset value to apply to synchronize the cameras.
+    - offset: int. Offset value to apply to synchronize the cameras.
-    float: The maximum correlation value.
+    - max_corr: float. Maximum correlation value.
-    """
+    '''
-
-    # Initial shift of camy to match camx
-    # initial_shift = -(camy_max_speed_index - camx_max_speed_index) + fps
-    # camy = camy.shift(initial_shift).dropna()
 
     max_lag = int(fps * lag_time)
     pearson_r = []
@@ -238,7 +266,6 @@ def plot_time_lagged_cross_corr(camx, camy, ax, fps, lag_time, camx_max_speed_in
     for lag in lags:
         if lag < 0:
             shifted_camy = camy.shift(lag).dropna() # shift the camy segment by lag
-
             corr = camx.corr(shifted_camy) # calculate the correlation between the camx segment and the shifted camy segment
         elif lag == 0:
             corr = camx.corr(camy)
@@ -265,19 +292,19 @@ def plot_time_lagged_cross_corr(camx, camy, ax, fps, lag_time, camx_max_speed_in
 
 
 def apply_offset(offset, json_dirs, reset_sync, cam1_nb, cam2_nb):
-    """
+    '''
     Apply the offset to synchronize the cameras.
     Offset is always applied to the second camera.
     Offset would be always negative if the first camera is the last to start recording.
     Delete the camy json files from initial frame to offset frame.
 
-    Args:
+    INPUT:
-    offset (int): The offset value to apply to synchronize the cameras.
+    - offset (int): The offset value to apply to synchronize the cameras.
-    json_dirs (list): List of directories containing the JSON files for each camera.
+    - json_dirs (list): List of directories containing the JSON files for each camera.
-    reset_sync (bool): Whether to reset the synchronization by deleting the .del files.
+    - reset_sync (bool): Whether to reset the synchronization by deleting the .del files.
-    cam1_nb (int): The number of the reference camera.
+    - cam1_nb (int): The number of the reference camera.
-    cam2_nb (int): The number of the other camera.
+    - cam2_nb (int): The number of the other camera.
-    """ 
+    ''' 
 
     if offset == 0:
         print(f"Cams {cam1_nb} and {cam2_nb} are already synchronized. No offset applied.")
@@ -300,59 +327,60 @@ def apply_offset(offset, json_dirs, reset_sync, cam1_nb, cam2_nb):
                 os.rename(os.path.join(json_dir_to_offset, json_files[i]), os.path.join(json_dir_to_offset, json_files[i] + '.del'))
 
 
-
-#################
-# Main Function #
-#################
-
 def synchronize_cams_all(config_dict):
+    '''
     
-    #############
+    '''
-    # CONSTANTS #
-    #############
     
     # get parameters from Config.toml
     project_dir = config_dict.get('project').get('project_dir')
     pose_dir = os.path.realpath(os.path.join(project_dir, 'pose'))
     fps =  config_dict.get('project').get('frame_rate') # frame rate of the cameras (Hz)
     reset_sync = config_dict.get('synchronization').get('reset_sync')  # Start synchronization over each time it is run
+    id_kpt = 4
+    weights_kpt = 1
+    filter_order = 4
+    filter_cutoff = 6
+    vmax = 20 # px/s
 
-    # Vertical speeds (on 'Y')
+    # List json files
-    speed_kind = config_dict.get('synchronization').get('speed_kind') # this maybe fixed in the future
-    id_kpt =  config_dict.get('synchronization').get('id_kpt') #  get the numbers from the keypoint names in skeleton.py: 'RWrist' BLAZEPOSE 16, BODY_25B 10, BODY_25 4 ; 'LWrist' BLAZEPOSE 15, BODY_25B 9, BODY_25 7
-    weights_kpt = config_dict.get('synchronization').get('weights_kpt') # only considered if there are multiple keypoints.
-
-    ######################################
-    # 0. CONVERTING JSON FILES TO PANDAS #
-    ######################################
-
-    # Also filter, and then save the filtered data
     pose_listdirs_names = next(os.walk(pose_dir))[1]
     pose_listdirs_names.sort(key=lambda name: int(re.search(r'(\d+)', name).group(1)))
     json_dirs_names = [k for k in pose_listdirs_names if 'json' in k]
     json_dirs = [os.path.join(pose_dir, j_d) for j_d in json_dirs_names] # list of json directories in pose_dir
+    cam_nb = len(json_dirs)
 
-    # keypoints coordinates
+    # Extract, interpolate, and filter keypoint coordinates
     df_coords = []
+    b, a = signal.butter(filter_order/2, filter_cutoff/(fps/2), 'low', analog = False) 
     for i, json_dir in enumerate(json_dirs):
-        df_coords.append(convert_json2csv(json_dir))
+        df_coords.append(convert_json2pandas(json_dir))
         df_coords[i] = drop_col(df_coords[i],3) # drop likelihood
+        df_coords[i] = df_coords[i].apply(interpolate_zeros_nans, axis=0, args = ['cubic'])
+        df_coords[i] = df_coords[i].apply(loess_filter_1d, axis=0, args = [30])
+        df_coords[i] = pd.DataFrame(signal.filtfilt(b, a, df_coords[i], axis=0))
 
-    ## To save it and reopen it if needed
+
+
+
+    # Save keypoint coordinates to pickle
     with open(os.path.join(pose_dir, 'coords'), 'wb') as fp:
         pk.dump(df_coords, fp)
-    with open(os.path.join(pose_dir, 'coords'), 'rb') as fp:
+    # with open(os.path.join(pose_dir, 'coords'), 'rb') as fp:
-        df_coords = pk.load(fp)
+    #     df_coords = pk.load(fp)
 
-    #############################
+    # Compute vertical speed
-    # 1. COMPUTING SPEEDS       #
-    #############################
-
-    # Vitesse verticale    
     df_speed = []
-    for i in range(len(json_dirs)):
+    for i in range(cam_nb):
-        if speed_kind == 'y':
         df_speed.append(speed_vert(df_coords[i]))
+        # df_speed[i] = df_speed[i].where(abs(df_speed[i])<vmax, other=np.nan) # replaces by nan if jumps in speed
+        # df_speed[i] = df_speed[i].apply(interpolate_nans, axis=0, args = ['cubic'])
+
+
+    # Frame with maximum of the sum of absolute speeds
+    max_speed_frame = []
+    for i in range(cam_nb):
+        max_speed_frame += [np.argmax(abs(df_speed[i].sum(axis=1)))]
 
     #############################################
     # 2. PLOTTING PAIRED CORRELATIONS OF SPEEDS #
@@ -368,26 +396,31 @@ def synchronize_cams_all(config_dict):
     lowest_frames, lowest_y_coords = find_highest_wrist_position(df_coords, id_kpt)
 
     # set reference camera
-    ref_cam_nb = 0
+    nb_frames_per_cam = [len(d) for d in df_speed]
+    ref_cam_id = nb_frames_per_cam.index(min(nb_frames_per_cam))
+    
+    
     max_speeds = []
 
-    for cam_nb in range(1, len(json_dirs)):
+    
+    cam_list = list(range(cam_nb))
+    cam_list.pop(ref_cam_id)
+    for cam_id in cam_list:
         # find the highest wrist position for each camera
-        camx_start_frame = lowest_frames[ref_cam_nb]
+        camx_start_frame = lowest_frames[ref_cam_id]
-        camy_start_frame = lowest_frames[cam_nb]
+        camy_start_frame = lowest_frames[cam_id]
 
-        camx_lowest_y = lowest_y_coords[ref_cam_nb]
+        camx_lowest_y = lowest_y_coords[ref_cam_id]
-        camy_lowest_y = lowest_y_coords[cam_nb]
+        camy_lowest_y = lowest_y_coords[cam_id]
 
-        camx_time = find_motion_end(df_coords[ref_cam_nb], id_kpt[0], camx_start_frame, camx_lowest_y, fps)
+        camx_time = find_motion_end(df_coords[ref_cam_id], id_kpt[0], camx_start_frame, camx_lowest_y, fps)
-        camy_time = find_motion_end(df_coords[cam_nb], id_kpt[0], camy_start_frame, camy_lowest_y, fps)
+        camy_time = find_motion_end(df_coords[cam_id], id_kpt[0], camy_start_frame, camy_lowest_y, fps)
 
         camx_end_frame = camx_start_frame + int(camx_time * fps)
         camy_end_frame = camy_start_frame + int(camy_time * fps)
 
-        camx_segment = df_speed[ref_cam_nb].iloc[camx_start_frame:camx_end_frame+1, id_kpt[0]]
+        camx_segment = df_speed[ref_cam_id].iloc[camx_start_frame:camx_end_frame+1, id_kpt[0]]
-        camy_segment = df_speed[cam_nb].iloc[camy_start_frame:camy_end_frame+1, id_kpt[0]]
+        camy_segment = df_speed[cam_id].iloc[camy_start_frame:camy_end_frame+1, id_kpt[0]]
-
 
         # Find the fastest speed and the frame
         camx_max_speed_index, camx_max_speed = find_fastest_frame([camx_segment])
@@ -410,15 +443,15 @@ def synchronize_cams_all(config_dict):
         camy_end_frame = camy_max_speed_index + (fps) * (lag_time)
 
         if len(id_kpt) == 1 and id_kpt[0] != 'all':
-            camx = df_speed[ref_cam_nb].iloc[camx_start_frame:camx_end_frame+1, id_kpt[0]]
+            camx = df_speed[ref_cam_id].iloc[camx_start_frame:camx_end_frame+1, id_kpt[0]]
-            camy = df_speed[cam_nb].iloc[camy_start_frame:camy_end_frame+1, id_kpt[0]]
+            camy = df_speed[cam_id].iloc[camy_start_frame:camy_end_frame+1, id_kpt[0]]
         elif id_kpt == ['all']:
-            camx = df_speed[ref_cam_nb].iloc[camx_start_frame:camx_end_frame+1].sum(axis=1)
+            camx = df_speed[ref_cam_id].iloc[camx_start_frame:camx_end_frame+1].sum(axis=1)
-            camy = df_speed[cam_nb].iloc[camy_start_frame:camy_end_frame+1].sum(axis=1)
+            camy = df_speed[cam_id].iloc[camy_start_frame:camy_end_frame+1].sum(axis=1)
         elif len(id_kpt) == 1 and len(id_kpt) == len(weights_kpt):
             dict_id_weights = {i:w for i, w in zip(id_kpt, weights_kpt)}
-            camx = df_speed[ref_cam_nb] @ pd.Series(dict_id_weights).reindex(df_speed[ref_cam_nb].columns, fill_value=0)
+            camx = df_speed[ref_cam_id] @ pd.Series(dict_id_weights).reindex(df_speed[ref_cam_id].columns, fill_value=0)
-            camy = df_speed[cam_nb] @ pd.Series(dict_id_weights).reindex(df_speed[cam_nb].columns, fill_value=0)
+            camy = df_speed[cam_id] @ pd.Series(dict_id_weights).reindex(df_speed[cam_id].columns, fill_value=0)
             camx = camx.iloc[camx_start_frame:camx_end_frame+1]
             camy = camy.iloc[camy_start_frame:camy_end_frame+1]
         else:
@@ -431,8 +464,8 @@ def synchronize_cams_all(config_dict):
         f, ax = plt.subplots(2,1)
 
         # speed
-        camx.plot(ax=ax[0], label = f'cam {ref_cam_nb+1}')
+        camx.plot(ax=ax[0], label = f'cam {ref_cam_id+1}')
-        camy.plot(ax=ax[0], label = f'cam {cam_nb+1}')
+        camy.plot(ax=ax[0], label = f'cam {cam_id+1}')
         ax[0].set(xlabel='Frame',ylabel='Speed (pxframe)')
         ax[0].legend()
   
@@ -440,9 +473,9 @@ def synchronize_cams_all(config_dict):
         offset, max_corr = plot_time_lagged_cross_corr(camx, camy, ax[1], fps, lag_time, camx_max_speed_index, camy_max_speed_index)
         f.tight_layout()
         plt.show()
-        print(f'Using number{id_kpt} keypoint, synchronized camera {ref_cam_nb+1} and camera {cam_nb+1}, with an offset of {offset} and a max correlation of {max_corr}.')
+        print(f'Using number{id_kpt} keypoint, synchronized camera {ref_cam_id+1} and camera {cam_id+1}, with an offset of {offset} and a max correlation of {max_corr}.')
 
         # apply offset
-        apply_offset(offset, json_dirs, reset_sync, ref_cam_nb, cam_nb)
+        apply_offset(offset, json_dirs, reset_sync, ref_cam_id, cam_id)
 
 

Pose2Sim/triangulation.py

@@ -18,7 +18,8 @@ cameras are removed than a predefined minimum, triangulation is skipped for
 the point and this frame. In the end, missing values are interpolated.
 
 In case of multiple subjects detection, make sure you first run the 
-personAssociation module.
+personAssociation module. It will then associate people across frames by 
+measuring the frame-by-frame distance between them.
 
 INPUTS: 
 - a calibration file (.toml extension)
@@ -108,6 +109,96 @@ def interpolate_zeros_nans(col, *args):
     return col_interp
 
 
+def min_with_single_indices(L, T):
+    '''
+    Let L be a list (size s) with T associated tuple indices (size s).
+    Select the smallest values of L, considering that 
+    the next smallest value cannot have the same numbers 
+    in the associated tuple as any of the previous ones.
+
+    Example:
+    L = [  20,   27,  51,    33,   43,   23,   37,   24,   4,   68,   84,    3  ]
+    T = list(it.product(range(2),range(3)))
+      = [(0,0),(0,1),(0,2),(0,3),(1,0),(1,1),(1,2),(1,3),(2,0),(2,1),(2,2),(2,3)]
+
+    - 1st smallest value: 3 with tuple (2,3), index 11
+    - 2nd smallest value when excluding indices (2,.) and (.,3), i.e. [(0,0),(0,1),(0,2),X,(1,0),(1,1),(1,2),X,X,X,X,X]:
+    20 with tuple (0,0), index 0
+    - 3rd smallest value when excluding [X,X,X,X,X,(1,1),(1,2),X,X,X,X,X]:
+    23 with tuple (1,1), index 5
+    
+    INPUTS:
+    - L: list (size s)
+    - T: T associated tuple indices (size s)
+
+    OUTPUTS: 
+    - minL: list of smallest values of L, considering constraints on tuple indices
+    - argminL: list of indices of smallest values of L
+    - T_minL: list of tuples associated with smallest values of L
+    '''
+
+    minL = [np.nanmin(L)]
+    argminL = [np.nanargmin(L)]
+    T_minL = [T[argminL[0]]]
+    
+    mask_tokeep = np.array([True for t in T])
+    i=0
+    while mask_tokeep.any()==True:
+        mask_tokeep = mask_tokeep & np.array([t[0]!=T_minL[i][0] and t[1]!=T_minL[i][1] for t in T])
+        if mask_tokeep.any()==True:
+            indicesL_tokeep = np.where(mask_tokeep)[0]
+            minL += [np.nanmin(np.array(L)[indicesL_tokeep]) if not np.isnan(np.array(L)[indicesL_tokeep]).all() else np.nan]
+            argminL += [indicesL_tokeep[np.nanargmin(np.array(L)[indicesL_tokeep])] if not np.isnan(minL[-1]) else indicesL_tokeep[0]]
+            T_minL += (T[argminL[i+1]],)
+            i+=1
+    
+    return np.array(minL), np.array(argminL), np.array(T_minL)
+
+
+def sort_people(Q_kpt_old, Q_kpt):
+    '''
+    Associate persons across frames
+    Persons' indices are sometimes swapped when changing frame
+    A person is associated to another in the next frame when they are at a small distance
+    
+    INPUTS:
+    - Q_kpt_old: list of arrays of 3D coordinates [X, Y, Z, 1.] for the previous frame
+    - Q_kpt: idem Q_kpt_old, for current frame
+    
+    OUTPUT:
+    - Q_kpt_new: array with reordered persons
+    - personsIDs_sorted: index of reordered persons
+    '''
+    
+    # Generate possible person correspondences across frames
+    if len(Q_kpt_old) < len(Q_kpt):
+        Q_kpt_old = np.concatenate((Q_kpt_old, [[0., 0., 0., 1.]]*(len(Q_kpt)-len(Q_kpt_old))))
+    personsIDs_comb = sorted(list(it.product(range(len(Q_kpt_old)),range(len(Q_kpt)))))
+    
+    # Compute distance between persons from one frame to another
+    frame_by_frame_dist = []
+    for comb in personsIDs_comb:
+        frame_by_frame_dist += [euclidean_distance(Q_kpt_old[comb[0]][:3],Q_kpt[comb[1]][:3])]
+        
+    # sort correspondences by distance
+    minL, _, associated_tuples = min_with_single_indices(frame_by_frame_dist, personsIDs_comb)
+    # print('Distances :', minL)
+    
+    # associate 3D points to same index across frames, nan if no correspondence
+    Q_kpt_new, personsIDs_sorted = [], []
+    for i in range(len(Q_kpt_old)):
+        id_in_old =  associated_tuples[:,1][associated_tuples[:,0] == i].tolist()
+        # print('id_in_old ', i, id_in_old)
+        if len(id_in_old) > 0:
+            personsIDs_sorted += id_in_old
+            Q_kpt_new += [Q_kpt[id_in_old[0]]]
+        else:
+            personsIDs_sorted += [-1]
+            Q_kpt_new += [Q_kpt_old[i]]
+    
+    return Q_kpt_new, personsIDs_sorted, associated_tuples
+
+
 def make_trc(config, Q, keypoints_names, f_range, id_person=-1):
     '''
     Make Opensim compatible trc file from a dataframe with 3D coordinates
@@ -267,7 +358,7 @@ def recap_triangulate(config, error, nb_cams_excluded, keypoints_names, cam_excl
         logging.info(f'In average, {mean_cam_excluded} cameras had to be excluded to reach these thresholds.')
         
         cam_excluded_count[n] = {i: v for i, v in zip(cam_names, cam_excluded_count[n].values())}
-        cam_excluded_count[n] = {i: cam_excluded_count[n][i] for i in sorted(cam_excluded_count[n].keys())}
+        cam_excluded_count[n] = {k: v for k, v in sorted(cam_excluded_count[n].items(), key=lambda item: item[1])[::-1]}
         str_cam_excluded_count = ''
         for i, (k, v) in enumerate(cam_excluded_count[n].items()):
             if i ==0:
@@ -584,7 +675,7 @@ def triangulate_all(config):
     
      INPUTS: 
     - a calibration file (.toml extension)
-    - json files for each camera with only one person of interest
+    - json files for each camera with indices matching the detected persons
     - a Config.toml file
     - a skeleton model
     
@@ -655,9 +746,7 @@ def triangulate_all(config):
     # Prep triangulation
     f_range = [[0,min([len(j) for j in json_files_names])] if frame_range==[] else frame_range][0]
     frames_nb = f_range[1]-f_range[0]
-    
     nb_persons_to_detect = max([len(json.load(open(json_fname))['people']) for json_fname in json_tracked_files[0]])
-    
     n_cams = len(json_dirs_names)
     
     # Check that camera number is consistent between calibration file and pose folders
@@ -667,8 +756,14 @@ def triangulate_all(config):
                     and {n_cams} cameras based on the number of pose folders.')
     
     # Triangulation
+    Q = [[[np.nan]*3]*keypoints_nb for n in range(nb_persons_to_detect)]
+    Q_old = [[[np.nan]*3]*keypoints_nb for n in range(nb_persons_to_detect)]
+    error = [[] for n in range(nb_persons_to_detect)]
+    nb_cams_excluded = [[] for n in range(nb_persons_to_detect)]
+    id_excluded_cams = [[] for n in range(nb_persons_to_detect)]
     Q_tot, error_tot, nb_cams_excluded_tot,id_excluded_cams_tot = [], [], [], []
     for f in tqdm(range(frames_nb)):
+        # print(f'\nFrame {f}:')        
         # Get x,y,likelihood values from files
         json_tracked_files_f = [json_tracked_files[c][f] for c in range(n_cams)]
         # print(json_tracked_files_f)
@@ -692,12 +787,19 @@ def triangulate_all(config):
                 y_files[n][likelihood_files[n] < likelihood_threshold] = np.nan
                 likelihood_files[n][likelihood_files[n] < likelihood_threshold] = np.nan
         
+        # Q_old = Q except when it has nan, otherwise it takes the Q_old value
+        nan_mask = np.isnan(Q)
+        Q_old = np.where(nan_mask, Q_old, Q)
+        error_old, nb_cams_excluded_old, id_excluded_cams_old = error.copy(), nb_cams_excluded.copy(), id_excluded_cams.copy()
         Q = [[] for n in range(nb_persons_to_detect)]
         error = [[] for n in range(nb_persons_to_detect)]
         nb_cams_excluded = [[] for n in range(nb_persons_to_detect)]
         id_excluded_cams = [[] for n in range(nb_persons_to_detect)]
+        
         for n in range(nb_persons_to_detect):
             for keypoint_idx in keypoints_idx:
+            # keypoints_nb = 2
+            # for keypoint_idx in range(2):
             # Triangulate cameras with min reprojection error
                 # print('\n', keypoints_names[keypoint_idx])
                 coords_2D_kpt = np.array( (x_files[n][:, keypoint_idx], y_files[n][:, keypoint_idx], likelihood_files[n][:, keypoint_idx]) )
@@ -710,6 +812,30 @@ def triangulate_all(config):
                 nb_cams_excluded[n].append(nb_cams_excluded_kpt)
                 id_excluded_cams[n].append(id_excluded_cams_kpt)
         
+        if multi_person:
+            # reID persons across frames by checking the distance from one frame to another
+            # print('Q before ordering ', np.array(Q)[:,:2])
+            if f !=0:
+                Q, personsIDs_sorted, associated_tuples = sort_people(Q_old, Q)
+                # print('Q after ordering ', personsIDs_sorted, associated_tuples, np.array(Q)[:,:2])
+                
+                error_sorted, nb_cams_excluded_sorted, id_excluded_cams_sorted = [], [], []
+                for i in range(len(Q)):
+                    id_in_old =  associated_tuples[:,1][associated_tuples[:,0] == i].tolist()
+                    if len(id_in_old) > 0:
+                        personsIDs_sorted += id_in_old
+                        error_sorted += [error[id_in_old[0]]]
+                        nb_cams_excluded_sorted += [nb_cams_excluded[id_in_old[0]]]
+                        id_excluded_cams_sorted += [id_excluded_cams[id_in_old[0]]]
+                    else:
+                        personsIDs_sorted += [-1]
+                        error_sorted += [error[i]]
+                        nb_cams_excluded_sorted += [nb_cams_excluded[i]]
+                        id_excluded_cams_sorted += [id_excluded_cams[i]]
+                error, nb_cams_excluded, id_excluded_cams = error_sorted, nb_cams_excluded_sorted, id_excluded_cams_sorted
+        
+        # TODO: if distance > threshold, new person
+        
         # Add triangulated points, errors and excluded cameras to pandas dataframes
         Q_tot.append([np.concatenate(Q[n]) for n in range(nb_persons_to_detect)])
         error_tot.append([error[n] for n in range(nb_persons_to_detect)])
@@ -717,6 +843,13 @@ def triangulate_all(config):
         id_excluded_cams = [[id_excluded_cams[n][k] for k in range(keypoints_nb)] for n in range(nb_persons_to_detect)]
         id_excluded_cams_tot.append(id_excluded_cams)
             
+    # fill values for if a person that was not initially detected has entered the frame 
+    Q_tot = [list(tpl) for tpl in zip(*it.zip_longest(*Q_tot, fillvalue=[np.nan]*keypoints_nb*3))]
+    error_tot = [list(tpl) for tpl in zip(*it.zip_longest(*error_tot, fillvalue=[np.nan]*keypoints_nb*3))]
+    nb_cams_excluded_tot = [list(tpl) for tpl in zip(*it.zip_longest(*nb_cams_excluded_tot, fillvalue=[np.nan]*keypoints_nb*3))]
+    id_excluded_cams_tot = [list(tpl) for tpl in zip(*it.zip_longest(*id_excluded_cams_tot, fillvalue=[np.nan]*keypoints_nb*3))]
+
+    # dataframes for each person
     Q_tot = [pd.DataFrame([Q_tot[f][n] for f in range(frames_nb)]) for n in range(nb_persons_to_detect)]
     error_tot = [pd.DataFrame([error_tot[f][n] for f in range(frames_nb)]) for n in range(nb_persons_to_detect)]
     nb_cams_excluded_tot = [pd.DataFrame([nb_cams_excluded_tot[f][n] for f in range(frames_nb)]) for n in range(nb_persons_to_detect)]

README.md

@@ -17,7 +17,7 @@
 ##### N.B:. Please set undistort_points and handle_LR_swap to false for now since it currently leads to inaccuracies. I'll try to fix it soon.
 
 > **_News_: Version 0.7:**\
-> **Multi-person analysis is now supported!**\
+> **Multi-person analysis is now supported!** Latest version is 100 times faster that the one before, and also more robust.\
 > Team sports, combat sports, and ballroom dancing can now take advantage of Pose2Sim full potential.\
 > **Other recently added features**: Automatic batch processing, Marker augmentation, Blender visualization.
 <!-- Incidentally, right/left limb swapping is now handled, which is useful if few cameras are used;\
@@ -307,27 +307,34 @@ All AlphaPose models are supported (HALPE_26, HALPE_68, HALPE_136, COCO_133, COC
 > _**Cameras need to be synchronized, so that 2D points correspond to the same position across cameras.**_\
 ***N.B.:** Skip this step if your cameras are already synchronized.*
 
+Open an Anaconda prompt or a terminal in a `Session`, `Participant`, or `Trial` folder.\
+Type `ipython`.
+
 ``` python
 from Pose2Sim import Pose2Sim
 Pose2Sim.synchronization()
 ```
 
 Reference camera (usally cam1) should start record at last between whole cameras.\
-Set fps, id_kpt, weight_kpt, reset_sync in Config.toml.\
+
 **How to get perfect sync point**
-1. Set cameras position where they can see person wrist clearly.
+1. Set cameras position where they can see `id_kpt` (default: `RWrist`) clearly.
 2. Press record button, and what pressed last time to be reference camera.
 3. Walk to proper location( See 1 ).
 4. Raise your hands.
 5. Downward your hand fastly.
 
+Check printed output. If results are not satisfying, try and release the constraints in the [Config.toml](https://github.com/perfanalytics/pose2sim/blob/main/Pose2Sim/S00_Demo_Session/Config.toml) file.
+
 Alternatively, use a flashlight or a clap to synchronize them. GoPro cameras can also be synchronized with a timecode, by GPS (outdoors) or with a remote control (slightly less reliable).
 
+
+
 </br>
 
 ## Camera calibration
 > _**Calculate camera intrinsic properties and extrinsic locations and positions.\
-> Convert a preexisting calibration file, or calculate intrinsic and extrinsic parameters from scratch.**_ \
+> Convert a preexisting calibration file, or calculate intrinsic and extrinsic parameters from scratch.**_
 
 Open an Anaconda prompt or a terminal in a `Session`, `Participant`, or `Trial` folder.\
 Type `ipython`.