Pose2Sim - Markerless kinematics, from OpenPose to OpenSim

• MacOS support, finally! Automatic device and backend selection, display issues are fixed. Thanks a lot to @Satya.P and @mumu-create for thoroughly testing it out https://github.com/perfanalytics/pose2sim/issues/230

• Enhanced sorting algorithm:

  • Optionally predicts next frame position (simple frame-by-frame linear translation)
  • Optionally tracks centroid or bounding boxes instead of keypoints.

  N.B.: Default behavior remains the same, ie still sorts based on keypoints and without prediction. Prediction based on bounding boxes is recommended when people move in sustained straight lines and are likely to be occluded a lot. N.B.: I plan to switch to the RF-DETR detector instead of YOLOX for even better results.

• Others:

  • No need to provide all the parameters when running with a config dictionary (eg Pose2Sim.poseEstimation(config_dict)). The unspecified parameters are taken from the closest Config.toml file (eg Pose2Sim.poseEstimation(r'pose2sim\Pose2Sim\Demo_SinglePerson\Config.toml'))
  • You can now pass a .toml path instead of a dictionary or nothing
  • Fixed zero mass and inertial parameters of patella
  • Fixed issue when reading caliscope calibration files
  • Logs are saved project_dir rather than to the current directory
  • Copy button on Code block on website
  • Got rid of customtkinter, which cause issues when using along with PySide6 (interactive matplotlib plots in particular)
  • Minor fixes in synchronization, markerAugmentation, etc

Full Changelog: https://github.com/perfanalytics/pose2sim/compare/v0.10.47...v0.10.49