techainer-ocsort 0.1

"Horus's general object tracker"

This is a fork from OC_SORT. Only the main implementations are used.

Adjustments

• Map 1-1 each input detection result with a tracked result so a detection will receive either a valid value or None. Check map_result variable in tracker.py for detailed implementation.
• Only accept post-processed detections (e.g: boxes are scaled to original image size).
• Adjust stale tracklet removal logic by using 2 new arguments:
  • hit_inertia_min: minimum number of hits for a tracklet to be considered alive. hits will be initiated with hit_inertia_min+1. A tracklet will update hits depending on input detection. Check KalmanBoxTracker.update and KalmanBoxTracker.is_alive function in tracker.py for detailed implementation.
  • hit_inertia_max: hard cap of hits.

Installation

pip install techainer-ocsort

Example

• Prepare a video for face detection. Reconfig video name accordingly in tests/test_ocsort.py.
• Replace our face detection with yours in tests/test_ocsort.py. Make sure that tracker input format is [[x1,y1,x2,y2,score],[x1,y1,x2,y2,score],...].
• Run python tests/test_ocsort.py. Output will be saved in your configed video name.