How tstanczyk95/mcbyteplusplus works
McByte++ is a training-free, tracking-by-detection system for long-term sports-video tracking. The user supplies frames and optionally detections; the demo runs detection, online association and identity management, then writes MOT records and visualizations. With re-ID enabled, it also uses appearance features to reconnect returning subjects; mask propagation and selective camera-motion compensation support the tracking process. YOLOX model code, the EdgeTAM-backed mask manager, and the re-ID package are represented as distinct internal layers. The README documents the overall workflow; some internal call ordering is not shown in the excerpts.
Main entry points: demo_track__with_reid.py Demo with re-ID · demo_track__no_reid.py Demo without re-ID · mcbyteplusplus_tracker__with_reid.py Online tracker with re-ID · yolo_head.py YOLOX detection head
Components 18
Demo entry
- Demo with re-ID
tools/demo_track__with_reid.py - Demo without re-ID
tools/demo_track__no_reid.py
Detection
- Detection input
- Detector configuration
yolox/exp/base_exp.py - YOLOX detection head
yolox/models/yolo_head.py
Tracking and identity
- Online tracker with re-ID
yolox/tracker/mcbyteplusplus_tracker__with_reid.py - Online tracker
yolox/tracker/mcbyteplusplus_tracker__no_reid.py - Motion filtering
yolox/tracker/kalman_filter.py - Track matching
yolox/tracker/matching.py - Camera motion compensation
yolox/tracker/gmc.py - Appearance features
deep-person-reid/torchreid/utils/feature_extractor.py
Mask propagation
- Mask manager
mask_propagation/mask_manager__edgetam.py - EdgeTAM predictor builder
mask_propagation/EdgeTAM/sam2/build_sam.py
Outputs
- Tracking visualization
yolox/utils/visualize.py - MOT records
Other components
- Video operator
- Input frames
- Offline global association
Connections 24
- Video operator → Demo with re-ID: runs
README.md - Video operator → Demo without re-ID: runs
README.md - Input frames → Demo with re-ID: provides frames
README.md - Input frames → Demo without re-ID: provides frames
README.md - Detection input → Demo with re-ID: optional detections
README.md - Demo with re-ID → Detector configuration: configures detector
tools/demo_track__with_reid.py - Demo with re-ID → YOLOX detection head: runs detection
- Demo without re-ID → YOLOX detection head: runs detection
- YOLOX detection head → Online tracker with re-ID: supplies detections
- YOLOX detection head → Online tracker: supplies detections
- Demo with re-ID → Online tracker with re-ID: runs tracking
tools/demo_track__with_reid.py - Demo without re-ID → Online tracker: runs tracking
README.md - Online tracker with re-ID → Motion filtering: uses motion filter
yolox/tracker/mcbyteplusplus_tracker__with_reid.py - Online tracker with re-ID → Track matching: uses matching
yolox/tracker/mcbyteplusplus_tracker__with_reid.py - Online tracker with re-ID → Camera motion compensation: uses camera compensation
yolox/tracker/mcbyteplusplus_tracker__with_reid.py - Online tracker with re-ID → Appearance features: extracts features
yolox/tracker/mcbyteplusplus_tracker__with_reid.py - Demo with re-ID → Mask manager: manages masks
tools/demo_track__with_reid.py - Demo without re-ID → Mask manager: manages masks
- Mask manager → EdgeTAM predictor builder: builds predictor
mask_propagation/mask_manager__edgetam.py - Demo with re-ID → Tracking visualization: renders tracks
tools/demo_track__with_reid.py - Demo without re-ID → Tracking visualization: renders tracks
README.md - Demo with re-ID → MOT records: writes MOT records
README.md - Demo without re-ID → MOT records: writes MOT records
README.md - MOT records → Offline global association: optional post-processing
README.md