How nezarsheikhi/resource-efficient-mtmc works
This is a Python multi-camera people-tracking system for assigning consistent global identities across non-overlapping video feeds. Users provide videos and tracking settings through the CLI, web UI, or Python API; the pipeline runs per-camera detection and tracking, extracts appearance embeddings, and matches them through a shared identity gallery. Results include per-camera MOT tracks, optional annotated videos, run metadata, and evaluation metrics. The README also documents model-weight and example-video downloads. Several subsystem links are supported by the source index rather than full excerpts, so those relationships are limited to its import references and README statements.
Main entry points: pipeline.py Tracking pipeline · cli.py CLI · camera.py Camera worker · ui.py Web UI
Components 20
User entry points
- CLI
src/mtmc_tracking/cli.py - Web UI
src/mtmc_tracking/ui.py - Python API
src/mtmc_tracking/__init__.py
Tracking and identity
- Tracking pipeline
src/mtmc_tracking/pipeline.py - Camera worker
src/mtmc_tracking/camera.py - YOLO and ByteTrack
- Global gallery
src/mtmc_tracking/gallery.py - Box geometry
src/mtmc_tracking/geometry.py - Motion prediction
src/mtmc_tracking/kalman.py
Models and settings
- Configuration
src/mtmc_tracking/config.py - Config presets
src/mtmc_tracking/configs - Weight manager
src/mtmc_tracking/weights.py - Re-ID extractor
src/mtmc_tracking/reid.py - OSNet backbone
src/mtmc_tracking/osnet.py
Results and evaluation
- Track outputs
src/mtmc_tracking/pipeline.py - Run summary
src/mtmc_tracking/summary.py - Metrics evaluation
src/mtmc_tracking/evaluation.py
Other components
- Operator
- Model releases
- Ground truth
Connections 29
- Operator → CLI: runs commands
README.md - Operator → Web UI: uses interface
README.md - Operator → Python API: calls API
README.md - CLI → Configuration: loads settings
src/mtmc_tracking/cli.py - CLI → Tracking pipeline: starts tracking
src/mtmc_tracking/cli.py - CLI → Metrics evaluation: runs evaluation
src/mtmc_tracking/cli.py - CLI → Web UI: launches UI
src/mtmc_tracking/cli.py - CLI → Weight manager: downloads weights
src/mtmc_tracking/cli.py - Web UI → Tracking pipeline: starts tracking
src/mtmc_tracking/ui.py - Web UI → Configuration: loads settings
src/mtmc_tracking/ui.py - Web UI → Run summary: formats results
src/mtmc_tracking/ui.py - Web UI → Weight manager: loads weights
src/mtmc_tracking/ui.py - Python API → Tracking pipeline: runs tracking
README.md - Python API → Metrics evaluation: evaluates tracks
README.md - Configuration → Config presets: loads presets
src/mtmc_tracking/config.py - Weight manager → Model releases: downloads files
src/mtmc_tracking/weights.py - Tracking pipeline → Camera worker: starts workers
src/mtmc_tracking/pipeline.py - Tracking pipeline → Global gallery: shares identities
src/mtmc_tracking/pipeline.py - Tracking pipeline → Re-ID extractor: creates extractor
src/mtmc_tracking/pipeline.py - Tracking pipeline → Track outputs: writes tracks
src/mtmc_tracking/pipeline.py - Camera worker → YOLO and ByteTrack: detects and tracks
README.md - Camera worker → Re-ID extractor: extracts features
src/mtmc_tracking/camera.py - Camera worker → Global gallery: requests identities
src/mtmc_tracking/camera.py - Camera worker → Box geometry: uses box geometry
src/mtmc_tracking/camera.py - Camera worker → Motion prediction: predicts motion
src/mtmc_tracking/camera.py - Re-ID extractor → OSNet backbone: uses backbone
src/mtmc_tracking/reid.py - Track outputs → Metrics evaluation: provides predictions
README.md - Ground truth → Metrics evaluation: supplies reference
README.md - Tracking pipeline → Run summary: summarizes run
src/mtmc_tracking/summary.py