How kamkyu94/tracktrack works
TrackTrack is an online multi-object tracking research system for MOT17, MOT20, and DanceTrack. Its documented workflow runs YOLOX detection, FastReID, then Tracker; the sampled detector writes detections, and the tracker consumes detection pickles, updates tracks, applies dataset-specific post-processing, and evaluates validation results. FastReID supplies a re-identification training and inference framework, but the excerpts do not establish how its outputs connect to the tracker. Some subsystem links below are documented only at the repository workflow level.
Components 29
YOLOX Detection
- Detection Runner
- Experiment Config
- Detection Data
- YOLOX Head
- Detection Losses
- Network Blocks
- Detection Evaluator
- Distributed Launch
- Learning Scheduler
FastReID Embeddings
- ReID Training
- Training Framework
- ReID Data
- Configuration
- Embedding Head
- Feature Extraction
- Embedding Computer
Tracking Pipeline
- Tracking Runner
- Track Association
- Track State
- Kalman Filter
- Track Linker
- Link Dataset
- Tracking Utilities
- Gap Interpolation
Evaluation
- Evaluation Runner
- MOT Dataset
- Dataset Base
- Track Metrics
Other components
- Operator
Connections 25
- Operator → Detection Runner: runs first
- Detection Runner → Experiment Config: loads config
- Detection Runner → Distributed Launch: launches
- Detection Runner → Detection Evaluator: detects with
- Detection Runner → Tracking Runner: writes detections
- YOLOX Head → Detection Losses: uses
- YOLOX Head → Network Blocks: uses
- Operator → ReID Training: runs second
- ReID Training → Configuration: loads config
- ReID Training → Training Framework: launches
- Training Framework → ReID Data: builds loader
- Training Framework → Embedding Head: builds model
- Feature Extraction → Embedding Computer: uses
- Operator → Tracking Runner: runs third
- Tracking Runner → Track Association: updates tracks
- Tracking Runner → Track Linker: links tracks
- Tracking Runner → Link Dataset: creates dataset
- Tracking Runner → Tracking Utilities: sets parameters
- Tracking Runner → Gap Interpolation: interpolates gaps
- Track Association → Track State: creates tracks
- Track State → Kalman Filter: filters motion
- Tracking Runner → Evaluation Runner: evaluates results
- Evaluation Runner → MOT Dataset: loads dataset
- MOT Dataset → Dataset Base: extends
- Track Metrics → Dataset Base: uses utilities