Research map / Detection and segmentation / Tracking
History-Aware Transformation of ReID Features for Multiple Object Tracking
Ruopeng Gao, Yuyao Wang, Chunxu Liu, Limin Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add history-aware FLD transformation (ablation) | FastReID-MOT · DanceTrack val (YOLOX) | HOTA: 51.1 -> 57.7 | Improved |
| add history-aware FLD transformation (ablation) | FastReID-MOT · SportsMOT val (YOLOX) | HOTA: 73.7 -> 81.1 | Improved |
| use PCA transformation instead of FLD (d'=N_id-1) (ablation) | FastReID-MOT · DanceTrack val (YOLOX) | HOTA: 51.1 -> 43.5 | Hurt |
| use PCA transformation (d'=N_id-1) with oracle detections (ablation) | FastReID-MOT · DanceTrack val (oracle) | HOTA: 74.9 -> 56.3 | Hurt |
| fit FLD on history instead of supervised training labels (ablation) | FastReID-MOT · DanceTrack val | HOTA: 51.4 -> 57.7 | Improved |
| fit FLD on history instead of supervised training labels (ablation) | FastReID-MOT · SportsMOT val | HOTA: 77.6 -> 81.1 | Improved |
| limit history length to T=60 instead of unlimited (ablation) | HAT-FastReID-MOT · DanceTrack val | HOTA: 55.3 -> 57.7 | Improved |
| use only T=10 recent features instead of 60 (ablation) | HAT-FastReID-MOT · DanceTrack val | HOTA: 57.7 -> 54.3 | Hurt |
| add temporally-weighted trajectory centroid (lambda0=0.9) (ablation) | HAT-FastReID-MOT · DanceTrack val | HOTA: 57.7 -> 59.3 | Improved |
| add knowledge integration with original space (alpha=0.9) (ablation) | HAT-FastReID-MOT · DanceTrack val | HOTA: 59.3 -> 60.6 | Improved |
| lower alpha to 0.4 (more weight on original space) (ablation) | HAT-FastReID-MOT · DanceTrack val | HOTA/MOTA: HOTA 59.3 -> 58.7, MOTA 86.9 -> 87.9 | Mixed |
| add HAT to MASA-R50 | MASA-R50 · DanceTrack test | HOTA: 50.8 -> 54.3 | Improved |
Closest papers by meaning
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.88)
- From Detection to Association: Learning Discriminative Object Embeddings for Multi-Object Tracking (CVPR 2026 · similarity 0.88)
- Looking Beyond Two Frames: End-to-End Multi-Object Tracking Using Spatial and Temporal Transformers (TPAMI 2022 · similarity 0.87)
- Dual-Path Temporal Decoder for End-to-End Multi-Object Tracking (NeurIPS 2025 · similarity 0.87)
- COVTrack++: Learning Open-Vocabulary Multi-Object Tracking from Continuous Videos via a Synergistic Paradigm (TPAMI 2026 · similarity 0.86)
- Dual-level Adaptation for Multi-Object Tracking: Building Test-Time Calibration from Experience and Intuition (CVPR 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
