Research map / Detection and segmentation / Tracking
PS-MOT: Cultivating Instance Awareness from Point Seeds for Multi-Object Tracking
Kai Luo, Fei Teng, Mengfei Duan, Wanjun Jia, Xu Wang, Hao Shi, Kunyu Peng, Zhiyong Li, Kailun Yang
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 Temporal-Feedback Prompting (negative cues + motion priors) to the naive point-to-box baseline (ablation) | PS-Track (MOTIP) · DanceTrack val | HOTA: 30.3 -> 49.0 | Improved |
| add Uncertainty-Guided Gaussian Learning (ablation) | PS-Track with TFP · DanceTrack val | HOTA: 49.0 -> 49.4 | Improved |
| add Point-Excited Wavelet Attention (ablation) | PS-Track with TFP+UGL · DanceTrack val | HOTA: 49.4 -> 50.3 | Improved |
| stochastic PEWA activation during training (p in 0.2 to 0.8) instead of always on (ablation) | PS-Track (2-epoch schedule) · DanceTrack val | HOTA: 44.2 -> 43.3 (p=0.4) | Hurt |
| PEWA always on versus removed (p=1.0 vs p=0.0) (ablation) | PS-Track (2-epoch schedule) · DanceTrack val | HOTA: 43.7 -> 44.2 | Improved |
| add Gaussian noise to training points, up to 16 px (ablation) | PS-Track (2-epoch schedule) · DanceTrack val | HOTA: 44.2 -> 44.1 | No clear effect |
| increase point noise to 48 px (ablation) | PS-Track (2-epoch schedule) · DanceTrack val | HOTA: 44.2 -> 39.8 | Hurt |
| vary regression loss weight lambda_reg from 1 to 2, 3, 4, 5 (ablation) | PS-Track (2-epoch schedule) · DanceTrack val | HOTA: 44.2 (1) / 42.7 (2) / 44.8 (3) / 40.8 (4) / 43.7 (5) | Mixed |
| use stronger SAM variant (v1 -> v2 -> v3) for offline pseudo-labels (ablation) | PS-Track · DanceTrack val | HOTA: 20.7 -> 39.0 -> 50.3 | Improved |
| train YOLOX with PS-Track plus BYTE instead of Point2RBox-v3 detector with off-the-shelf associators | TBD pipeline · DanceTrack (point-supervised detector, TBD) | HOTA: 10.8 -> 42.9 | Improved |
| point supervision via PS-Track instead of box supervision | MOTIP · DanceTrack val | HOTA: 62.2 -> 50.3 | Hurt |
| point supervision via PS-Track instead of box supervision | BYTE · DanceTrack val | HOTA: 47.1 -> 42.9 | Hurt |
Closest papers by meaning
- Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · similarity 0.88)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · similarity 0.88)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.88)
- M3SOT: Multi-Frame, Multi-Field, Multi-Space 3D Single Object Tracking (AAAI 2024 · similarity 0.87)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.87)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.87)
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.
