Research map / Detection and segmentation / Open-set and few-shot detection
Track, Check, Repeat: An EM Approach to Unsupervised Tracking
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Iterate from Round1 to Round3 of EM pseudo-labeling and detector retraining (ablation) | Track, Check, Repeat · KITTI (2D), mAP@0.7 | mAP@0.7: 0.01 -> 0.21 | Helped |
| Iterate from Round1 to Round3 of EM pseudo-labeling and detector retraining (ablation) | Track, Check, Repeat · CATER (BEV), mAP@0.4 | mAP@0.4: 0.57 -> 0.76 | Helped |
| Remove the ensemble agreement check when promoting predictions to pseudo-labels (ablation) | Track, Check, Repeat · KITTI (BEV), Round2, mAP@0.5 | mAP@0.5: 0.28 -> 0.17 | Hurt |
| Add the trajectory library to short-range tracking (ablation) | Track, Check, Repeat · CATER long-range tracking | Recall: 0.53 -> 0.64 | Helped |
| Add the trajectory library to short-range tracking (ablation) | Track, Check, Repeat · CATER long-range tracking | Precision: 0.94 -> 0.91 | Hurt |
| Fine-tune the pretrained optical flow model with self-supervision (ablation) | RAFT optical flow module · Optical flow estimation | accuracy | No clear effect |
| Use ImageNet pretraining for the 2D ResNet segmentor (ablation) | 2D objectness segmentation · CATER and KITTI | performance | No clear effect |
| Use heavy color, scale, translation, and occlusion augmentation when training detectors | Track, Check, Repeat · Unlabelled RGB-D video object detection | generalization to unlabelled data | Helped |
| Add a trajectory library for tracking through partial and full occlusions (ablation) | Track, Check, Repeat · CATER partially and fully occluded objects | tracking recall | Helped |
Closest papers by meaning
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.89)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.88)
- GMOS: Grounding Moving Object Segmentation in 3D Space and Time (NeurIPS 2026 · similarity 0.87)
- Leveraging Temporal Cues for Semi-Supervised Multi-View 3D Object Detection (CVPR 2025 · similarity 0.87)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.87)
- Single-Model and Any-Modality for Video Object Tracking (CVPR 2024 · 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.