Research map / Detection and segmentation / Tracking
Learning To Track Objects From Unlabeled Videos
Jilai Zheng, Chao Ma, Houwen Peng, Xiaokang 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 |
|---|---|---|---|
| generate pseudo boxes from optical flow and dynamic programming instead of random boxes (ablation) | USOT* · VOT2017/18 | EAO: 0.195 -> 0.263 | Improved |
| add naive Siamese tracker initialization before cycle memory training (ablation) | USOT* · VOT2017/18 | EAO: 0.306 -> 0.344 | Improved |
| add naive Siamese tracker initialization before cycle memory training (ablation) | USOT* · VOT2017/18 | accuracy (A): 0.575 -> 0.578 | No clear effect |
| add GOT-10k to VID as training data (ablation) | USOT* · VOT2017/18 | EAO: 0.315 -> 0.320 | Improved |
| add LaSOT to VID and GOT-10k training data (ablation) | USOT* · VOT2017/18 | EAO: 0.320 -> 0.337 | Improved |
| add YouTube-VOS to training data (ablation) | USOT* · VOT2017/18 | EAO: 0.337 -> 0.344 | Improved |
| set online weight w to 0.7 instead of 0.3 at queue length 7 (ablation) | USOT* · VOT2017/18 | EAO: 0.310 -> 0.344 | Improved |
| set memory queue length to 7 instead of 5 at w=0.7 (ablation) | USOT* · VOT2017/18 | EAO: 0.313 -> 0.344 | Improved |
| set memory queue length to 8 instead of 7 at w=0.7 (ablation) | USOT* · VOT2017/18 | EAO: 0.344 -> 0.341 | No clear effect |
| use supervised ImageNet backbone pretraining (USOT* vs USOT) | USOT · VOT2017/18 | EAO: 0.290 -> 0.344 | Improved |
| use supervised ImageNet backbone pretraining (USOT* vs USOT) | USOT · OTB2015 | AUC success: 58.9 -> 57.4 | Hurt |
| use supervised ImageNet backbone pretraining (USOT* vs USOT) | USOT · VOT2020 | EAO: 0.222 -> 0.219 | No clear effect |
Closest papers by meaning
- Unsupervised Deep Representation Learning for Real-Time Tracking (IJCV 2020 · similarity 0.93)
- Unsupervised Deep Tracking (CVPR 2019 · similarity 0.92)
- Learning a Visual Tracker from a Single Movie without Annotation (AAAI 2019 · similarity 0.91)
- Unsupervised Learning of Accurate Siamese Tracking (CVPR 2022 · similarity 0.91)
- Track, Check, Repeat: An EM Approach to Unsupervised Tracking (CVPR 2021 · similarity 0.90)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 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.
