Research map / Detection and segmentation / Tracking
Unsupervised Deep Representation Learning for Real-Time Tracking
Ning Wang, Wengang Zhou, Yibing Song, Chao Ma, Wei Liu, Houqiang Li
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 trajectory enlargement (TE) (ablation) | UDT · OTB-2015 | AUC (%): 59.4 -> 59.8 | Improved |
| add RoI selection (RS) (ablation) | UDT · OTB-2015 | AUC (%): 59.4 -> 60.0 | Improved |
| add both TE and RS (ablation) | UDT · OTB-2015 | AUC (%): 59.4 -> 60.2 | Improved |
| use 3 frames instead of 2 frames validation (ablation) | LUDT · OTB-2015 | AUC (%): 57.4 -> 59.4 | Improved |
| use 4 frames instead of 3 frames validation (ablation) | LUDT · OTB-2015 | AUC (%): 59.4 -> 59.8 | Improved |
| use 5 frames instead of 4 frames validation (ablation) | LUDT · OTB-2015 | AUC (%): 59.8 -> 59.7 | No clear effect |
| add hard sample reweighing (A_motion) (ablation) | LUDT · OTB-2015 | DP / AUC (%): drops about 1.5% DP and 1% AUC without it | Improved |
| train with forward-backward unsupervised loss instead of encoder-decoder (ablation) | DCFNet · OTB-2015 | AUC (%): 54.5 -> 60.2 | Improved |
| use unsupervised deep features instead of HOG (ablation) | DCFNet · OTB-2015 | AUC (%): 52.1 -> 60.2 | Improved |
| use RoI selection instead of center cropping (ablation) | LUDT · OTB-2015 | AUC (%): 59.4 -> 60.0 | Improved |
| use full supervision instead of unsupervised RoI selection | LUDT · OTB-2015 | AUC (%): 60.0 -> 62.6 | Improved |
| few-shot fine-tune on first 5 frames of OTB videos (ablation) | LUDT · OTB-2015 | AUC (%): 60.2 -> 61.5 | Improved |
Closest papers by meaning
- Unsupervised Deep Tracking (CVPR 2019 · similarity 0.97)
- Unsupervised Learning of Accurate Siamese Tracking (CVPR 2022 · similarity 0.93)
- Learning To Track Objects From Unlabeled Videos (ICCV 2021 · similarity 0.93)
- Progressive Unsupervised Learning for Visual Object Tracking (CVPR 2021 · similarity 0.93)
- Learning a Visual Tracker from a Single Movie without Annotation (AAAI 2019 · similarity 0.93)
- 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.
