Research map / Video / Video segmentation and tracking
Self-Supervised Video Copy Localization with Regional Token Representation
Minlong Lu, Yichen Lu, Siwei Nie, Xudong Yang, Xiaobo Zhang
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 Transitivity Property to generate extra copied video pairs (ablation) | self-supervised detector · VCSL test (RTR feature) | segment-level F-score: 66.96 -> 68.83 | Improved |
| add Transitivity Property pairs, effect on video-level F-score (ablation) | self-supervised detector · VCSL test (RTR feature) | video-level F-score: 95.08 -> 95.42 | Improved |
| add a Regional Token to ViT (ablation) | ViT-small feature extractor · VCSL test | segment-level F-score: 67.96 -> 68.83 | Improved |
| add a Regional Token to ViT (ablation) | ViT-small feature extractor · DISC | µAP: 55.39 -> 58.66 | Improved |
| add more tokens beyond one Regional Token (ablation) | ViT-small feature extractor · VCSL test | segment-level F-score: 68.83 -> 68.93 | No clear effect |
| add more tokens beyond one Regional Token (ablation) | ViT-small feature extractor · DISC | µAP: 58.66 -> 59.40 | Improved |
| use asymmetric two-stage training rather than symmetric single-stage (ablation) | Regional Token · VCSL PiP pairs (200 pairs) | Regional Token Hit Rate: 75% -> 94.36% | Improved |
| fine-tune detector with 1% of human-labeled VCSL data | Ours-ssl (RTR) · VCSL test | segment-level F-score: 68.83 -> 69.90 | Improved |
| train detector on self-supervised pairs instead of human labels | Ours (RTR) · VCSL test | segment-level F-score: 69.71 -> 68.83 | Mixed |
| replace dual-softmax similarity map with cosine similarity map | TransVCL · VCSL test | segment-level F-score: 66.51 -> 68.76 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TransVCL — TransVCL: Attention-Enhanced Video Copy Localization Network with Flexible Supervision (AAAI 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TransVCL: Attention-Enhanced Video Copy Localization Network with Flexible Supervision (AAAI 2023 · similarity 0.91)
- FiGVCL: Fine-Grained Benchmark and Method for Video Copy Localization (TPAMI 2025 · similarity 0.89)
- A Large-Scale Comprehensive Dataset and Copy-Overlap Aware Evaluation Protocol for Segment-Level Video Copy Detection (CVPR 2022 · similarity 0.89)
- Video Re-localization (ECCV 2018 · similarity 0.87)
- Efficient Long Video Tokenization via Coordinate-based Patch Reconstruction (CVPR 2025 · similarity 0.84)
- Dense Video Object Captioning from Disjoint Supervision (ICLR 2025 · similarity 0.84)
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.
