Research map / Video / Video-language understanding
DeformTrace: A Deformable State Space Model with Relay Tokens for Temporal Forgery Localization
Xiaodong Zhu, Suting Wang, Yuanming Zheng, Junqi Yang, Yangxu Liao, Yuhong Yang, Weiping Tu, Zhongyuan Wang
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 Deformable Self-SSM instead of vanilla SSM (ablation) | DeformTrace · AV-Deepfake1M test | mAP (%): 41.2 -> 44.4 | Improved |
| add Deformable Cross-SSM instead of vanilla SSM (ablation) | DeformTrace · AV-Deepfake1M test | mAP (%): 41.2 -> 47.8 | Improved |
| combine DS-SSM and DC-SSM (ablation) | DeformTrace · AV-Deepfake1M test | mAP (%): 47.8 -> 49.8 | Improved |
| add enhance loss on relay tokens (ablation) | DeformTrace · AV-Deepfake1M test | mAP (%): 49.8 -> 51.4 | Improved |
| add cooperation loss on relay tokens (ablation) | DeformTrace · AV-Deepfake1M test | mAR (%): 78.2 -> 78.9 | Improved |
| add both relay losses (ablation) | DeformTrace · AV-Deepfake1M test | mAP (%): 49.8 -> 52.9 | Improved |
| use 1 or 2 relay tokens instead of 8 (ablation) | DeformTrace · AV-Deepfake1M test | mAP | No clear effect |
| increase relay tokens beyond 8 (ablation) | DeformTrace · AV-Deepfake1M test | mAP | Hurt |
| add relay tokens on short videos (ablation) | DeformTrace · AV-Deepfake1M by duration | mAP | Mixed |
| replace Transformer blocks with SSM modules | DeformTrace · AV-Deepfake1M | mAP avg (%): 40.4 -> 52.9 | Improved |
| replace Transformer blocks with SSM modules | DeformTrace · LAV-DF | inference time (ms): 126 -> 104 | Improved |
| DeformTrace vs UMMAFormer on LAV-DF | DeformTrace · LAV-DF | mAP@0.75 (%): 95.5 -> 90.7 | Hurt |
Closest papers by meaning
- MG-RWKV: Multi-Grained Context-Aware RWKV for Temporal Forgery Localization (ECCV 2026 · similarity 0.87)
- GEM-TFL: Bridging Weak and Full Supervision for Forgery Localization through EM-Guided Decomposition and Temporal Refinement (CVPR 2026 · similarity 0.87)
- Toward Real-world Text Image Forgery Localization: Structured and Interpretable Data Synthesis (NeurIPS 2025 · similarity 0.84)
- SUMI-IFL: An Information-Theoretic Framework for Image Forgery Localization with Sufficiency and Minimality Constraints (AAAI 2025 · similarity 0.83)
- DiffForensics: Leveraging Diffusion Prior to Image Forgery Detection and Localization (CVPR 2024 · similarity 0.83)
- ForensicHub: A Unified Benchmark & Codebase for All-Domain Fake Image Detection and Localization (NeurIPS 2025 · similarity 0.83)
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.
