Research map / Video / Video-language understanding
Explainable Forensics of Manipulated Segments in Untrimmed Long Videos
Yue Feng, Jingjing Li, Qijia Lu, Wei Ji, Jingrou Zhang, Fei Shen, Xiao Li, Yizhen Jia, Qiang Chen, Limin Wang, Wentong Li, Jie Qin
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 |
|---|---|---|---|
| fine-tune on TASLE | BusterX++ · TASLE in-domain | F1Det: 8.3 -> 33.6 | Improved |
| use four-class boundary-aware labels instead of binary real/fake (ablation) | DeMamba · TASLE in-domain | F1Det: 54.9 -> 67.5 | Improved |
| add MSLoc-PG proposals before Trace (ablation) | Trace · TASLE in-domain | F1Loc: 41.4 -> 70.9 | Improved |
| use DeMamba proposals instead of none for Trace (ablation) | Trace · TASLE in-domain | F1Loc: 41.4 -> 59.1 | Improved |
| add boundary-focused MSLoc-PR refinement over Trace w. MSLoc-PG (ablation) | MSLoc · TASLE in-domain | F1Loc: 70.9 -> 72.2 | Improved |
| add MLLM refinement over MSLoc-PG on out-of-domain data (ablation) | MSLoc · TASLE out-of-domain | F1Det / F1Loc: +7.8 F1Det, +17.6 F1Loc | Improved |
| add DAM (boundary dense sampling) to backbone (ablation) | MSLoc-PR · TASLE test | F1Loc / RQ: F1Loc 62.3 -> 63.0; RQ 3.77 -> 3.56 | Mixed |
| add EAM on top of DAM (ablation) | MSLoc-PR · TASLE test | RQ: 3.56 -> 3.79 | Improved |
| add anomaly-aware loss LAA (ablation) | MSLoc-PR · TASLE test | RQ: 3.79 -> 3.99 | Improved |
| reduce boundary frames from 16 to 8 (ablation) | MSLoc-PR · TASLE test | F1Det / F1Loc / RQ: 64.0 -> 63.2 F1Det; 63.8 -> 63.0 F1Loc | Hurt |
| widen boundary region phi from 10% to 30% (ablation) | MSLoc-PR · TASLE test | RQ: 3.80 -> 4.01 | Improved |
| widen boundary region phi to 30% for localization (ablation) | MSLoc-PR · TASLE test | F1Loc: 64.0 -> 63.7 | No clear effect |
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.
- BusterX++ — BusterX: MLLM-Powered AI-Generated Video Forgery Detection and Explanation (NeurIPS 2026) · better on 4 of 4 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
- BusterX: MLLM-Powered AI-Generated Video Forgery Detection and Explanation (NeurIPS 2026 · similarity 0.88)
- Seeing What Matters: Generalizable AI-generated Video Detection with Forensic-Oriented Augmentation (NeurIPS 2025 · similarity 0.86)
- Locate-Then-Examine: Grounded Region Reasoning Improves Detection of AI-Generated Images (CVPR 2026 · similarity 0.85)
- FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models (ICLR 2025 · similarity 0.85)
- Less Supervision, Better Generalization: Weakly Supervised Fake Region Localization in Diffusion-Edited Images (NeurIPS 2026 · similarity 0.85)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.85)
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.
