Research map / Video / Action recognition
GEM-TFL: Bridging Weak and Full Supervision for Forgery Localization through EM-Guided Decomposition and Temporal Refinement
Xiaodong Zhu, Yuanming Zheng, Suting Wang, Junqi Yang, 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 Latent Attribute Decomposition (EM) to base MIL model (ablation) | GEM-TFL base · AV-Deepfake1M | average mAP: 5.12 -> 23.8 | Improved |
| add Temporal Consistency Refinement (ablation) | GEM-TFL +LAD · AV-Deepfake1M | average mAP: 23.8 -> 27.3 | Improved |
| add Graph-based Proposal Refinement (ablation) | GEM-TFL +LAD+TCR · AV-Deepfake1M | average mAP: 27.3 -> 31.9 | Improved |
| add Localization Phase regression (UMMAFormer) (ablation) | GEM-TFL +LAD+TCR+GPR · AV-Deepfake1M | average mAP: 31.9 -> 42.7 | Improved |
| use m=3 latent attributes instead of binary classification (ablation) | LAD · AV-Deepfake1M classification | AUC: 98.70 -> 99.07 | Improved |
| use m=1 or 2 latent attributes (ablation) | LAD · AV-Deepfake1M classification | AUC: 98.70 -> 98.58 (m=1), 98.48 (m=2) | Hurt |
| use m>3 latent attributes (ablation) | LAD · AV-Deepfake1M classification | AUC: 99.07 -> 98.90 (m=4) | Hurt |
| GEM-TFL vs best weak baseline PseudoFormer on LAV-DF | WS-TFL · LAV-DF | average mAP: 64.7 -> 77.6 | Improved |
| GEM-TFL vs WMMT on AV-Deepfake1M (WMMT uses richer quadruple labels) | WS-TFL · AV-Deepfake1M | average mAP: 34.3 -> 42.7 | Improved |
| GEM-TFL vs fully supervised MFMS | WS-TFL · AV-Deepfake1M | average mAP: 42.7 vs 94.1 | Hurt |
Closest papers by meaning
- Less Supervision, Better Generalization: Weakly Supervised Fake Region Localization in Diffusion-Edited Images (NeurIPS 2026 · similarity 0.87)
- Toward Real-world Text Image Forgery Localization: Structured and Interpretable Data Synthesis (NeurIPS 2025 · similarity 0.85)
- M²RL-Net: Multi-View and Multi-Level Relation Learning Network for Weakly-Supervised Image Forgery Detection (AAAI 2025 · similarity 0.85)
- Learning Discriminative Noise Guidance for Image Forgery Detection and Localization (AAAI 2024 · similarity 0.85)
- BusterX: MLLM-Powered AI-Generated Video Forgery Detection and Explanation (NeurIPS 2026 · similarity 0.84)
- SUMI-IFL: An Information-Theoretic Framework for Image Forgery Localization with Sufficiency and Minimality Constraints (AAAI 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.
