Research map / Video / Video-language understanding
HumanSAM: Classifying Human-centric Forgery Videos in Human Spatial, Appearance, and Motion Anomaly
Chang Liu, Yunfan Ye, Fan Zhang, Qingyang Zhou, Yuchuan Luo, Zhiping Cai
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 |
|---|---|---|---|
| use HumanSAM dual-branch model instead of TimeSformer for 4-class forgery classification | HumanSAM · HFV multi-class | ACC (%): 64.0 -> 69.8 | Improved |
| add the Cat module combining attention-pooled and average-pooled features (ablation) | InternVideo2 backbone · HFV multi-class | ACC (%): 63.3 -> 65.6 ACC; 87.4 -> 86.3 AUC | Mixed |
| add monocular depth branch (Depth Pro) (ablation) | InternVideo2 + Cat · HFV multi-class | ACC (%): 65.6 -> 69.2 | Improved |
| add rank-based confidence loss (ablation) | InternVideo2 + Cat + Depth · HFV multi-class | ACC (%): 69.2 -> 69.8 | Improved |
| train on multi-class labels instead of binary on weak detectors (Uni-FD) | Uni-FD · HFV binary | ACC (%): 80.9 -> 92.3 | Improved |
| train on multi-class labels instead of binary on strong detectors | TimeSformer, MM-Det · HFV binary | ACC (%): 95.2 -> 94.8 (TimeSformer) | No clear effect |
| apply Gaussian blur sigma=3 to test videos | HumanSAM · CogVideoX-5B multi-class | AUC (%): 92.1 -> 88.4 | Hurt |
| apply JPEG or resize post-processing to test videos | HumanSAM · CogVideoX-5B multi-class | AUC (%): 92.1 -> 91.8 (JPEG), 91.9 (resize) | No clear effect |
| train on MiniMax or Kling forgeries instead of lower-ranked CogVideoX-2B (ablation) | HumanSAM · HFV multi-class | ACC (%): 69.8 -> 66.5 (MinMax), 64.1 (Kling) | Hurt |
| HumanSAM versus MM-Det on Sora external dataset | HumanSAM · Sora | ACC (%): 81.0 -> 95.3 | Improved |
Closest papers by meaning
- On Learning Multi-Modal Forgery Representation for Diffusion Generated Video Detection (NeurIPS 2024 · similarity 0.87)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.87)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.86)
- Forensics-Bench: A Comprehensive Forgery Detection Benchmark Suite for Large Vision Language Models (CVPR 2025 · similarity 0.86)
- BusterX: MLLM-Powered AI-Generated Video Forgery Detection and Explanation (NeurIPS 2026 · similarity 0.86)
- Face Forgery Video Detection via Temporal Forgery Cue Unraveling (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.
