Research map / Label-efficient and robust learning / Recognition and long-tail
Divide and Conquer: Reliable Multi-View Evidential Learning for Deepfake Detection
Xiaolu Kang, Zhongyuan Wang, Jikang Cheng, Baojin Huang, Zhanhe Lei, Gang Wu, Qin Zou, Qian 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 orthogonal projection to purify the residual artifact view (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.922 -> 0.938 | Improved |
| add artifact view alongside semantic view (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.905 -> 0.938 | Improved |
| use heterogeneous artifact view instead of duplicate semantic view (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.917 -> 0.938 | Improved |
| use evidential loss with Dempster-Shafer fusion instead of cross-entropy with mean fusion (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.924 -> 0.938 | Improved |
| replace mean fusion with Dempster-Shafer fusion (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.927 -> 0.938 | Improved |
| add cosine alignment loss to the manifold projector (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.921 -> 0.938 | Improved |
| add KL divergence regularization to the beta-VAE (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.922 -> 0.938 | Improved |
| use CLIP ViT-L/14 instead of ViT-B/16 (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.829 -> 0.938 | Improved |
| use CLIP instead of DINOv2 ViT-L as backbone (ablation) | DiCoME · Ablation avg (CDFv2, DFDC, MFS) | AUC: 0.808 -> 0.938 | Improved |
| use evidential Dempster-Shafer framework instead of baseline (D) (ablation) | DiCoME · ECE avg (CDFv2, DFD, MFS) | expected calibration error: 0.16 -> 0.07 | Improved |
| DiCoME vs GenD on cross-dataset average | DiCoME · Cross-dataset avg | video-level AUC: 0.923 -> 0.939 | Improved |
Closest papers by meaning
- Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning (ICLR 2026 · similarity 0.88)
- Critical Forgetting-Based Multi-Scale Disentanglement for Deepfake Detection (AAAI 2025 · similarity 0.86)
- Pixels Don't Lie (But Your Detector Might): Bootstrapping MLLM-as-a-Judge for Trustworthy Deepfake Detection and Reasoning Supervision (CVPR 2026 · similarity 0.86)
- PRPO: Paragraph-level Policy Optimization for Vision-Language Deepfake Detection (ICML 2026 · similarity 0.86)
- Exposing the Deception: Uncovering More Forgery Clues for Deepfake Detection (AAAI 2024 · similarity 0.86)
- Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector (CVPR 2025 · similarity 0.86)
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.
