Research map / Video / Video segmentation and tracking
Hierarchical Contrastive Inconsistency Learning for Deepfake Video Detection
Zhihao Gu, Taiping Yao, Yang Chen, Shouhong Ding, Lizhuang Ma
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 RAI module to ResNet-50 baseline (ablation) | ResNet-50 · FF++ NT | accuracy: 87.50% -> 92.14% | Improved |
| add RAI module to ResNet-50 baseline (ablation) | ResNet-50 · FF++ F2F | accuracy: 88.93% -> 93.21% | Improved |
| add ISF module to ResNet-50 baseline (ablation) | ResNet-50 · FF++ NT | accuracy: 0.8750 -> 0.9178 | Improved |
| add RAI and ISF together (ablation) | ResNet-50 · FF++ NT | accuracy: 0.9214 -> 0.9250 | Improved |
| add hierarchical contrast losses Ll and Lg to baseline features (ablation) | ResNet-50 · FF++ F2F | accuracy: 88.93% -> 91.43% | Improved |
| add hierarchical contrast on top of RAI and ISF (all components) (ablation) | HCIL · FF++ NT | accuracy: 0.9250 -> 0.9464 | Improved |
| add snippet-level contrast (vanilla NCE) (ablation) | HCIL · FF++ F2F | accuracy: 0.9464 -> 0.9536 | Improved |
| use weighted NCE (beta) instead of vanilla NCE (ablation) | HCIL · FF++ F2F | accuracy: 0.9536 -> 0.9571 | Improved |
| use weighted NCE instead of vanilla NCE (ablation) | HCIL · Celeb-DF | accuracy: 0.9981 -> 0.9981 | No clear effect |
| use weighted NCE instead of vanilla NCE (ablation) | HCIL · DFDC (partial forgery) | accuracy: 0.9395 -> 0.9511 | Improved |
| add snippet contrast Ll on RAI vs baseline without snippet contrast (ablation) | HCIL · Cross-dataset DFDC | AUC: 0.6323 -> 0.6898 | Improved |
| add all components (RAI, ISF, Ll, Lg) (ablation) | HCIL · Cross-dataset Celeb-DF | AUC: 0.6956 -> 0.7900 | Improved |
Closest papers by meaning
- Exploiting Style Latent Flows for Generalizing Deepfake Video Detection (CVPR 2024 · similarity 0.88)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.87)
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes (NeurIPS 2025 · similarity 0.87)
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · similarity 0.86)
- AVFF: Audio-Visual Feature Fusion for Video Deepfake Detection (CVPR 2024 · similarity 0.86)
- Contrastive Learning for DeepFake Classification and Localization via Multi-Label Ranking (CVPR 2024 · 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.
