Research map / Video / Video segmentation and tracking
FakeRadar: Probing Forgery Outliers to Detect Unknown Deepfake Videos
Zhaolun Li, Jichang Li, Yinqi Cai, Junye Chen, Xiaonan Luo, Guanbin Li, Rushi Lan
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 outlier-conditioned cross-entropy loss (ablation) | FakeRadar · Cross-dataset average (train FF++ HQ) | average AUC (%): 87.9 -> 90.1 | Improved |
| add outlier-driven contrastive loss (ablation) | FakeRadar · Cross-dataset average (train FF++ HQ) | average AUC (%): 88.0 -> 90.1 | Improved |
| add Forgery Outlier Probing on top of both losses (ablation) | FakeRadar · Cross-dataset average (train FF++ HQ) | average AUC (%): 86.4 -> 90.1 | Improved |
| add outlier-conditioned CE loss alone to baseline (ablation) | baseline w/o FOP/ODCL · Cross-dataset average (train FF++ HQ) | average AUC (%): 86.4 -> 86.7 | Improved |
| add outlier-driven contrastive loss alone to baseline (ablation) | baseline w/o FOP/OCCE · Cross-dataset average (train FF++ HQ) | average AUC (%): 86.4 -> 87.6 | Improved |
| add dynamic subcluster modeling (ablation) | FakeRadar · CDFv2 (train FF++ HQ) | AUC (%): 90.2 -> 91.7 | Improved |
| add dynamic subcluster modeling (ablation) | FakeRadar · DFDC (train FF++ HQ) | AUC (%): 82.3 -> 84.1 | Improved |
| add cluster-conditional outlier generation (vs Gaussian noise outliers) (ablation) | FakeRadar · CDFv2 (train FF++ HQ) | AUC (%): +2.0% AUC | Improved |
| use dynamic number of subclusters instead of fixed K=5 (ablation) | FakeRadar · Cross-dataset average (train FF++ HQ) | average AUC (%): 87.4 -> 90.1 | Improved |
| add fine-grained forgery subtype labels as prior (ablation) | FakeRadar · Cross-dataset average (train FF++ HQ) | average AUC (%): 89.5 -> 90.1 | Improved |
| use three-class (real/fake/outlier) classifier instead of binary (ablation) | FakeRadar · Cross-dataset prediction probabilities | forgery probability on fake samples | Improved |
| fine-tune with ST-Adapter and binary classification vs frozen CLIP (ablation) | FakeRadar (Frozen) · CDF | AUC (%): 60.0 -> 88.2 | Improved |
Closest papers by meaning
- Exposing the Deception: Uncovering More Forgery Clues for Deepfake Detection (AAAI 2024 · similarity 0.91)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.89)
- Exploring Unbiased Deepfake Detection via Token-Level Shuffling and Mixing (AAAI 2025 · similarity 0.89)
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · similarity 0.89)
- Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning (ICLR 2026 · similarity 0.88)
- FreqBlender: Enhancing DeepFake Detection by Blending Frequency Knowledge (NeurIPS 2024 · similarity 0.88)
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.
