Research map / Video / Video segmentation and tracking
Vulnerability-Aware Spatio-Temporal Learning for Generalizable Deepfake Video Detection
Dat Nguyen, Marcella Astrid, Anis Kacem, Enjie Ghorbel, Djamila Aouada
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 SBV pseudo-fake training data (ablation) | TimeSformer baseline · Cross-dataset avg (FF++ c23) | AUC (%): 63.2 -> 85.1 | Improved |
| add vulnerability-driven Cutout (ablation) | TimeSformer + SBV · Cross-dataset avg (FF++ c23) | AUC (%): 85.1 -> 85.2 | No clear effect |
| add spatial branch g (ablation) | TimeSformer + SBV + V-CutOut · Cross-dataset avg (FF++ c23) | AUC (%): 85.2 -> 85.7 | Improved |
| add temporal branch h (ablation) | TimeSformer + SBV · Cross-dataset avg (FF++ c23) | AUC (%): 85.1 -> 86.7 | Improved |
| add V-CutOut and g to SBV + h (ablation) | TimeSformer + SBV + h · Cross-dataset avg (FF++ c23) | AUC (%): 86.7 -> 87.8 | Improved |
| add Consistent Synthesized Parameters (CSP) (ablation) | Stacked SBIs · CDF | AUC (%): 48.4 -> 84.1 | Improved |
| add Landmark Interpolation (LI) (ablation) | SBV with CSP · CDF | AUC (%): 84.1 -> 90.7 | Improved |
| use raw c0 data instead of c23 for SBV training | SBV · CDF | AUC (%): 90.7 -> 94.9 | Improved |
| increase number of frames T from 4 to 16 | FakeSTormer · CDF | AUC (%): 92.4 -> 92.8 | Improved |
| loss weights 0.8/100/0.2 instead of 0.9/1/0.1 (ablation) | FakeSTormer · Cross-dataset avg (FF++ c23) | AUC (%): 85.1 -> 87.8 | Improved |
| MeanStd instead of MinMax normalization of temporal GT (ablation) | FakeSTormer · Cross-dataset avg (FF++ c23) | AUC (%): 84.8 -> 87.8 | Improved |
| MeanStd instead of unnormalized 3D Gaussian (ablation) | FakeSTormer · Cross-dataset avg (FF++ c23) | AUC (%): 85.2 -> 87.8 | Improved |
Closest papers by meaning
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.89)
- D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · similarity 0.86)
- Exploiting Style Latent Flows for Generalizing Deepfake Video Detection (CVPR 2024 · similarity 0.86)
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes (NeurIPS 2025 · similarity 0.86)
- Exposing the Deception: Uncovering More Forgery Clues for Deepfake Detection (AAAI 2024 · similarity 0.85)
- Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model (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.
