Research map / Video / Video segmentation and tracking
Exploring Temporal Coherence for More General Video Face Forgery Detection
Yinglin Zheng, Jianmin Bao, Dong Chen, Ming Zeng, Fang Wen
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 |
|---|---|---|---|
| set all spatial convolution kernels to 1 (FTCN) (ablation) | 3D R50 · FF++ HQ train on F2F, test avg | video-level AUC (%): 90.3 -> 97.5 | Improved |
| use spatial-only convolutions instead of spatio-temporal (ablation) | 3D R50 · FF++ HQ train on F2F, test avg | video-level AUC (%): 90.3 -> 77.7 | Hurt |
| add 3x3 spatial kernel in first FTCN conv layer (ablation) | 3D R50-FTCN · FF++ HQ train on F2F, test avg | video-level AUC (%): 97.5 -> 96.8 | Hurt |
| add 5x5 spatial kernel in first FTCN conv layer (ablation) | 3D R50-FTCN · FF++ HQ train on F2F, test avg | video-level AUC (%): 97.5 -> 95.1 | Hurt |
| spatially shuffle the input pixels of test/training clips (ablation) | 3D R50-FTCN · FF++ HQ train on F2F, test avg | video-level AUC (%): 97.5 -> 95.8 | Hurt |
| use FTCN design instead of parameter-matched channel-reduced 3D R50 (ablation) | 3D R50-SP · FF++ HQ train on F2F, test avg | video-level AUC (%): 89.6 -> 97.5 | Improved |
| use height-only (FHCN) convolutions instead of temporal-only (ablation) | 3D R50-FTCN · FF++ HQ train on F2F, test avg | video-level AUC (%): 97.5 -> 77.1 | Hurt |
| use width-only (FWCN) convolutions instead of temporal-only (ablation) | 3D R50-FTCN · FF++ HQ train on F2F, test avg | video-level AUC (%): 97.5 -> 83.4 | Hurt |
| increase clip size from 8 to 32 frames (ablation) | 3D R50-FTCN · FF++ HQ train on F2F, test avg | video-level AUC (%): 87.7 -> 97.5 | Improved |
| increase clip size from 32 to 64 frames (ablation) | 3D R50-FTCN · FF++ HQ train on F2F, test avg | video-level AUC (%): 97.5 -> 97.9 | No clear effect |
| add one-layer Temporal Transformer after FTCN (ablation) | 3D R50-FTCN · FF++ HQ train on F2F, test avg | video-level AUC (%): 97.5 -> 98.9 | Improved |
| increase Temporal Transformer layers from 1 to 2 (ablation) | FTCN+TT · FF++ HQ train on F2F, test avg | video-level AUC (%): 98.9 -> 98.5 | Hurt |
Closest papers by meaning
- Delving Deep Into Unraveling Spatio-Temporal Forgery Cues for Face Forgery Video Detection (TPAMI 2026 · similarity 0.90)
- Face Forgery Video Detection via Temporal Forgery Cue Unraveling (CVPR 2025 · similarity 0.90)
- AltFreezing for More General Video Face Forgery Detection (CVPR 2023 · similarity 0.89)
- Learning Natural Consistency Representation for Face Forgery Video Detection (ECCV 2024 · similarity 0.88)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.87)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · 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.
