Research map / Video / Video segmentation and tracking
TALL: Thumbnail Layout for Deepfake Video Detection
Yuting Xu, Jian Liang, Gengyun Jia, Ziming Yang, Yanhao Zhang, Ran He
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 TALL thumbnail layout to Swin-B | Swin-B · FF++(HQ) -> Celeb-DF | AUC: 83.13 -> 90.79 | Improved |
| add TALL to ViT-B, ResNet50, EfficientNet-B4 | image-level backbones · FF++(HQ) -> Celeb-DF | AUC: 82.33 -> 86.58 (ViT-B); 76.38 -> 80.90 (ResNet50); 78.19 -> 83.37 (EffNetB4) | Improved |
| use TALL with 2x2 resize layout rather than 4x4 layout (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 80.18 -> 87.60 | Improved |
| resize frames instead of random crop for sub-images (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 78.55 -> 87.60 | Improved |
| keep forward temporal order of frames instead of random (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 85.85 -> 87.60 | Improved |
| use all four sub-images rather than dropping the last two (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 84.22 -> 87.60 | Improved |
| use TALL fixed-position mask instead of Cutout (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 89.33 -> 90.79 | Improved |
| add TALL mask over Mixup+CutMix (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 89.75 -> 90.79 | Improved |
| enlarge Swin window to (14,14,14,7) from (7,7,7,7) (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 85.60 -> 87.60 | Improved |
| enlarge windows further to (28,28,28,7) (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 87.60 -> 86.65 | Hurt |
| use full-resolution sub-images (224x224, 2x2) (ablation) | TALL-Swin · FF++(HQ) -> Celeb-DF | AUC: 87.60 -> 88.15 | No clear effect |
| apply TALL on low-quality compressed FF++ (LQ) | TALL-Swin · FF++ (LQ) | Acc.: 92.82 vs ISTVT 96.15 | Hurt |
Closest papers by meaning
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.84)
- Towards a Universal Synthetic Video Detector: From Face or Background Manipulations to Fully AI-Generated Content (CVPR 2025 · similarity 0.83)
- Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model (CVPR 2025 · similarity 0.83)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.83)
- AVFF: Audio-Visual Feature Fusion for Video Deepfake Detection (CVPR 2024 · similarity 0.82)
- Exploring Unbiased Deepfake Detection via Token-Level Shuffling and Mixing (AAAI 2025 · similarity 0.82)
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.
