Research map / Video / Video segmentation and tracking
Beyond Spatial Frequency: Pixel-wise Temporal Frequency-based Deepfake Video Detection
Taehoon Kim, Jongwook Choi, Yonghyun Jeong, Haeun Noh, Jaejun Yoo, Seungryul Baek, Jongwon Choi
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 part-based frequency features via APM to global features (ablation) | Ours · CDF | AUC: 88.2 -> 89.7 | Improved |
| add global frequency features to part-based features (ablation) | Ours · CDF | AUC: 85.6 -> 89.7 | Improved |
| use APM-proposed regions instead of landmark regions (ablation) | Ours · CDF | AUC: 81.6 -> 89.7 | Improved |
| add feature blender to MLP classifier (ablation) | Ours · CDF | AUC: 59.7 -> 66.6 | Improved |
| add TTE to STE-only joint transformer (ablation) | Ours · CDF | AUC: 70.7 -> 89.7 | Improved |
| add STE to TTE-only joint transformer (ablation) | Ours · CDF | AUC: 88.2 -> 89.7 | Improved |
| add auxiliary classifier loss (ablation) | Ours · CDF | AUC: 87.2 -> 89.7 | Improved |
| delay lambda=1 until after 4 epochs instead of from the start (ablation) | Ours · CDF | AUC: 86.9 -> 89.7 | Improved |
| add median filtering preprocessing (ablation) | ResNet-50 global temporal frequency · FF++ cross-synthesis Avg | AUC: 62.25 -> 92.34 | Improved |
| median filter instead of mean filter (ablation) | ResNet-50 global temporal frequency · FF++ cross-synthesis Avg | AUC: 89.81 -> 92.34 | Improved |
| use magnitude instead of phase of temporal spectrum (ablation) | ResNet-50 global temporal frequency · FF++ cross-synthesis Avg | AUC: 55.34 -> 92.34 | Improved |
| use magnitude only instead of magnitude and phase (ablation) | ResNet-50 global temporal frequency · FF++ cross-synthesis Avg | AUC: 89.36 -> 92.34 | Improved |
Closest papers by meaning
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.91)
- Frequency-Aware Deepfake Detection: Improving Generalizability through Frequency Space Domain Learning (AAAI 2024 · similarity 0.88)
- Exploiting Style Latent Flows for Generalizing Deepfake Video Detection (CVPR 2024 · similarity 0.88)
- Through the Lens: Benchmarking Deepfake Detectors Against Moiré-Induced Distortions (NeurIPS 2025 · similarity 0.86)
- Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model (CVPR 2025 · similarity 0.86)
- Exposing and Mitigating Temporal Attack in Deepfake Video Detection (NeurIPS 2026 · 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.
