Research map / Detection and segmentation / CNN features
Thinking in Frequency: Face Forgery Detection by Mining Frequency-aware Clues
Yuyang Qian, Guojun Yin, Lu Sheng, Zixuan Chen, Jing Shao
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 FAD (frequency-aware decomposition) to Xception (ablation) | Xception · FF++ LQ | accuracy: 86.86% -> 87.95% | Improved |
| add learnable filters to FAD (ablation) | Xception+FAD · FF++ LQ | accuracy: 87.12% -> 87.95% | Improved |
| replace fixed PAFilters with FAD (ablation) | Xception · FF++ LQ | accuracy: 87.16% -> 87.95% | Improved |
| use all frequency bands in FAD instead of high band only (ablation) | FAD · FF++ LQ | accuracy: 87.77% -> 87.95% | Improved |
| use sliding window DCT instead of full-image DCT in LFS (ablation) | LFS branch · FF++ LQ | accuracy: 76.16% -> 82.47% | Improved |
| add frequency statistics to LFS with SWDCT (ablation) | LFS branch · FF++ LQ | accuracy: 82.47% -> 84.89% | Improved |
| add adaptive frequency statistics (D-Stat) to LFS (ablation) | LFS branch · FF++ LQ | accuracy: 84.89% -> 86.16% | Improved |
| add FAD and LFS two-stream frequency branches to Xception | Xception · FF++ LQ | accuracy: 86.86% -> 90.43% | Improved |
| add FAD and LFS to Slowfast video model | Slowfast-R101 · FF++ LQ | accuracy: 90.53% -> 93.02% | Improved |
| add FAD and LFS to Xception on NeuralTextures | Xception · FF++ LQ, NeuralTextures only | accuracy: 79.14% -> 83.32% | Improved |
| enlarge SWDCT window from 10 to 30 (ablation) | LFS branch · FF++ LQ | AUC: 0.889 -> 0.853 | Hurt |
| enlarge SWDCT window from 2 to 10 (ablation) | LFS branch · FF++ LQ | AUC: 0.836 -> 0.889 | Improved |
Closest papers by meaning
- Frequency-Aware Discriminative Feature Learning Supervised by Single-Center Loss for Face Forgery Detection (CVPR 2021 · similarity 0.91)
- Exploring Frequency Adversarial Attacks for Face Forgery Detection (CVPR 2022 · similarity 0.91)
- Local Relation Learning for Face Forgery Detection (AAAI 2021 · similarity 0.90)
- FreqBlender: Enhancing DeepFake Detection by Blending Frequency Knowledge (NeurIPS 2024 · similarity 0.89)
- DevFD : Developmental Face Forgery Detection by Learning Shared and Orthogonal LoRA Subspaces (NeurIPS 2025 · similarity 0.88)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · 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.
