Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Leveraging Failed Samples: A Few-Shot and Training-Free Framework for Generalized Deepfake Detection
Shibo Yao, Renshuai Tao, Xiaolong Zheng, Chao Liang, Chunjie Zhang
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 |
|---|---|---|---|
| fine-tune the cache keys (FTNet-T) instead of training-free FTNet (ablation) | FTNet · UniversalFakeDetect | mAcc: 94.03 -> 97.14 | Improved |
| fine-tune the cache keys (FTNet-T) instead of training-free FTNet (ablation) | FTNet · GenImage cross-domain | mAcc: 90.7 -> 94.2 | Improved |
| fine-tune the cache keys (FTNet-T) instead of training-free FTNet (ablation) | FTNet · OpenSDI | mAcc: 79.94 -> 83.16 | Improved |
| use 8-shot fine-tuned baselines UniFD/NPR vs FTNet-T with only 4 shots | FTNet-T · GenImage, UniFD, OpenSDI average | mAcc: 69.3 -> 90.3 | Improved |
| FTNet-T trained on 500 images vs baselines trained on 320k SDv1.4 images | FTNet-T · GenImage zero-shot generalization | mAcc: 84.6 -> 86.0 | Improved |
| training-free FTNet with 500 images vs UniFD trained on 320k images | FTNet · GenImage zero-shot generalization | mAcc: 84.6 -> 80.8 | Hurt |
| use more shots in the cache (ablation) | FTNet · GenImage, UniFD, OpenSDI | accuracy | Mixed |
| use CLIP middle layer L12 instead of L6, L18 or L24 features (ablation) | FTNet · GenImage 4-shot | average accuracy | Improved |
| use ViT-L/14 backbone instead of ViT-B/16 (ablation) | FTNet · GenImage | accuracy: 60.4 -> 90.7 | Improved |
| use ViT-L/14 backbone instead of ViT-B/32 (ablation) | FTNet · GenImage | accuracy: 61.2 -> 90.7 | Improved |
| use ViT-L/14 backbone instead of ViT-B/16 (ablation) | FTNet · Chameleon | accuracy: 77.4 -> 80.7 | Improved |
| apply JPEG compression and Gaussian blur to test images | FTNet · Chameleon | accuracy | Hurt |
Closest papers by meaning
- D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · similarity 0.91)
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.90)
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · similarity 0.89)
- Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model (CVPR 2025 · similarity 0.89)
- Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning (ICLR 2026 · similarity 0.88)
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes (NeurIPS 2025 · 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.
