Research map / Label-efficient and robust learning / Recognition and long-tail
DeepFake-Adapter: Dual-Level Adapter for DeepFake Detection
Rui Shao, Tianxing Wu, Liqiang Nie, Ziwei Liu
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 GBA to ViT with LSA (ablation) | DeepFake-Adapter · FF++ c40 train DF, test FS | AUC: 73.16 -> 79.51 | Improved |
| add LSA to ViT with GBA (ablation) | DeepFake-Adapter · FF++ c40 train DF, test FS | AUC: 75.03 -> 79.51 | Improved |
| add LSA to ViT with GBA (ablation) | DeepFake-Adapter · FF++ c40 train FS, test DF | AUC: 84.90 -> 88.57 | Improved |
| add either adapter to linear-probed ViT (ablation) | DeepFake-Adapter · FF++ c40 train DF, test FS | AUC: 67.20 -> 73.16 | Improved |
| use parameter-efficient dual adapters instead of full-tuning | DeepFake-Adapter vs full-tuning · FF++ c40 train DF, test FS | AUC: 74.25 -> 76.85 | Improved |
| use dual-level adapters instead of ViT-Adapter | DeepFake-Adapter vs ViT-Adapter · FF++ c40 train FS, test DF | AUC: 87.65 -> 88.57 | Improved |
| supervised ImageNet-21K weights instead of MAE self-supervised weights (ablation) | DeepFake-Adapter · FF++ c40 train DF, test FS | AUC: 73.12 -> 79.51 | Improved |
| 3 stages with 4 blocks instead of 2 stages with 6 blocks (ablation) | DeepFake-Adapter · FF++ c40 train DF, test FS | AUC: 75.44 -> 79.51 | Improved |
| 3 stages with 4 blocks instead of 6 stages with 2 blocks (ablation) | DeepFake-Adapter · FF++ c40 train DF, test FS | AUC: 78.69 -> 79.51 | Improved |
| scale backbone from ViT-Base to ViT-Large (ablation) | DeepFake-Adapter · FF++ c40 train DF, test FS | AUC: 79.51 -> 78.60 | No clear effect |
| scale backbone from ViT-Base to ViT-Large (ablation) | DeepFake-Adapter · FF++ c40 train FS, test DF | AUC: 88.57 -> 85.08 | Hurt |
| use Ours* variant for cross-manipulation generalization trained with F2F | DeepFake-Adapter · FF++ c40 train F2F, average over DF/FS/NT | AUC: 70.29 -> 71.14 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- STIL — STiL: Semi-supervised Tabular-Image Learning for Comprehensive Task-Relevant Information Exploration in Multimodal Classification (CVPR 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Exposing the Deception: Uncovering More Forgery Clues for Deepfake Detection (AAAI 2024 · similarity 0.91)
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · similarity 0.91)
- Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning (ICLR 2026 · similarity 0.91)
- Transcending Forgery Specificity with Latent Space Augmentation for Generalizable Deepfake Detection (CVPR 2024 · similarity 0.91)
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.91)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.91)
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.
