Research map / Image generation / Autoregressive and latent generation
Self-Supervised AI-Generated Image Detection: A Camera Metadata Perspective
Nan Zhong, Mian Zou, Yiran Xu, Zhenxing Qian, Xinpeng Zhang, Baoyuan Wu, Kede Ma
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 |
|---|---|---|---|
| use EXIF pretext features instead of CLIP features (ablation) | one-class detector · 17 generators average | accuracy: 53.28 -> 81.02 | Improved |
| use EXIF classification and ranking instead of EAL image-EXIF text pairs (ablation) | one-class detector · 17 generators average | accuracy: 55.83 -> 81.02 | Improved |
| pairwise ranking instead of regression for continuous tags (ablation) | SDAIE · 17 generators average | accuracy: 79.55 -> 81.02 | Improved |
| pairwise ranking instead of regression for continuous tags (ablation) | SDAIE-dagger · 17 generators average | accuracy: 91.38 -> 94.30 | Improved |
| add high-pass filtering (ablation) | SDAIE · 17 generators average | accuracy: 66.84 -> 81.02 | Improved |
| add patch scrambling (ablation) | SDAIE · 17 generators average | accuracy: 80.56 -> 81.02 | Improved |
| add patch scrambling (ablation) | SDAIE-dagger · 17 generators average | accuracy: 80.02 -> 94.30 | Improved |
| add covariance pooling (ablation) | SDAIE-dagger · 17 generators average | accuracy: 89.14 -> 94.30 | Improved |
| patch size 64 instead of 16 (ablation) | SDAIE-dagger · 17 generators average | accuracy: 84.09 -> 94.30 | Improved |
| full 30-filter bank instead of Gabor filters (ablation) | SDAIE-dagger · 17 generators average | accuracy: 88.34 -> 94.30 | Improved |
| EXIF regularizer instead of fine-tuning extractor with head (ablation) | SDAIE-dagger · 17 generators average | accuracy: 91.78 -> 94.30 | Improved |
| data augmentation (ablation) | SDAIE-dagger · clean images | accuracy: 76.48 -> 94.30 | Improved |
Closest papers by meaning
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.88)
- A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 2025 · similarity 0.87)
- WildFake: A Large-Scale and Hierarchical Dataset for AI-Generated Images Detection (AAAI 2025 · similarity 0.87)
- Forensic Self-Descriptions Are All You Need for Zero-Shot Detection, Open-Set Source Attribution, and Clustering of AI-generated Images (CVPR 2025 · similarity 0.87)
- A Sanity Check for AI-generated Image Detection (ICLR 2025 · similarity 0.86)
- CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI (CVPR 2025 · 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.
