Research map / Image generation / Autoregressive and latent generation
Enabling Supervised Learning of Generative Signatures for Generalized AI-Generated Images Detection
Jianwei Fei, Yunshu Dai, Xiaoyu Zhou, Zhihua Xia, Alessandro Piva
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 self-augmentation with reconstructed images (p_aug 20%) (ablation) | GenSign detector · mean mAP across 3 benchmarks | mAP: 88.80 -> 95.79 | Improved |
| increase reconstructor downsampling levels from 1 to 3 (ablation) | DAIR · mean AP | AP: 76.6 -> 95.8 | Improved |
| feed GenSign instead of RGB to CNN from scratch (ablation) | ResNet50 · mean mAP | mAP: 68.93 -> 78.86 | Improved |
| feed GenSign instead of RGB to CNN from scratch (ablation) | EfficientNet-B0 · mean mAP | mAP: 80.88 -> 83.59 | Improved |
| replace RGB with GenSign input (ablation) | CLIP ViT-L/14 · mean mAP | mAP: 88.39 -> 87.93 | Hurt |
| add GenSign stream to RGB stream (dual-stream) (ablation) | CLIP detector · mean mAP | mAP: 88.39 -> 95.79 | Improved |
| use GenSign extractor instead of SRM Conv (ablation) | dual-stream detector · mean mAP | mAP: 95.27 -> 95.79 | Improved |
| use GenSign extractor instead of Noiseprint++ (ablation) | dual-stream detector · mean mAP | mAP: 86.18 -> 95.79 | Improved |
| use GenSign detector instead of FatFormer | FatFormer · UniversalFakeDetect | mean AP: 98.16 -> 99.37 | Improved |
| use GenSign detector instead of AIDE | AIDE · AIGIBenchmark | mean AP: 75.36 -> 89.96 | Improved |
| use GenSign detector instead of Effort | Effort · GenImage | mean accuracy: 91.1 -> 96.6 | Improved |
| use GenSign detector instead of UniFD | UniFD · AIGCDetectBenchmark: Midjourney | AP: 74.00 -> 81.17 | Improved |
Closest papers by meaning
- A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 2025 · similarity 0.89)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.89)
- Seeing What Matters: Generalizable AI-generated Video Detection with Forensic-Oriented Augmentation (NeurIPS 2025 · similarity 0.88)
- CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI (CVPR 2025 · similarity 0.87)
- Stay-Positive: A Case for Ignoring Real Image Features in Fake Image Detection (ICML 2025 · similarity 0.87)
- Transfer Learning of Real Image Features with Soft Contrastive Loss for Fake Image Detection (AAAI 2025 · similarity 0.87)
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.
