Research map / Image generation / Autoregressive and latent generation
Few-Shot Synthetic Image Attribution: Identifying Unseen Generators with Limited Samples
Shiyu Wu, Shuyan Li, Jing Li, Jing Liu, Yequan Wang
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 the OmniDFA metric-learning model instead of prototypical classification on backbones of existing detectors/attributors | few-shot attribution · OmniFake 5-way 10-shot | accuracy: 73.31 -> 75.34 (vs FSD) | Improved |
| use OmniDFA instead of FSD on 15-way attribution | few-shot attribution · OmniFake 15-way 10-shot | accuracy: 52.28 -> 53.54 average; Part III 60.81 -> 60.26 | Mixed |
| train detectors on diverse multi-generator data rather than single-generator data | AIGI detection · OmniFake detection | accuracy: 57.24 (UnivFD) vs 95.58 (OmniDFA) | Improved |
| add sphere center loss on real class plus Tukey boundary threshold for detection | OmniDFA · OmniFake detection | accuracy: avg Acc 95.58 vs ComFor 89.75; real acc 95.32 vs ComFor 98.96 | Mixed |
| OmniDFA vs prior detectors under zero-shot Chameleon | OmniDFA · Chameleon | accuracy: 65.77 -> 83.48 | Improved |
| OmniDFA on GenImage zero-shot | OmniDFA · GenImage | accuracy: 90.53 -> 95.86 (average vs AIDE) | Improved |
| increase support shots from 1 to 10 per class (ablation) | OmniDFA · OmniFake attribution | accuracy | Improved |
| increase support shots beyond 10 up to 200 (ablation) | OmniDFA · OmniFake attribution | accuracy | No clear effect |
| use dual-branch (local and global) instead of single-branch (ablation) | OmniDFA · OmniFake detection and attribution | accuracy | Improved |
| use supervised contrastive learning instead of plain binary classifier (ablation) | OmniDFA · OmniFake detection | accuracy | Improved |
| train with JPEG and blur augmentation (ablation) | OmniDFA · OmniFake Part I robustness | accuracy | Mixed |
Closest papers by meaning
- Few-Shot Learner Generalizes Across AI-Generated Image Detection (ICML 2025 · similarity 0.91)
- Is Artificial Intelligence Generated Image Detection a Solved Problem? (NeurIPS 2025 · similarity 0.88)
- Provably Improving Generalization of Few-shot models with Synthetic Data (ICML 2025 · similarity 0.87)
- CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI (CVPR 2025 · similarity 0.87)
- Are High-Quality AI-Generated Images More Difficult for Models to Detect? (ICML 2025 · similarity 0.86)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (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.
