Research map / Image generation / Autoregressive and latent generation
Scaling Up AI-Generated Image Detection with Generator-Aware Prototypes
Ziheng Qin, Yuheng Ji, Renshuai Tao, Yuxuan Tian, Yuyang Liu, Yipu Wang, Xiaolong Zheng
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 prototype mapping (cross-attention to prototypes) (ablation) | GAPL w/o PCA, LoRA · 4-benchmark mean | mAcc: 60.05 -> 68.59 | Improved |
| add PCA-derived generator-aware prototypes to prototype mapping (ablation) | GAPL w/o LoRA · 4-benchmark mean | mAcc: 68.59 -> 71.88 | Improved |
| add LoRA finetuning to prototype mapping (ablation) | GAPL · 4-benchmark mean | mAcc: 68.59 -> 90.35 | Improved |
| add PCA prototypes on top of PM + LoRA (ablation) | GAPL · 4-benchmark mean | mAcc: 90.35 -> 95.54 | Improved |
| add LoRA alone (no PM, no PCA) (ablation) | GAPL · 4-benchmark mean | mAcc: 60.05 -> 88.52 | Improved |
| use PM+LoRA without PCA, mAP (ablation) | GAPL · 4-benchmark mean | mAP: 97.91 -> 95.40 (LoRA only vs PM+LoRA); full 98.97 | Mixed |
| increase number of prototypes from 16 to 64 (ablation) | GAPL · 4-benchmark mean | mAcc: 95.28 -> 95.54 | No clear effect |
| build prototypes from real generator types instead of random vectors (ablation) | GAPL · 4-benchmark mean | mAcc: 90.35 -> 95.54 (3 types) | Improved |
| add a fourth generator type to prototype set (ablation) | GAPL · 4-benchmark mean | mAcc: 95.54 -> 95.29 | No clear effect |
| train all detectors on large multi-generator data (scaling up) | previous AIGI detectors (e.g. Co-SPY, NPR, UniFD) · 4-benchmark mean | accuracy: e.g. UniFD 59.8 vs Swin-T 89.7 | Hurt |
| use GAPL rather than CommForen with 4.7K generators | GAPL · 6-benchmark mean | accuracy: 86.9 -> 90.4 | Improved |
| GAPL on Chameleon vs B-Free | GAPL · Chameleon | accuracy: 78.2 -> 71.0 | Hurt |
Closest papers by meaning
- Where Detectors Fail: Probing Generative Space for Generalizable AI-Generated Image Detection (ICML 2026 · similarity 0.88)
- SAIDO: Generalizable Detection of AI-Generated Images via Scene-Aware and Importance-Guided Dynamic Optimization in Continual Learning (CVPR 2026 · similarity 0.88)
- GenImage: A Million-Scale Benchmark for Detecting AI-Generated Image (NeurIPS 2023 · similarity 0.88)
- RealNet: Efficient and Unsupervised Detection of AI-Generated Images via Real-Only Representation Learning (AAAI 2026 · similarity 0.88)
- Is Artificial Intelligence Generated Image Detection a Solved Problem? (NeurIPS 2025 · similarity 0.87)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 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.
