Research map / Image generation / Autoregressive and latent generation
GenImage: A Million-Scale Benchmark for Detecting AI-Generated Image
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| evaluate on a different generator than the one trained on (SD V1.4 to Midjourney) | ResNet-50 · GenImage | accuracy (%): 99.9 -> 54.9 | Hurt |
| use Swin-T instead of ResNet-50 | detector backbone · GenImage cross-generator | avg accuracy (%): 72.1 -> 74.8 | Helped |
| use DeiT-S instead of ResNet-50 | detector backbone · GenImage cross-generator | avg accuracy (%): 72.1 -> 71.6 | No clear effect |
| use GAN-oriented CNNSpot instead of plain ResNet-50 | detector · GenImage cross-generator | avg accuracy (%): 72.1 -> 64.2 | Hurt |
| use Spec frequency input instead of ResNet-50 | detector · GenImage cross-generator | avg accuracy (%): 72.1 -> 68.8 | Hurt |
| add blur and JPEG augmentation (CNNSpot) | ResNet-50 · GenImage JPEG q=30 | accuracy (%): 51.2 -> 97.3 | Helped |
| add blur and JPEG augmentation (CNNSpot) | ResNet-50 · GenImage low resolution 112 | accuracy (%): 96.2 -> 50.0 | Hurt |
| JPEG compression of test images | ResNet-50 · GenImage | accuracy (%): 99.9 -> 51.2 | Hurt |
| increase classes from 10 to 100 at 160 images per class (ablation) | ResNet-50 · GenImage same-generator | accuracy (%): 73.6 -> 97.7 | Helped |
| increase images per class and classes (1600 to 162000 total) (ablation) | ResNet-50 · GenImage cross-generator | accuracy (%): 60.9 -> 72.1 | Helped |
| increase images per class from 128 to 1300 (100 classes) (ablation) | ResNet-50 · GenImage all generators | accuracy (%): 81.2 -> 98.4 | Helped |
| increase classes from 10 to 100 at fixed training size (ablation) | ResNet-50 · GenImage all generators | accuracy (%): 78.8 -> 81.2 | Helped |
Closest papers by meaning
- Is Artificial Intelligence Generated Image Detection a Solved Problem? (NeurIPS 2025 · similarity 0.91)
- Beyond Generation: A Diffusion-based Low-level Feature Extractor for Detecting AI-generated Images (CVPR 2025 · similarity 0.91)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.90)
- WildFake: A Large-Scale and Hierarchical Dataset for AI-Generated Images Detection (AAAI 2025 · similarity 0.89)
- A Sanity Check for AI-generated Image Detection (ICLR 2025 · similarity 0.89)
- Few-Shot Learner Generalizes Across AI-Generated Image Detection (ICML 2025 · similarity 0.88)
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.