Research map / Image generation / Autoregressive and latent generation
CNN-Generated Images Are Surprisingly Easy to Spot... for Now
Check 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 |
|---|---|---|---|
| Add blur and JPEG augmentation during training (ablation) | ProGAN-trained detector · BigGAN | AP: 72.2→88.2 | Helped |
| Add blur and JPEG augmentation during training (ablation) | ProGAN-trained detector · CycleGAN | AP: 84.0→96.8 | Helped |
| Add blur and JPEG augmentation during training (ablation) | ProGAN-trained detector · GauGAN | AP: 67.0→98.1 | Helped |
| Add blur and JPEG augmentation during training (ablation) | ProGAN-trained detector · SAN | AP: 93.6→63.9 | Hurt |
| Add blur and JPEG augmentation during training (ablation) | ProGAN-trained detector · DeepFake | AP: 98.2→66.3 | Hurt |
| Use JPEG augmentation only rather than no augmentation (ablation) | ProGAN-trained detector · ForenSynths, 11 generators | mAP: 90.1→93.0 | Helped |
| Increase training diversity from 2 to 16 ProGAN classes (ablation) | ProGAN-trained detector · ForenSynths, 11 generators | mAP: 81.3→91.4 | Helped |
| Increase training diversity from 16 to 20 ProGAN classes (ablation) | ProGAN-trained detector · ForenSynths, 11 generators | mAP: 91.4→90.8 | No clear effect |
| Reduce the training dataset to 10% (ablation) | ProGAN-trained detector · ForenSynths, 11 generators | mAP: 90.8→87.8 | Hurt |
| Train on BigGAN instead of ProGAN (ablation) | CNN-image detector · ForenSynths, 11 generators | mAP: 90.8→83.9 | Hurt |
| Train on deep image prior reconstructions instead of ProGAN samples (ablation) | CNN-image detector · ForenSynths, 11 generators | mAP: 90.8→70.3 | Hurt |
Closest papers by meaning
- Towards Universal Fake Image Detectors That Generalize Across Generative Models (CVPR 2023 · similarity 0.90)
- GenImage: A Million-Scale Benchmark for Detecting AI-Generated Image (NeurIPS 2023 · similarity 0.87)
- RealNet: Efficient and Unsupervised Detection of AI-Generated Images via Real-Only Representation Learning (AAAI 2026 · similarity 0.85)
- CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI (CVPR 2025 · similarity 0.85)
- Few-Shot Learner Generalizes Across AI-Generated Image Detection (ICML 2025 · similarity 0.85)
- Is Artificial Intelligence Generated Image Detection a Solved Problem? (NeurIPS 2025 · similarity 0.85)
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.