Research map / Image generation / GANs
Think Twice Before Detecting GAN-Generated Fake Images From Their Spectral Domain Imprints
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 |
|---|---|---|---|
| Apply Method 1 with power distribution correction to fake images | GAN-generated images evaluated by the shallow CNN spectrum detector · BigGAN | accuracy (%): 94.59 -> 68.44 | Helped |
| Apply Method 1 with power distribution correction to fake images | GAN-generated images evaluated by the shallow CNN spectrum detector · CycleGAN, StarGAN, StyleGAN, and Fingerprint | accuracy (%): CycleGAN 99.48 -> 53.30; StarGAN 98.08 -> 54.73; StyleGAN 85.48 -> 64.05; Fingerprint 100.0 -> 51.09 | Helped |
| Apply Method 2 with power distribution correction to fake images | GAN-generated images evaluated by the shallow CNN spectrum detector · CRN, IMLE, ProGAN, and StyleGAN2 | accuracy (%): CRN 96.32 -> 72.67; IMLE 95.11 -> 57.79; ProGAN 97.62 -> 61.32; StyleGAN2 92.55 -> 62.63 | Helped |
| Use power distribution correction alone (ablation) | GAN-generated images evaluated by the shallow CNN spectrum detector · BigGAN, CRN, CycleGAN, IMLE, ProGAN, StarGAN, StyleGAN, StyleGAN2 | accuracy (%): 94.59 -> 93.28; 96.32 -> 91.50; 99.48 -> 94.33; 95.11 -> 92.68; 97.62 -> 91.49; 98.08 -> 93.85; 85.48 -> 80.13; 92.55 -> | No clear effect |
| Add Method 1 artifact suppression to power distribution correction (ablation) | GAN-generated images evaluated by the shallow CNN spectrum detector · CycleGAN, StarGAN, StyleGAN, Fingerprint versus other classes | accuracy (%): CycleGAN: 59.18 -> 53.30; StarGAN: 60.28 -> 54.73; StyleGAN: 67.19 -> 64.05; Fingerprint: 54.55 -> 51.09 | Mixed |
| Add Method 2 spectrum difference normalization to power distribution correction (ablation) | GAN-generated images evaluated by the shallow CNN spectrum detector · BigGAN, CRN, IMLE, ProGAN, StyleGAN2 | accuracy (%): BigGAN: 68.44 -> 60.34; CRN: 81.09 -> 72.67; IMLE: 61.79 -> 57.79; ProGAN: 64.58 -> 61.32; StyleGAN2: 69.28 -> 62.63 | Helped |
Closest papers by meaning
- Leveraging Frequency Analysis for Deep Fake Image Recognition (ICML 2020 · similarity 0.92)
- Secret Lies in Color: Enhancing AI-Generated Images Detection with Color Distribution Analysis (CVPR 2025 · similarity 0.86)
- Transfer Learning of Real Image Features with Soft Contrastive Loss for Fake Image Detection (AAAI 2025 · similarity 0.85)
- Spectrum Translation for Refinement of Image Generation (STIG) Based on Contrastive Learning and Spectral Filter Profile (AAAI 2024 · similarity 0.84)
- Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection (CVPR 2024 · similarity 0.84)
- Frequency-Aware Deepfake Detection: Improving Generalizability through Frequency Space Domain Learning (AAAI 2024 · similarity 0.83)
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.