Research map / Image generation / GANs
Fourier Spectrum Discrepancies in Deep Network Generated Images
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 |
|---|---|---|---|
| Use Fourier high-frequency decay parameters with a KNN classifier | Real versus generated image classification · 1024² images, 100% quality | classification accuracy: 99.2% | Helped |
| Compress images to 95% quality | Fourier-spectrum KNN · 1024² images | classification accuracy: 99.2% -> 94.4% | Hurt |
| Compress images to 85% quality | Fourier-spectrum KNN · 1024² images | classification accuracy: 99.2% -> 83.9% | Hurt |
| Crop images from 1024² to 768² | Fourier-spectrum KNN · Uncompressed images | classification accuracy: 99.2% -> 98.5% | Hurt |
| Reduce resolution from 1024² to 256² | Fourier-spectrum KNN · Uncompressed images | classification accuracy: 99.2% -> 88.8% | Hurt |
| Compress images to 85% quality | Fourier-spectrum KNN · 256² images | classification accuracy: 88.8% -> 87.4% | Hurt |
| Apply 95% compression | Fourier-spectrum KNN · VQ-VAE2 images, 1024² and 768² | classification accuracy: 99.8% -> 100 % | No clear effect |
| Apply 85% compression | Fourier-spectrum KNN · VQ-VAE2 images, 1024² and 768² | classification accuracy: 99.8% -> 99.6% | No clear effect |
| Train on 1024² images and test on cropped 768² images | Fourier-spectrum KNN · 1024² to 768² resolution transfer | classification accuracy: The classifier ... was able to maintain a similar classification accuracy | No clear effect |
| Modify generated-image high-frequency spectra to match a real-image target | Generated images and the Fourier-spectrum classifier · Spectrum synthesis demonstration | classification outcome | Helped |
| Continue ALAE training with versus without regularization (ablation) | ALAE · Generated-image high-frequency decay rates | decay-rate difference | No clear effect |
Closest papers by meaning
- A Closer Look at Fourier Spectrum Discrepancies for CNN-Generated Images Detection (CVPR 2021 · similarity 0.91)
- On the Frequency Bias of Generative Models (NeurIPS 2021 · similarity 0.89)
- Leveraging Frequency Analysis for Deep Fake Image Recognition (ICML 2020 · similarity 0.89)
- SSD-GAN: Measuring the Realness in the Spatial and Spectral Domains (AAAI 2021 · similarity 0.88)
- Frequency-Aware Deepfake Detection: Improving Generalizability through Frequency Space Domain Learning (AAAI 2024 · similarity 0.88)
- Spectrum Translation for Refinement of Image Generation (STIG) Based on Contrastive Learning and Spectral Filter Profile (AAAI 2024 · 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.