Research map / Image generation / GANs
Blur, Noise, and Compression Robust Generative Adversarial Networks
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 |
|---|---|---|---|
| Use BR-GAN rather than standard GAN for blur-robust generation | CIFAR-10 · Blur (D), focus or motion | FID: 43.1 -> 23.4 | Helped |
| Use CR-GAN rather than standard GAN for compression-robust generation | CIFAR-10 · Compression (J), quality factor 60--100 | FID: 33.3 -> 26.3 | Helped |
| Use BNCR-GAN rather than single-degradation models | CIFAR-10 generation with blur, noise, and compression · All images degraded (N) | FID: 51.1 -> 34.1 | Helped |
| Add the masking architecture (ablation) | BR-GAN · CIFAR-10 blur (D) | FID: 26.6 -> 23.4 | Helped |
| Add the masking architecture (ablation) | CR-GAN · CIFAR-10 compression (J) | FID: 38.6 -> 26.3 | Helped |
| Add adaptive consistency losses (ablation) | BNCR-GAN · CIFAR-10 blur, noise, and compression (N) | FID: 41.0 -> 34.1 | Helped |
| Add adaptive consistency losses (ablation) | BNCR-GAN · CIFAR-10 compression (J) | FID: 33.4 -> 26.6 | Helped |
| Add the blur adaptive consistency loss (ablation) | BNCR-GAN · CIFAR-10 blur (D) | FID: 25.7 -> 26.8 | Hurt |
| Use adaptive rather than non-adaptive consistency losses (ablation) | BNCR-GAN · CIFAR-10, across six clean and degraded settings | FID: BNCR-GAN outperformed the non-adaptive variant in 5/6 settings | Helped |
| Use adaptive rather than non-adaptive consistency losses (ablation) | BNCR-GAN · CIFAR-10 noise (R) | FID: 20.5 | No clear effect |
| Use BNCR-GAN+UNIR rather than CycleGAN | CIFAR-10 image restoration · Three-quarter degraded images (Q), FID | FID: 33.7 -> 20.4 | Helped |
| Use BNCR-GAN+UNIR rather than UNIR | CIFAR-10 image restoration · All images degraded (N), LPIPS | LPIPS: 0.0146 -> 0.0206 | Hurt |
Closest papers by meaning
- Noise Robust Generative Adversarial Networks (CVPR 2020 · similarity 0.93)
- GAN2GAN: Generative Noise Learning for Blind Denoising with Single Noisy Images (ICLR 2021 · similarity 0.89)
- Image Blind Denoising With Generative Adversarial Network Based Noise Modeling (CVPR 2018 · similarity 0.89)
- Label-Noise Robust Generative Adversarial Networks (CVPR 2019 · similarity 0.89)
- sRGB Real Noise Modeling via Noise-Aware Sampling with Normalizing Flows (ICLR 2024 · similarity 0.86)
- Uncertainty-Aware GAN for Single Image Super Resolution (AAAI 2024 · similarity 0.86)
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.