Research map / Image generation / GANs
RoCGAN: Robust Conditional GAN
Grigorios G. Chrysos, Jean Kossaifi, Stefanos P. Zafeiriou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 autoencoder pathway with shared decoder (RoCGAN) to cGAN | 4layer cGAN · Natural scenes denoising 25% | SSIM: 0.628 -> 0.668 | Improved |
| add autoencoder pathway with shared decoder (RoCGAN) to cGAN | 4layer cGAN · Faces denoising 25% | SSIM: 0.803 -> 0.834 | Improved |
| add autoencoder pathway under extra test noise | 4layer cGAN · Faces denoising 35% (extra noise) | SSIM: 0.765 -> 0.821 | Improved |
| add autoencoder pathway on sparse inpainting | 4layer cGAN · Faces sparse inpainting 50% | SSIM: 0.801 -> 0.804 | No clear effect |
| add autoencoder pathway under extra inpainting corruption | 4layer cGAN · Natural scenes sparse inpainting 75% | SSIM: 0.542 -> 0.548 | No clear effect |
| add autoencoder pathway on ImageNet | 4layer cGAN · ImageNet denoising 25% | SSIM: 0.657 -> 0.679 | Improved |
| add autoencoder pathway to 5layer network | 5layer cGAN · Faces sparse inpainting 75% | SSIM: 0.707 -> 0.818 | Improved |
| add autoencoder pathway to 4layer-skip network | 4layer-skip cGAN · Faces sparse inpainting 50% | SSIM: 0.855 -> 0.859 | No clear effect |
| use 6layer instead of 5layer RoCGAN | RoCGAN · Faces denoising 25% | SSIM: 0.890 -> 0.881 | No clear effect |
| train with additional unlabelled data for AE pathway | RoCGAN · Faces sparse inpainting 75% (50k labelled) | SSIM: 0.617 -> 0.681 | Improved |
| use OneNet instead of the cGAN baseline | cGAN · Faces denoising 25% | SSIM: 0.803 -> 0.758 | Hurt |
Later papers that compared with this method
1 later accepted paper reports RoCGAN as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Reconstructing the Noise Variance Manifold for Image Denoising (ECCV 2020 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Robust Conditional Generative Adversarial Networks (ICLR 2019 · similarity 0.99)
- Rethinking conditional GAN training: An approach using geometrically structured latent manifolds (NeurIPS 2021 · similarity 0.93)
- Diversity-Sensitive Conditional Generative Adversarial Networks (ICLR 2019 · similarity 0.92)
- Robustness of conditional GANs to noisy labels (NeurIPS 2018 · similarity 0.92)
- CcGAN: Continuous Conditional Generative Adversarial Networks for Image Generation (ICLR 2021 · similarity 0.92)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.89)
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.
