Research map / Image generation / GANs
CLR-GAN: Improving GANs Stability and Quality via Consistent Latent Representation and Reconstruction
Shengke Sun, Ziqian Luan, Zhanshan Zhao, Shijie Luo, Shuzhen Han
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 latent consistency and reconstruction losses (lambda1=5, lambda2=0.5) | DCGAN · CIFAR-10 | FID: 38.56 -> 23.3 | Improved |
| Add latent consistency and reconstruction losses | StyleGAN-V2 · FFHQ 256 | FID: 3.86 -> 3.44 | Improved |
| Add both losses on top of StyleGAN-V2-ADA | StyleGAN-V2-ADA · AFHQ-Cat | FID: 6.05 -> 4.45 | Improved |
| Use only the latent consistency loss (lambda2=0) instead of both losses (ablation) | CLR-GAN · CIFAR-10 | FID: 23.3 -> 25.15 | Hurt |
| Use only the reconstruction loss (lambda1=0) instead of both losses (ablation) | CLR-GAN · CIFAR-10 | FID: 23.3 -> 26.09 | Hurt |
| Add only one of the two losses on high-resolution data (ablation) | StyleGAN-V2 · FFHQ 256 | FID: 3.86 -> 3.75 and 3.77 | No clear effect |
| Raise latent loss weight lambda1 from 5 to 15 (ablation) | CLR-GAN · CIFAR-10 | FID: 25.20 -> 26.30 | Hurt |
| Raise reconstruction weight lambda2 from 0.5 to 5 at lambda1=5 (ablation) | CLR-GAN · CIFAR-10 | FID: 23.3 -> 25.55 | Hurt |
| Use CLR-GAN instead of AdaptiveMix on CelebA | DCGAN · CelebA 64x64 | FID: 12.43 -> 13.63 | Hurt |
| Use the CLR-GAN trained discriminator as a feature extractor with a linear head | Discriminator · CIFAR-10 classification | accuracy: 97.7% (HCGNet) vs 98.8% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- AdaptiveMix — AdaptiveMix: Improving GAN Training via Feature Space Shrinkage (CVPR 2023) · better on 1 of 2 shared results
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
- Uncertainty Principles of Encoding GANs (ICML 2021 · similarity 0.92)
- Are GANs Created Equal? A Large-Scale Study (NeurIPS 2018 · similarity 0.91)
- Generalized Adversarially Learned Inference (AAAI 2021 · similarity 0.91)
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.91)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.90)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · similarity 0.90)
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.
