Research map / Image generation / GANs
CoopInit: Initializing Generative Adversarial Networks via Cooperative Learning
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 |
|---|---|---|---|
| Add cooperative initialization before adversarial training | StyleGAN2 · CIFAR-10 unconditional, no ADA | FID: 6.40 -> 4.34 | Helped |
| Add CoopInit to CUT | CUT · Cat→Dog | FID: 76.2 -> 61.3 | Helped |
| Add CoopInit to CUT | CUT · Horse→Zebra | FID: 45.5 -> 38.7 | Helped |
| Add CoopInit to conditional StyleGAN2 | StyleGAN2 w/ labels · ImageNet 32×32 conditional | FID: 3.87 -> 3.84 | No clear effect |
| Add CoopInit to BigGAN | BigGAN w/ labels · ImageNet 64×64 conditional | FID: 10.55 -> 10.63 | Hurt |
| Add CoopInit across adversarial loss variants | StyleGAN2 · CIFAR-10 generation | FID: NS: 13.95 -> 5.85; Hinge: 11.64 -> 5.09; WAS-GP: 13.21 -> 11.83 | Helped |
| Replace R1 regularization with CoopInit | StyleGAN2-NS · CIFAR-10 generation | FID: 15.8 -> 4.34 | Helped |
| Use CoopInit with ADA | StyleGAN2 · CIFAR-100 generation, 50K training examples | FID: 5.00 -> 5.10 | Hurt |
| Add CoopInit while varying the learning rate (ablation) | StyleGAN2-NS · CIFAR-10 generation, γ=0.00 | FID: 13.21 -> 6.16; 13.95 -> 6.07; 14.94 -> 5.85 | Helped |
| Use CoopInit with discriminator conditional layers (ablation) | StyleGAN2 · CIFAR-10 generation, conditional discriminator | FID: 7.94 -> 6.67; CoopInit conditional FID 6.67 vs unconditioned Orig 6.07 | Mixed |
| Use CoopInit with varying discriminator capacity while generator scale is fixed (ablation) | StyleGAN2 · CIFAR-10 generation, unbalanced D and G | FID: 35.52 -> 16.45; 22.14 -> 9.00; 13.95 -> 6.07; 14.02 -> 5.32; 13.15 -> 5.26 | Helped |
| Increase Langevin step size from 0.5 to 1.0 (ablation) | CoopInit · CIFAR-10 generation | FID: 6.61 -> 6.07 | Helped |
Closest papers by meaning
- CoT: Cooperative Training for Generative Modeling of Discrete Data (ICML 2019 · similarity 0.90)
- Generative Cooperative Networks for Natural Language Generation (ICML 2022 · similarity 0.89)
- SGAN: An Alternative Training of Generative Adversarial Networks (CVPR 2018 · similarity 0.89)
- Towards Principled Methods for Training Generative Adversarial Networks (ICLR 2017 · similarity 0.89)
- Training Generative Adversarial Networks in One Stage (CVPR 2021 · similarity 0.88)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · 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.