Research map / Image generation / GANs
GAN Cocktail: Mixing GANs without Dataset Access
Omri Avrahami, Dani Lischinski, Ohad Fried
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 |
|---|---|---|---|
| train on generated images instead of real data (ablation) | UnionGAN (N=1) · FFHQ | FID: 5.58 -> 8.84 | Hurt |
| train on generated images instead of real data (ablation) | UnionGAN (N=1) · LSUN cat | FID: 17.37 -> 21.78 | Hurt |
| train on generated images instead of real data (ablation) | UnionGAN (N=1) · LSUN car | FID: 7.12 -> 12.79 | Hurt |
| add model rooting and weight averaging over from-scratch training | GAN Cocktail · Union FID, cat+dog | FID: 27.58 -> 20.77 | Improved |
| add model rooting and averaging over TransferGAN | GAN Cocktail · Union FID, cat+dog | FID: 22.17 -> 20.77 | Improved |
| add model rooting and averaging over Freeze-D | GAN Cocktail · Union FID, FFHQ+cat | FID: 18.17 -> 16.44 | Improved |
| add model rooting and averaging over EWC | GAN Cocktail · Union FID, FFHQ+cat+dog+car | FID: 19.14 -> 18.44 | Improved |
| add more source datasets (FFHQ+cat+dog to four datasets) | GAN Cocktail · Union FID | FID: 18.98 -> 18.44 | Hurt |
| add EWC regularization to TransferGAN | TransferGAN · LSUN cat+FFHQ, cat class | FID: 32.08 -> 35.36 (FFHQ-root cat) | Hurt |
| choose LSUN dog rather than LSUN cat as root | GAN Cocktail · LSUN cat + dog | FID: 20.77 -> 20.03 | Improved |
| choose LSUN cat rather than FFHQ as root | GAN Cocktail · LSUN cat + FFHQ | FID: 16.44 -> 15.60 | Improved |
| root with a custom architecture instead of StyleGAN2 | GAN Cocktail · LSUN cat + dog | FID: 19.42 -> 21.28 | Hurt |
Closest papers by meaning
- AdaGAN: Boosting Generative Models (NeurIPS 2017 · similarity 0.89)
- InfoGAN-CR and ModelCentrality: Self-supervised Model Training and Selection for Disentangling GANs (ICML 2020 · similarity 0.88)
- Rethinking Layer-wise Model Merging through Chain of Merges (NeurIPS 2026 · similarity 0.87)
- GLeaD: Improving GANs With a Generator-Leading Task (CVPR 2023 · similarity 0.87)
- RG-GAN: Dynamic Regenerative Pruning for Data-Efficient Generative Adversarial Networks (AAAI 2024 · similarity 0.87)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.87)
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.
