Research map / Label-efficient and robust learning / Recognition and long-tail
When, Why, and Which Pretrained GANs Are Useful?
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 |
|---|---|---|---|
| Initialize transfer from a pretrained checkpoint rather than random weights | StyleGAN2 · Flowers | FID: 10.73 -> 8.31 | Helped |
| Initialize transfer from a pretrained checkpoint rather than random weights | StyleGAN2 · Satellite-Landscapes | FID: 26.6 -> 21.0 | Helped |
| Choose a diverse Imagenet source instead of random initialization | StyleGAN2 · CIFAR-10 | FID: 9.33 -> 6.2 | Helped |
| Choose a suitable pretrained source for the target rather than random initialization | StyleGAN2 · Simpsons | FID: 8.42 -> 7.87 | Helped |
| Use a source checkpoint that passed two times fewer training steps instead of a converged checkpoint (ablation) | StyleGAN2 transferred to Flowers from Imagenet · Flowers | FID: 8.31 -> 8.54 | Hurt |
| Use a source checkpoint that passed two times fewer training steps instead of a converged checkpoint (ablation) | StyleGAN2 transferred to Flowers from FFHQ · Flowers | FID: 9.47 -> 9.5 | No clear effect |
| Initialize both generator and discriminator from suitable pretrained checkpoints rather than initializing only one or neither (ablation) | GAN finetuning · Ten-Gaussian synthetic target | Wasserstein-1 distance | Helped |
| Select a checkpoint using Recall rather than Precision | GAN transfer checkpoint selection · Target datasets | Best-source identification failures: 2 vs 11 | Helped |
Closest papers by meaning
- Ensembling Off-the-Shelf Models for GAN Training (CVPR 2022 · similarity 0.92)
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.90)
- Are GANs Created Equal? A Large-Scale Study (NeurIPS 2018 · similarity 0.89)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · similarity 0.88)
- RG-GAN: Dynamic Regenerative Pruning for Data-Efficient Generative Adversarial Networks (AAAI 2024 · 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.