Research map / Image generation / GANs
ScoreMix: A Scalable Augmentation Strategy for Training GANs With Limited Data
Jie Cao, Mandi Luo, Junchi Yu, Ming–Hsuan Yang, Ran He
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 ScoreMix (ablation) | StyleGAN2 · FFHQ-1K | FID: 58.11 -> 24.21 | Improved |
| add ScoreMix | StyleGAN2+DiffAug · FFHQ-1K | FID: 26.13 -> 19.41 | Improved |
| add ScoreMix to wider StyleGAN2 with DiffAug (ablation) | StyleGAN2_wider + DiffAug · FFHQ-1K | FID: 26.13 -> 19.41 | Improved |
| add ScoreMix to narrower StyleGAN2 (ablation) | StyleGAN2_narrower · FFHQ-1K | FID: 55.91 -> 25.04 | Improved |
| add DiffAug on top of ScoreMix (ablation) | StyleGAN2 + ScoreMix · FFHQ-1K | FID: 24.21 -> 21.13 | Improved |
| replace mixup/CutMix/AugMix with ScoreMix | StyleGAN2+DiffAug · FFHQ-1K | FID: 22.95 -> 19.41 | Improved |
| increase static augmentation ratio mu from 0 to 10 (ablation) | ScoreMix · FFHQ-1K | FID: 58.10 -> 24.21 | Improved |
| increase mu from 10 to 100 (ablation) | ScoreMix · FFHQ-1K | FID: 1.31 lower | No clear effect |
| make augmentation ratio grow during training (ablation) | ScoreMix · FFHQ-1K | FID | Hurt |
| add ScoreMix with 70K data | StyleGAN2+DiffAug · FFHQ 70k | FID: 4.56 -> 4.32 | Improved |
| add ScoreMix with 10% ImageNet | BigGAN+DiffAug · ImageNet 10% data | FID: 49.67 -> 24.67 | Improved |
| add LC regularization to ScoreMix | ScoreMix · ImageNet 10% data | FID: 24.67 -> 22.51 | Improved |
Closest papers by meaning
- Improved Techniques for Training Score-Based Generative Models (NeurIPS 2020 · similarity 0.90)
- SMaRt: Improving GANs with Score Matching Regularity (ICML 2024 · similarity 0.89)
- AdaptiveMix: Improving GAN Training via Feature Space Shrinkage (CVPR 2023 · similarity 0.89)
- Adversarial Score Distillation: When score distillation meets GAN (CVPR 2024 · similarity 0.88)
- StudioGAN: A Taxonomy and Benchmark of GANs for Image Synthesis (TPAMI 2023 · similarity 0.87)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.86)
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.
