Research map / Image generation / GANs
Feature Statistics Mixing Regularization for Generative Adversarial Networks
Check 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 FSMR to StyleGAN2 | StyleGAN2 · CIFAR-10, FFHQ, AFHQ, MetFaces, AFHQ Dog, AFHQ Cat, AFHQ Wild | FID: 3.89±0.07 -> 3.76±0.03; 5.62±0.10 -> 3.74±0.03; 11.37±0.03 -> 8.59±0.03; 51.88±0.44 -> 45.47±0.42; 19.65±0.07 -> 18.08±0 | Helped |
| Add FSMR to StyleGAN2-ADA | StyleGAN2-ADA · CIFAR-10, FFHQ, AFHQ, MetFaces, AFHQ Dog, AFHQ Cat, AFHQ Wild | FID: 3.23±0.06 -> 2.90±0.08; 4.05±0.07 -> 3.91±0.06; 7.73±0.11 -> 6.12±0.10; 29.17±0.08 -> 27.81±0.11; 13.56±0.10 -> 11.76±0. | Helped |
| Add FSMR to StyleGAN2-DiffAug | StyleGAN2-DiffAug · CIFAR-10, FFHQ, AFHQ, MetFaces, AFHQ Dog, AFHQ Cat, AFHQ Wild | FID: 3.23±0.08 -> 2.93±0.05; 5.35±0.09 -> 4.99±0.08; 7.52±0.08 -> 6.53±0.05; 32.96±0.08 -> 29.98±0.15; 16.92±0.06 -> 14.55±0. | Helped |
| Add FSMR to StyleGAN2 | StyleGAN2, unconditional CIFAR-10 generation · CIFAR-10 | inception score: 9.36 -> 9.58 | Helped |
| Add FSMR to StyleGAN2-ADA | StyleGAN2-ADA, unconditional CIFAR-10 generation · CIFAR-10 | inception score: 9.47 -> 9.68 | Helped |
| Add FSMR to StyleGAN2-DiffAug | StyleGAN2-DiffAug, unconditional CIFAR-10 generation · CIFAR-10 | inception score: 9.63 -> 9.81 | Helped |
| Add FSMR to the discriminator | DCGAN and bCRGAN · CIFAR-10, FFHQ, AFHQ, CelebA-HQ, LSUN Church | FID: DCGAN: 15.89±0.12 -> 14.98±0.09; 7.82±0.10 -> 6.76±0.08; 17.27±0.13 -> 13.19±0.09; 6.71±0.09 -> 5.23±0.10; 17.33±0.11 -> | Mixed |
| Use FSMR instead of on-the-fly stylization consistency regularization | DCGAN and bCRGAN · CIFAR-10, FFHQ, AFHQ, CelebA-HQ, LSUN Church | FID: DCGAN: 15.88±0.11 -> 14.98±0.09; 7.33±0.17 -> 6.76±0.08; 14.22±0.15 -> 13.19±0.09; 5.41±0.10 -> 5.23±0.10; 26.05±0.14 -> | Mixed |
| Use arbitrary mean and standard deviation in AdaIN instead of in-domain feature statistics mixing (ablation) | FSMR discriminator regularization · GAN training | training stability | Hurt |
| Add FSMR to StyleGAN2 for conditional CIFAR-10 generation | StyleGAN2 · CIFAR-10 | FID: 3.52 -> 3.35 | Helped |
| Add FSMR to StyleGAN2-ADA for conditional CIFAR-10 generation | StyleGAN2-ADA · CIFAR-10 | FID: 2.76 -> 2.63 | Helped |
| Add FSMR to StyleGAN2-DiffAug for conditional CIFAR-10 generation | StyleGAN2-DiffAug · CIFAR-10 | FID: 3.10 -> 2.87 | Helped |
Closest papers by meaning
- Stabilizing Training of Generative Adversarial Networks through Regularization (NeurIPS 2017 · similarity 0.90)
- Mixed batches and symmetric discriminators for GAN training (ICML 2018 · similarity 0.89)
- Refining Deep Generative Models via Discriminator Gradient Flow (ICLR 2021 · similarity 0.89)
- Regularizing Generative Adversarial Networks Under Limited Data (CVPR 2021 · similarity 0.89)
- NICE: NoIse-modulated Consistency rEgularization for Data-Efficient GANs (NeurIPS 2023 · similarity 0.89)
- Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · 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.