Research map / Label-efficient and robust learning / Recognition and long-tail
Whitening and Coloring Batch Transform for GANs
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 |
|---|---|---|---|
| Replace generator BN with WC | SN GAN generators · CIFAR-10 and STL-10 | IS: CIFAR-10: 7.58 ± .12 -> 7.84 ± .10; 8.22 ± .05 -> 8.66 ± .11; STL-10: 8.79 ± .14 -> 9.45 ± 0.18; 9.10 ± .04 -> 9.93 ± .1 | Helped |
| Replace generator BN with WC | GP + ResNet · CIFAR-10 | IS: 7.86 ± .07 -> 8.20 ± .08 | Helped |
| Replace generator cBN with cWC | SN + Projection Discriminator · CIFAR-10 and CIFAR-100 | IS: CIFAR-10: 8.62 -> 8.97 ± .11; CIFAR-100: 9.04 -> 9.52 ± .07 | Helped |
| Use class-filter soft assignment | cWC generator · Tiny ImageNet | IS: 10.43 ± .13 -> 11.78 ± .25 | Helped |
| Use class-filter soft assignment instead of full per-class filters | cWC generator · CIFAR-10 | IS: 8.97 ± .11 -> 8.85 ± .10 | Mixed |
| Add coloring after whitening (ablation) | WC generator · CIFAR-10 | IS: 6.63 ± .07 -> 8.66 ± .11 | Helped |
| Use full-matrix coloring instead of diagonal coloring (ablation) | WC generator · CIFAR-10 | IS: 7.00 ± .09 -> 8.66 ± .11 | Helped |
| Add whitening before coloring rather than coloring unnormalized features (ablation) | WC generator · CIFAR-10 | IS: 7.86 ± .08 -> 8.66 ± .11 | Helped |
| Use whitening rather than standardization before coloring (ablation) | WC generator · CIFAR-10 | IS: 8.34 ± .08 -> 8.66 ± .11 | Helped |
| Use Cholesky whitening rather than ZCA whitening (ablation) | WC generator · CIFAR-10 | IS: 8.29 ± .08 -> 8.66 ± .11 | Helped |
| Add the class-agnostic coloring branch (ablation) | cWC generator · CIFAR-100 | IS: 7.10 ± .06 -> 9.52 ± .07 | Helped |
| Use full-matrix class-specific coloring instead of diagonal matrices (ablation) | cWC generator · CIFAR-100 | IS: 8.82 ± .12 -> 9.52 ± .07 | Helped |
Closest papers by meaning
- An Investigation Into the Stochasticity of Batch Whitening (CVPR 2020 · similarity 0.92)
- Stochastic Whitening Batch Normalization (CVPR 2021 · similarity 0.88)
- Decorrelated Batch Normalization (CVPR 2018 · similarity 0.88)
- CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization (CVPR 2024 · similarity 0.88)
- Spectral Normalization for Generative Adversarial Networks (ICLR 2018 · similarity 0.88)
- Diversity-Sensitive Conditional Generative Adversarial Networks (ICLR 2019 · 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.