Research map / Image generation / GANs
Disentangled Image Generation Through Structured Noise Injection
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 |
|---|---|---|---|
| use structured noise injection and separate spatially-variable codes from style codes | GAN generation on FFHQ · FFHQ 256×256 | Z path length: 280.58 -> 93.05 | Helped |
| use structured noise injection and separate spatially-variable codes from style codes | GAN generation on FFHQ · FFHQ 256×256 | FID: 10.73 -> 12.08 | Hurt |
| add per-pixel noise codes (ablation) | Ours without per-pixel noise · FFHQ 1024×1024 | Z linear separability: 45.36 | Helped |
| add per-pixel noise codes (ablation) | Ours without per-pixel noise · FFHQ 1024×1024 | FID: 5.92 | Mixed |
| start style modulation at 16×16 rather than applying it at all layers (ablation) | Ours without per-pixel noise · FFHQ 256×256 | Z path length: 258.69 -> 208.04 | Helped |
| start style modulation at 16×16 rather than 128×128 (ablation) | Ours without per-pixel noise · FFHQ 256×256 | W path length: 577.33 -> 111.13 | Helped |
| start style modulation at 16×16 rather than 64×64 (ablation) | Ours · FFHQ 256×256 | Z path length: 302.39 -> 93.05 | Helped |
| start style modulation at 16×16 rather than 64×64 (ablation) | Ours · FFHQ 256×256 | FID: 16.05 -> 12.08 | Helped |
| increase local code dimension from 8 to 16 (ablation) | Ours · FFHQ 256×256 | Z path length: 103.82 -> 93.05 | Helped |
| increase local code dimension from 16 to 32 (ablation) | Ours · FFHQ 256×256 | Z path length: 93.05 -> 108.44 | Hurt |
| increase local code dimension from 16 to 32 (ablation) | Ours · FFHQ 256×256 | FID: 12.08 -> 11.44 | Helped |
| add another 4×4 shared-code scale (ablation) | Ours · FFHQ | local change variance | Hurt |
Closest papers by meaning
- OOGAN: Disentangling GAN with One-Hot Sampling and Orthogonal Regularization (AAAI 2020 · similarity 0.89)
- MixNMatch: Multifactor Disentanglement and Encoding for Conditional Image Generation (CVPR 2020 · similarity 0.88)
- Interpreting the Latent Space of GANs for Semantic Face Editing (CVPR 2020 · similarity 0.88)
- IB-GAN: Disentangled Representation Learning with Information Bottleneck Generative Adversarial Networks (AAAI 2021 · similarity 0.87)
- InfoGAN-CR and ModelCentrality: Self-supervised Model Training and Selection for Disentangling GANs (ICML 2020 · similarity 0.87)
- Uncovering the Disentanglement Capability in Text-to-Image Diffusion Models (CVPR 2023 · 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.