Research map / Image generation / GANs
PandA: Unsupervised Learning of Parts and Appearances in the Feature Maps of 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 |
|---|---|---|---|
| use PandA feature-map editing instead of the strongest baseline | StyleGAN2 · FFHQ, Eyes | ROIR: 1.26±0.70 -> 1.04 ± 0.33 | Helped |
| use PandA feature-map editing instead of the strongest baseline | StyleGAN2 · FFHQ, Nose | ROIR: 1.70±0.82 -> 1.17 ± 0.44 | Helped |
| use PandA feature-map editing instead of the strongest baseline | StyleGAN2 · FFHQ, Open mouth | ROIR: 1.24±0.44 -> 1.04 ± 0.39 | Helped |
| use PandA feature-map editing instead of the strongest baseline | StyleGAN2 · FFHQ, Smile | ROIR: 1.87±0.75 -> 1.05 ± 0.38 | Helped |
| use PandA feature-map editing instead of StyleSpace | StyleGAN2 · FFHQ, additional six local edits | ROIR: 2.52 ± 0.91 -> 0.80 ± 0.23; 1.99 ± 1.28 -> 0.51 ± 0.19; 1.95 ± 1.13 -> 0.41 ± 0.13; 1.54 ± 1.14 -> 0.68 ± 0.24; 1.50 ± 1 | Helped |
| use PandA instead of the strongest baseline | StyleGAN2 · FFHQ, Eyes | FID: 29.37 -> 29.91 | Hurt |
| use PandA instead of the strongest baseline | StyleGAN2 · FFHQ, Nose | FID: 29.18 -> 29.77 | Hurt |
| use PandA instead of the strongest baseline | StyleGAN2 · FFHQ, Open mouth | FID: 29.63 -> 29.26 | Helped |
| use PandA instead of the strongest baseline | StyleGAN2 · FFHQ, Smile | FID: 29.45 -> 28.91 | Helped |
| add non-negativity constraints to parts factors (ablation) | PandA · feature-map part discovery | semantic locality | Helped |
| combine the spatial modes before factorization instead of decomposing the original tensor (ablation) | PandA · feature-map part discovery | semantic locality | Helped |
| refine global parts into sample-specific parts (ablation) | PandA · unaligned samples | part alignment | Helped |
Closest papers by meaning
- Region-Based Semantic Factorization in GANs (ICML 2022 · similarity 0.91)
- Editing in Style: Uncovering the Local Semantics of GANs (CVPR 2020 · similarity 0.90)
- EditGAN: High-Precision Semantic Image Editing (NeurIPS 2021 · similarity 0.90)
- Closed-Form Factorization of Latent Semantics in GANs (CVPR 2021 · similarity 0.90)
- Low-Rank Subspaces in GANs (NeurIPS 2021 · similarity 0.89)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · 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.