Research map / Image generation / GANs
Extracting Semantic Knowledge From GANs With Unsupervised Learning
Jianjin Xu, Zhaoxiang Zhang, Xiaolin Hu
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 |
|---|---|---|---|
| use KLiSH clustering instead of AHC (arccos) | GAN feature clustering · StyleGAN2-FFHQ, K=26 | MIoU: 43.8 -> 56.3 | Improved |
| use KLiSH instead of K-means | GAN feature clustering · StyleGAN-CelebAHQ, K=30 | MIoU: 38.5 -> 45.7 | Improved |
| use KLiSH instead of AHC (arccos) | GAN feature clustering · PGGAN-CelebAHQ, K=26 | AMI: 36.0 -> 51.0 | Improved |
| use KLiSH instead of AHC (arccos) | GAN feature clustering · StyleGAN2-FFHQ, K=26 | ARI: 69.2 vs 65.4 | Hurt |
| use KLiSH labels instead of AHC labels for unsupervised segmentation | UFGS · CelebAMask-HQ | mIoU: 36.6 -> 47.2 | Improved |
| compare zero-annotation UFGS against one-shot LSE | UFGS (KLiSH) · CelebAMask-HQ | mIoU: 47.2 vs 47.1 | No clear effect |
| compare UFGS against 10-shot LSE | UFGS (KLiSH) · CelebAMask-HQ | mIoU: 47.2 vs 58.5 | Hurt |
| replace SVM with softmax-trained LSE (classical linear separability) (ablation) | feature segmentation probe · StyleGAN2-FFHQ | mIoU: 81.7 vs 82.7 | No clear effect |
| replace SVM with LSE (ablation) | feature segmentation probe · StyleGAN-Bedroom | mIoU: 30.9 vs 37.3 | Hurt |
| use centroids of features instead of linear classifier (ablation) | K-means-style clustering assumption · StyleGAN2-FFHQ | mIoU: 82.7 -> 49.6 | Hurt |
Closest papers by meaning
- Linear Semantics in Generative Adversarial Networks (CVPR 2021 · similarity 0.90)
- Repurposing GANs for One-Shot Semantic Part Segmentation (CVPR 2021 · similarity 0.90)
- KE-GAN: Knowledge Embedded Generative Adversarial Networks for Semi-Supervised Scene Parsing (CVPR 2019 · similarity 0.89)
- Region-Based Semantic Factorization in GANs (ICML 2022 · similarity 0.88)
- Semi Supervised Semantic Segmentation Using Generative Adversarial Network (ICCV 2017 · similarity 0.88)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · 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.
