Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised StyleGAN for Disentanglement Learning
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 |
|---|---|---|---|
| add the smoothness consistency regularization (ablation) | Semi-StyleGAN · Isaac3D (η = 0.1%) | MIG-gen: 0.945 $\pm$ 0.011 -> 0.965 $\pm$ 0.014 | Helped |
| add the smoothness consistency regularization (ablation) | Semi-StyleGAN · Falcor3D (η = 0.5%) | MIG-gen: 0.632 $\pm$ 0.058 -> 0.754 $\pm$ 0.017 | Helped |
| increase the encoder unsupervised loss weight to γ_E = 10 (ablation) | Semi-StyleGAN · Isaac3D (η = 0.1%) | MIG-gen: 0.965 $\pm$ 0.014 -> 0.888 $\pm$ 0.087 | Hurt |
| increase the encoder unsupervised loss weight to γ_E = 10 (ablation) | Semi-StyleGAN · Isaac3D (η = 0.1%) | MIG: 0.812 $\pm$ 0.020 -> 0.880 $\pm$ 0.120 | Mixed |
| increase the encoder unsupervised loss weight to γ_E = 10 (ablation) | Semi-StyleGAN · Falcor3D (η = 0.5%) | MIG-gen: 0.754 $\pm$ 0.017 -> 0.636 $\pm$ 0.065 | Hurt |
| remove the generator unsupervised loss (ablation) | Semi-StyleGAN · Isaac3D (η = 0.1%) | MIG-gen: 0.965 $\pm$ 0.014 -> 0.130 $\pm$ 0.054 | Hurt |
| remove the generator unsupervised loss (ablation) | Semi-StyleGAN · Falcor3D (η = 0.5%) | MIG-gen: 0.754 $\pm$ 0.017 -> 0.034 $\pm$ 0.028 | Hurt |
| use limited labeled data (0.25%) rather than no labels | Semi-StyleGAN · Isaac3D | encoder and generator disentanglement | Helped |
| use limited labeled data (2.5%) rather than no labels | Semi-StyleGAN · Falcor3D | disentanglement | Helped |
| use StyleGAN architecture with mutual-information code reconstruction | Info-StyleGAN · Isaac3D (128x128) | FID: 80.72 $\pm$ 30.79 -> 2.19 $\pm$ 0.48 | Helped |
Closest papers by meaning
- Unsupervised Disentanglement of Linear-Encoded Facial Semantics (CVPR 2021 · similarity 0.90)
- StyleSpace Analysis: Disentangled Controls for StyleGAN Image Generation (CVPR 2021 · similarity 0.89)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.89)
- Tripod: Three Complementary Inductive Biases for Disentangled Representation Learning (ICML 2024 · similarity 0.88)
- Disentangled Representation Learning via Flow Matching (NeurIPS 2026 · similarity 0.87)
- SDGAN: Disentangling Semantic Manipulation for Facial Attribute Editing (AAAI 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.