Research map / Detection and segmentation / Open-set and few-shot detection
FineGAN: Unsupervised Hierarchical Disentanglement for Fine-Grained Object Generation and Discovery
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 |
|---|---|---|---|
| hierarchically disentangle background, shape, and appearance with FineGAN | Fine-grained image generation · CUB, Stanford Dogs, and Stanford Cars | Inception Score: 52.53 ± 0.45 vs 47.32 ± 0.77; 46.92 ± 0.61 vs 43.16 ± 0.42; 32.62 ± 0.37 vs 33.69 ± 0.44 | Helped |
| hierarchically disentangle background, shape, and appearance with FineGAN | Fine-grained image generation · CUB, Stanford Dogs, and Stanford Cars | FID: 11.25 vs 13.20; 25.66 vs 29.34; 16.03 vs 16.28 | Helped |
| use FineGAN's learned hierarchical features for clustering | Unsupervised fine-grained clustering · CUB, Stanford Dogs, and Stanford Cars | Accuracy: 0.126 vs 0.061; 0.079 vs 0.054; 0.078 vs 0.063 | Helped |
| set parent-code count Np to 10 rather than 20 (ablation) | FineGAN · CUB | Inception Score: 52.53 -> 52.11 | No clear effect |
| set parent-code count Np to 40 rather than 20 (ablation) | FineGAN · CUB | Inception Score: 52.53 -> 49.62 | Hurt |
| set parent-code count Np to 5 rather than 20 (ablation) | FineGAN · CUB | Inception Score: 52.53 -> 46.68 | Hurt |
| use variable numbers of children per parent (ablation) | FineGAN · CUB | Inception Score: 52.53 -> 51.83 | No clear effect |
| use only child-code features instead of concatenated parent and child features (ablation) | FineGAN clustering · Birds | Accuracy: drops 0.017 in Accuracy | Hurt |
| add a parent code to InfoGAN and train it to predict parent and child codes (ablation) | InfoGAN · Birds | Inception Score: 43.16 -> 48.06 | Helped |
| add a parent code to InfoGAN and train it to predict parent and child codes (ablation) | InfoGAN · Birds | FID: 13.20 -> 12.84 | Helped |
| apply hard-negative training to degenerate generated images (ablation) | FineGAN · Fine-grained image generation | degenerate image occurrence | Helped |
| apply hard-negative training to InfoGAN (ablation) | InfoGAN · Image generation | performance | No clear effect |
Closest papers by meaning
- ComGAN: Unsupervised Disentanglement and Segmentation via Image Composition (NeurIPS 2022 · similarity 0.87)
- OOGAN: Disentangling GAN with One-Hot Sampling and Orthogonal Regularization (AAAI 2020 · similarity 0.86)
- MixNMatch: Multifactor Disentanglement and Encoding for Conditional Image Generation (CVPR 2020 · similarity 0.86)
- When, Why, and Which Pretrained GANs Are Useful? (ICLR 2022 · similarity 0.86)
- Context-Aware Layout to Image Generation With Enhanced Object Appearance (CVPR 2021 · similarity 0.86)
- PandA: Unsupervised Learning of Parts and Appearances in the Feature Maps of GANs (ICLR 2023 · similarity 0.86)
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.