Research map / Image generation / Autoregressive and latent generation
GANSeg: Learning To Segment by Unsupervised Hierarchical Image Generation
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 hierarchical GANSeg instead of the strongest listed part-segmentation baseline | CelebA-in-the-wild · CelebA-in-the-wild (K=8) | landmark regression error: 8.40% -> 6.18% | Helped |
| use hierarchical GANSeg instead of the strongest listed baseline | CUB-2011 · CUB-2011 | foreground IoU: 0.551 -> 0.629 | Helped |
| use hierarchical GANSeg instead of the strongest listed baseline | Flowers · Flowers | foreground IoU: 0.789 -> 0.739 | Hurt |
| use hierarchical GANSeg instead of video-trained Siarohin et al. | Taichi · Taichi (K=10) | IoU: 0.7686 -> 0.8538 | Helped |
| use hierarchical GANSeg instead of video-trained Siarohin et al. | Taichi · Taichi (K=10) | MAE: 389.78 -> 417.17 | Hurt |
| estimate σ per part and object rather than use a fixed σ (ablation) | GANSeg · CelebA-in-the-wild | landmark regression error: 26.32% -> 6.18% | Helped |
| estimate σ per part and object rather than use a fixed σ (ablation) | GANSeg · CUB-2011 | foreground IoU: 0.467 -> 0.629 | Helped |
| estimate σ per part and object rather than use a fixed σ (ablation) | GANSeg · Taichi | IoU: 0.7452 -> 0.8538 | Helped |
| use 6 points per part instead of 4 (ablation) | GANSeg · CUB-2011 | foreground IoU: 0.629 -> 0.682 | Helped |
| use 3 points per part instead of 4 (ablation) | GANSeg · CelebA-in-the-wild | landmark regression error: 6.18% -> 6.08% | Helped |
| use 3 points per part instead of 4 (ablation) | GANSeg · Taichi | IoU: 0.8538 -> 0.5863 | Hurt |
| add the geometric concentration loss (ablation) | GANSeg · generated masks across datasets | mask coherence | Helped |
Closest papers by meaning
- Object Segmentation Without Labels with Large-Scale Generative Models (ICML 2021 · similarity 0.90)
- Repurposing GANs for One-Shot Semantic Part Segmentation (CVPR 2021 · similarity 0.89)
- Finding an Unsupervised Image Segmenter in each of your Deep Generative Models (ICLR 2022 · similarity 0.89)
- A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract) (AAAI 2025 · similarity 0.86)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.85)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.85)
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.