Research map / Detection and segmentation / Open-set and few-shot detection
ILSGAN: Independent Layer Synthesis for Unsupervised Foreground-Background Segmentation
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 ILS-MI loss with weight 1.0 (ablation) | ILSGAN · Cars | IoU: 80.6 -> 83.5; 2.9 IoU | Helped |
| Add ILS-MI loss with weight 1.0 (ablation) | ILSGAN · CUB | IoU: 68.6 -> 72.3; 3.7 IoU | Helped |
| Add ILS-MI loss with weight 1.0 (ablation) | ILSGAN · Dogs | IoU: 81.7 -> 83.8; 2.1 IoU | Helped |
| Increase ILS-MI weight to 5.0 (ablation) | ILSGAN · Cars, CUB, and Dogs | IoU: 83.5 -> 61.7; 72.3 -> 49.3; 83.8 -> 44.2 | Hurt |
| Use ILS-L1 instead of ILS-MI (ablation) | ILSGAN · Cars, CUB, and Dogs | IoU: 80.6 -> 82.1; 68.6 -> 72.5; 81.7 -> 82.9 | Mixed |
| Optimize visible regions with ILS-MI (ablation) | ILSGAN · Cars, CUB, and Dogs | IoU: 79.6 -> 83.5; 70.9 -> 72.3; 84.4 -> 83.8 | Mixed |
| Optimize the mask with ILS-MI (ablation) | ILSGAN · Cars, CUB, and Dogs | IoU: 81.3 -> 83.5; 70.2 -> 72.3; 82.9 -> 83.8 | Helped |
| Separate visible and invisible regions for the MI loss (ablation) | ILSGAN · Cars, CUB, and Dogs | IoU: 82.0 -> 83.5; 69.5 -> 72.3; 82.9 -> 83.8 | Mixed |
| Use Gaussian rather than Laplace for the variational distribution (ablation) | ILSGAN · Cars, CUB, and Dogs | IoU: 77.3; 59.6; 79.0 | Hurt |
| Use ILSGAN rather than DRC | unsupervised segmentation · Cars | IoU: +10.4 IoU | Helped |
| Use ILSGAN rather than Impr.LGAN | unsupervised segmentation · CUB | IoU: +2.4 IoU | Helped |
| Use ILSGAN rather than DRC | unsupervised segmentation · Dogs | IoU: +14.0 IoU | Helped |
Closest papers by meaning
- Finding an Unsupervised Image Segmenter in each of your Deep Generative Models (ICLR 2022 · similarity 0.89)
- Learning Segmentation Masks with the Independence Prior (AAAI 2019 · similarity 0.86)
- Object Segmentation Without Labels with Large-Scale Generative Models (ICML 2021 · similarity 0.86)
- BachGAN: High-Resolution Image Synthesis From Salient Object Layout (CVPR 2020 · similarity 0.85)
- ComGAN: Unsupervised Disentanglement and Segmentation via Image Composition (NeurIPS 2022 · similarity 0.85)
- UGround: Towards Unified Visual Grounding with Unrolled Transformers (ICML 2026 · similarity 0.83)
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.