Research map / Detection and segmentation / Open-set and few-shot detection
Object Segmentation Without Labels with Large-Scale Generative Models
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 |
|---|---|---|---|
| Sample latent codes from the target dataset's encoder embeddings instead of the standard prior (ablation) | BigBiGAN segmentation pipeline · ECSSD | max Fβ: 0.737 -> 0.773 | Helped |
| Sample latent codes from the target dataset's encoder embeddings instead of the standard prior (ablation) | BigBiGAN segmentation pipeline · DUTS-test | max Fβ: 0.575 -> 0.616 | Helped |
| Sample latent codes from the target dataset's encoder embeddings instead of the standard prior (ablation) | BigBiGAN segmentation pipeline · DUT-OMRON | max Fβ: 0.498 -> 0.533 | Helped |
| Add mask-size filtering (ablation) | E-BigBiGAN · DUT-OMRON | max Fβ: 0.533 -> 0.552 | Helped |
| Add histogram filtering (ablation) | E-BigBiGAN · ECSSD | max Fβ: 0.781 -> 0.779 | Hurt |
| Add histogram filtering (ablation) | E-BigBiGAN · DUTS-test | max Fβ: 0.62 -> 0.621 | Helped |
| Add connected-components filtering (ablation) | E-BigBiGAN · ECSSD | max Fβ: 0.779 -> 0.797 | Helped |
| Add connected-components filtering (ablation) | E-BigBiGAN · DUTS-test | max Fβ: 0.621 -> 0.624 | Helped |
| Add connected-components filtering (ablation) | E-BigBiGAN · DUT-OMRON | max Fβ: 0.555 -> 0.563 | Helped |
| Add latent-space noise around dataset embeddings (α=0.2) (ablation) | E-BigBiGAN · CUB-200-2011 | max Fβ: 0.750 -> 0.834 | Helped |
| Add latent-space noise around dataset embeddings (α=0.2) (ablation) | E-BigBiGAN · Flowers | IoU: 0.689 -> 0.804 | Helped |
| Use U²Net-generated masks instead of the latent-shift masks (ablation) | E-BigBiGAN · DUT-OMRON | IoU: 0.464 -> 0.559 | Helped |
Closest papers by meaning
- Finding an Unsupervised Image Segmenter in each of your Deep Generative Models (ICLR 2022 · similarity 0.91)
- GANSeg: Learning To Segment by Unsupervised Hierarchical Image Generation (CVPR 2022 · similarity 0.90)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.88)
- unMORE: Unsupervised Multi-Object Segmentation via Center-Boundary Reasoning (ICML 2025 · similarity 0.87)
- Unsupervised Universal Image Segmentation (CVPR 2024 · similarity 0.87)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · 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.