Research map / Detection and segmentation / Open-set and few-shot detection
Progressive Proxy Anchor Propagation for Unsupervised Semantic Segmentation
Hyun Seok Seong, WonJun Moon, SuBeen Lee, Jae-Pil Heo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 trustworthy positive set via proxy anchor propagation (TPS) (ablation) | PPAP contrastive baseline · COCO-stuff ViT-S/16 | unsupervised mIoU: 22.9 -> 25.0 | Improved |
| add ambiguity-excluded negative set (ANS) (ablation) | PPAP contrastive baseline · COCO-stuff ViT-S/16 | unsupervised mIoU: 22.9 -> 25.1 | Improved |
| combine TPS and ANS (ablation) | PPAP · COCO-stuff ViT-S/16 | unsupervised accuracy: 49.1 -> 62.9 | Improved |
| apply CRF post-processing (ablation) | PPAP · COCO-stuff ViT-S/16 | unsupervised accuracy: 59.9 -> 62.9 | Improved |
| replace HP with PPAP, both without CRF | COCO-stuff ViT-S/16 · COCO-stuff | unsupervised accuracy: 54.5 -> 59.9 | Improved |
| use PPAP instead of HP on larger-patch ViT-S/16 backbone | unsupervised segmentation · COCO-stuff | unsupervised accuracy: 54.5 -> 62.9 | Improved |
| use PPAP instead of HP on Potsdam-3 (only 3 classes) | ViT-B/8 · Potsdam-3 | unsupervised mIoU: 69.7 -> 71.0 | Improved |
| use PPAP instead of HP on Cityscapes ViT-S/8 linear probe | ViT-S/8 · Cityscapes | linear accuracy: 91.2 -> 90.8 | Hurt |
| use PPAP instead of HP on ImageNet-S | ViT-S/16 · ImageNet-S val | mIoU: 8.6 -> 25.7 | Improved |
| apply PPAP on iBoT ViT-B/16 backbone instead of HP | iBoT · COCO-stuff | unsupervised mIoU: 22.4 -> 27.6 | Improved |
| use PPAP instead of HP on few-shot (rare) classes | COCO-stuff · COCO-stuff class frequency | mIoU (Few): 28.9 -> 32.9 | Improved |
| relocate proxy anchor progressively (PPAP) instead of k-means grouping for positives | positive set selection · COCO-stuff ViT-S/16 | positive precision: 22% -> 72% | Improved |
Closest papers by meaning
- Leveraging Hidden Positives for Unsupervised Semantic Segmentation (CVPR 2023 · similarity 0.89)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.87)
- Category Anchor-Guided Unsupervised Domain Adaptation for Semantic Segmentation (NeurIPS 2019 · similarity 0.87)
- Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation (AAAI 2025 · similarity 0.86)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.86)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 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.
