Research map / Detection and segmentation / CNN features
Leveraging Hidden Positives 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 global hidden positives and local hidden positives to STEGO | DINO ViT-S/8 segmentation · COCO-stuff | unsupervised accuracy: 48.3 -> 57.2 | Improved |
| Add hidden-positive training | DINO ViT-B/8 segmentation · Potsdam-3 | unsupervised accuracy: 77.0 -> 82.4 | Improved |
| Add task-specific GHP and LHP together (ablation) | HP · COCO-stuff | unsupervised accuracy: 49.3 -> 57.2 | Improved |
| Add LHP (ablation) | HP without task-specific GHP and LHP · COCO-stuff | unsupervised accuracy: 49.3 -> 52.5 | Improved |
| Add task-specific GHP (ablation) | HP without task-specific GHP and LHP · COCO-stuff | unsupervised accuracy: 49.3 -> 55.0 | Improved |
| Use symmetric GHP assignment (ablation) | HP · COCO-stuff | unsupervised accuracy: 54.0 -> 57.2 | Improved |
| Add consistency regularizer (ablation) | HP · COCO-stuff | unsupervised accuracy: 55.8 -> 57.2 | Improved |
| Replace hidden-positive training with naive contrastive learning and photometric perturbations (ablation) | DINO ViT-S/8 segmentation · COCO-stuff | unsupervised accuracy: 37.8 -> 57.2 | Hurt |
| Replace gradient propagation with loss propagation (ablation) | DINO ViT-S/16 HP · COCO-stuff | unsupervised mIoU: 24.3 -> 23.2 | No clear effect |
| Replace gradient propagation with loss propagation (ablation) | SelfPatch HP · COCO-stuff | unsupervised accuracy: 56.1 -> 54.5 | No clear effect |
| Replace gradient propagation with loss propagation (ablation) | DINO ViT-S/16 HP · COCO-stuff | linear mIoU: 39.1 -> 40.5 | Improved |
| Replace gradient propagation with loss propagation (ablation) | SelfPatch HP · COCO-stuff | linear mIoU: 41.3 -> 41.4 | No clear effect |
Closest papers by meaning
- UniMatch V2: Pushing the Limit of Semi-Supervised Semantic Segmentation (TPAMI 2025 · similarity 0.90)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.89)
- Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation (AAAI 2025 · similarity 0.89)
- Label-Efficient Few-Shot Semantic Segmentation with Unsupervised Meta-Training (AAAI 2024 · similarity 0.88)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.88)
- Learn to Rectify the Bias of CLIP for Unsupervised Semantic Segmentation (CVPR 2024 · similarity 0.87)
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.
