Research map / Detection and segmentation / Open-set and few-shot detection
Deep Metric Learning for Open World Semantic Segmentation
Jun Cen, Peng Yun, Junhao Cai, Michael Yu Wang, Ming Liu
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 DCE loss to VL (hybrid loss) (ablation) | DMLNet with EDS · StreetHazards | AUROC: 91.5 -> 93.7 | Improved |
| add hybrid VL+DCE loss instead of VL only (ablation) | DMLNet · StreetHazards | mIoU: 49.0 -> 53.9 | Improved |
| use EDS instead of MMSP criterion (ablation) | DMLNet (VL+DCE) · StreetHazards | AUPR: 8.8 -> 14.7 | Improved |
| combine MMSP with EDS (ablation) | DMLNet · StreetHazards | AUPR: 14.7 -> 14.1 | No clear effect |
| set prototype non-zero element T to 3 instead of 1 (ablation) | DMLNet · StreetHazards | FPR95: 35.1 -> 17.3 | Improved |
| vary gamma and beta of MMSP (ablation) | DMLNet · StreetHazards | AUPR: 12.0 to 14.7 | No clear effect |
| use PLM all heads instead of latest head only (ablation) | PLM · Cityscapes 16+1 (5 shot) | mIoU harm: 41.3 -> 69.2 | Improved |
| set lambda_novel to 1.5 instead of 1 (ablation) | NPM · Cityscapes 16+1 (5 shot) | mIoU harm: 50.8 -> 66.1 | Improved |
| increase training samples Q from 1 to 2975 (ablation) | PLM · Cityscapes 16+1 | mIoU harm: 62.2 -> 80.1 | Improved |
| use ground-truth labels instead of pseudo labels (ablation) | PLM · Cityscapes 16+1 | mIoU harm: 80.1 -> 80.0 | No clear effect |
| use DMLNet instead of MaxLogit | anomaly segmentation · StreetHazards | FPR95: 26.5 -> 17.3 | Improved |
Closest papers by meaning
- Open-World Semantic Segmentation Including Class Similarity (CVPR 2024 · similarity 0.87)
- Leveraging Depth and Language for Open-Vocabulary Domain-Generalized Semantic Segmentation (NeurIPS 2025 · similarity 0.86)
- Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models (CVPR 2024 · similarity 0.85)
- Segment Every Out-of-Distribution Object (CVPR 2024 · similarity 0.85)
- PDF: A Probability-Driven Framework for Open World 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.85)
- Distilling Spectral Graph for Object-Context Aware Open-Vocabulary Semantic Segmentation (CVPR 2025 · 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.
