Research map / Detection and segmentation / Open-set and few-shot detection
Rethinking Evaluation Metrics of Open-Vocabulary Segmentation
Hao Zhou, Lu Qi, Tiancheng Shen, Hai Huang, Xu Yang, Xiangtai Li, Ming–Hsuan Yang
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 |
|---|---|---|---|
| replace vanilla mIoU with Open mIoU | FC-CLIP · ADE20K A-150 | mIoU: 34.1 -> 40.4 (+6.3) | Improved |
| replace vanilla mIoU with Open mIoU | SimSeg · A-847 | mIoU: 6.7 -> 12.9 (+6.2) | Improved |
| replace vanilla mIoU with Open mIoU on easy dataset | SAN · PAS-20 | mIoU: 95.2 -> 96.0 (+0.8) | Improved |
| replace vanilla PQ with Open PQ | OPSNet · ADE20K A-150 (target) | PQ: 17.8 -> 21.9 (+4.1) | Improved |
| replace vanilla PQ with Open PQ | FC-CLIP · COCO (source) | PQ: 54.4 -> 55.3 (+0.9) | Improved |
| replace vanilla SQ with Open SQ | OPSNet · COCO (source) | SQ: 83.5 -> 83.4 (-0.1) | No clear effect |
| replace vanilla SQ with Open SQ | OPSNet · ADE20K A-150 (target) | SQ: 54.6 -> 76.0 (+21.4) | Improved |
| add class-agnostic global ranking (vanilla AP dagger) then Open AP | ODISE · ADE20K A-150 | AP: 12.7 -> 14.3 (+1.6) | Improved |
| replace vanilla AP with Open AP | RegionCLIP · COCO OVOD generalized novel | AP50: 30.1 -> 33.6 (+3.5) | Improved |
| use GloVe similarity instead of Path similarity in Open PQ (ablation) | FC-CLIP · ADE20K A-150 | PQ: 28.8 -> 31.5 (+4.7 over vanilla 26.8) | Hurt |
| use T5 similarity instead of Path similarity in Open PQ (ablation) | FC-CLIP · ADE20K A-150 | PQ: 28.8 -> 30.7 (+3.9 over vanilla 26.8) | Hurt |
| use Path similarity rather than GloVe in user study (ablation) | Open PQ evaluation · User study satisfaction | satisfaction score: 3.05 -> 6.12 | Improved |
Closest papers by meaning
- Open-Vocabulary Segmentation with Semantic-Assisted Calibration (CVPR 2024 · similarity 0.87)
- Transferable and Principled Efficiency for Open-Vocabulary Segmentation (CVPR 2024 · similarity 0.87)
- Renovating Names in Open-Vocabulary Segmentation Benchmarks (NeurIPS 2024 · similarity 0.86)
- Mask-Adapter: The Devil is in the Masks for Open-Vocabulary Segmentation (CVPR 2025 · similarity 0.86)
- Excluding the Impossible for Open Vocabulary Semantic Segmentation (AAAI 2025 · similarity 0.86)
- USE: Universal Segment Embeddings for Open-Vocabulary Image Segmentation (CVPR 2024 · 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.
