Research map / Detection and segmentation / Open-set and few-shot detection
MasQCLIP for Open-Vocabulary Universal Image Segmentation
Xin Xu, Tianyi Xiong, Zheng Ding, Zhuowen Tu
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 object score head to class-agnostic mask proposal network (ablation) | MasQCLIP (Mask2Former) · COCO instance novel classes | mask AP50: 11.3 -> 22.6 | Improved |
| add two rounds of progressive distillation of novel masks (ablation) | MasQCLIP (Mask2Former) · COCO instance novel classes | mask AP50: 22.6 -> 31.9 | Improved |
| add a third round of progressive distillation (ablation) | MasQCLIP (Mask2Former) · COCO instance novel classes | mask AP50: 31.9 -> 31.2 | No clear effect |
| progressive distillation on base classes (ablation) | MasQCLIP (Mask2Former) · COCO instance base classes | mask AP50: 53.8 -> 51.0 | Hurt |
| train teacher longer (200k) instead of distillation (ablation) | MasQCLIP (Mask2Former) · COCO instance novel classes | mask AP50: 22.6 -> 18.2 | Hurt |
| tune query projections of Mask Class Tokens versus freezing CLIP (ablation) | MasQ-Tuning · COCO panoptic | PQ: 19.4 -> 48.5 | Improved |
| tune query only instead of value projections (ablation) | MasQ-Tuning · ADE20k panoptic | PQ: 14.4 -> 23.3 | Improved |
| tune only query instead of full CLIP copy (ablation) | MasQ-Tuning · COCO panoptic | PQ: 49.0 -> 48.5 | No clear effect |
| tune query only instead of query and key (ablation) | MasQ-Tuning · ADE20k panoptic | PQ: 22.9 -> 23.3 | Improved |
| use Mask R-CNN instead of Mask2Former as proposal network | MasQCLIP · ADE20k panoptic | PQ: 23.3 -> 14.9 | Hurt |
Closest papers by meaning
- Open-Vocabulary Universal Image Segmentation with MaskCLIP (ICML 2023 · similarity 0.91)
- USE: Universal Segment Embeddings for Open-Vocabulary Image Segmentation (CVPR 2024 · similarity 0.88)
- Towards Open-Vocabulary Semantic Segmentation Without Semantic Labels (NeurIPS 2024 · similarity 0.88)
- Effective SAM Combination for Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.86)
- OpenWorldSAM: Extending SAM2 for Universal Image Segmentation with Language Prompts (NeurIPS 2025 · similarity 0.86)
- Demystifying CLIP Data (ICLR 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.
