Research map / Detection and segmentation / Open-set and few-shot detection
ProposalCLIP: Unsupervised Open-Category Object Proposal Generation via Exploiting CLIP Cues
Hengcan Shi, Munawar Hayat, Yicheng Wu, Jianfei Cai
Check 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 |
|---|---|---|---|
| Use CLIP-based proposal selection, graph merging, and proposal regression | Edge Boxes proposals · PASCAL VOC 2007 test, 10 proposals | Recall@0.5 (%): 42.5 -> 52.1 | Improved |
| Use CLIP-based proposal selection, graph merging, and proposal regression | Edge Boxes proposals · COCO validation, 10 proposals | Recall@0.5 (%): 17.1 -> 27.1 | Improved |
| Use ProposalCLIP instead of Edge Boxes | Edge Boxes proposals · Visual Genome mini, 10 proposals | Recall@0.5 (%): 14.5 -> 24.1 | Improved |
| Add CLIP proposal selection (ablation) | Edge Boxes on PASCAL VOC 2007 test · PASCAL VOC 2007 test | Recall@0.5 (%): 15.2 -> 21.0; 42.5 -> 51.8; 58.3 -> 62.1; 64.7 -> 68.5 | Improved |
| Add graph-based proposal merging (ablation) | CLIP-selected proposals on PASCAL VOC 2007 test · PASCAL VOC 2007 test | Recall@0.5 (%): 21.0 -> 21.6; 51.8 -> 52.4; 62.1 -> 64.6; 68.5 -> 70.8 | Improved |
| Add proposal regression (ablation) | CLIP selection and graph merging on PASCAL VOC 2007 test · PASCAL VOC 2007 test, 1 proposal | Recall@0.5 (%): 21.6 -> 22.1 | Hurt |
| Add proposal regression (ablation) | CLIP selection and graph merging on PASCAL VOC 2007 test · PASCAL VOC 2007 test, 10 proposals | Recall@0.5 (%): 52.4 -> 52.1 | Hurt |
| Add proposal regression (ablation) | CLIP selection and graph merging on PASCAL VOC 2007 test · PASCAL VOC 2007 test, 30 proposals | Recall@0.5 (%): 64.6 -> 65.8 | Improved |
| Use the initial score alone instead of the final objectness score (ablation) | ProposalCLIP on PASCAL VOC 2007 test · PASCAL VOC 2007 test, 10 proposals | Recall@0.5 (%): 51.8 -> 46.5 | Hurt |
| Use maximum CLIP similarity instead of the final objectness score (ablation) | ProposalCLIP on PASCAL VOC 2007 test · PASCAL VOC 2007 test, 10 proposals | Recall@0.5 (%): 51.8 -> 50.7 | Hurt |
| Use CLIP similarity entropy alone instead of the final objectness score (ablation) | ProposalCLIP on PASCAL VOC 2007 test · PASCAL VOC 2007 test, 10 proposals | Recall@0.5 (%): 46.5 -> 52.5 | Improved |
| Combine entropy, maximum similarity, and the initial score into the final objectness score (ablation) | ProposalCLIP on PASCAL VOC 2007 test · PASCAL VOC 2007 test, 10 proposals | Recall@0.5 (%): 52.5 -> 51.8 | Hurt |
Closest papers by meaning
- Object-Proposal Evaluation Protocol is 'Gameable' (CVPR 2016 · similarity 0.89)
- Relationship Proposal Networks (CVPR 2017 · similarity 0.86)
- Proposal-Free Network for Instance-Level Object Segmentation (TPAMI 2017 · similarity 0.86)
- DetCLIPv3: Towards Versatile Generative Open-vocabulary Object Detection (CVPR 2024 · similarity 0.85)
- Generative Region-Language Pretraining for Open-Ended Object Detection (CVPR 2024 · similarity 0.84)
- Just a Few Glances: Open-Set Visual Perception with Image Prompt Paradigm (AAAI 2025 · similarity 0.84)
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.
