Research map / Detection and segmentation / Open-set and few-shot detection
RegionCLIP: Region-Based Language-Image Pretraining
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| pretrain with region-text alignment instead of image-level CLIP alignment | RegionCLIP · COCO open-vocabulary (Novel) | AP50: 31.4 vs. 14.2 | Helped |
| add image-text pairs to region-text pretraining (ablation) | RegionCLIP · COCO open-vocabulary (Novel) | AP50: 21.4 -> 26.8 | Helped |
| replace randomly sampled regions with RPN proposals (ablation) | RegionCLIP pretraining · COCO zero-shot (GT boxes) | AP50 All: 60.8 -> 62.8 | Helped |
| replace randomly sampled regions with RPN proposals (ablation) | RegionCLIP pretraining · COCO open-vocabulary (Generalized All) | AP50: 46.9 vs. 47.5 | No clear effect |
| add region-level contrastive loss to distillation loss (ablation) | RegionCLIP pretraining · COCO open-vocabulary (Novel) | AP50: 24.1 -> 26.8 | Helped |
| add distillation loss to contrastive loss (ablation) | RegionCLIP pretraining · COCO open-vocabulary (Generalized All) | AP50: 45.8 -> 47.5 | Helped |
| replace RN50 teacher with RN50x4 teacher (ablation) | RegionCLIP · COCO zero-shot (GT boxes) | AP50 All: 61.4 -> 65.6 | Helped |
| replace RN50 student with RN50x4 student (ablation) | RegionCLIP · COCO open-vocabulary (Generalized All) | AP50: 50.4 -> 55.7 | Helped |
| replace RN50 student with RN50x4 student (ablation) | RegionCLIP · COCO zero-shot (RPN proposals) | AP50 All: 29.3 -> 29.6 | No clear effect |
| add focal scaling to detector training (ablation) | RegionCLIP COCO detector · COCO open-vocabulary (Novel) | AP50: 22.6 -> 31.4 | Helped |
| add focal scaling to detector training (ablation) | RegionCLIP COCO detector · COCO open-vocabulary (Base) | AP50: 58.5 -> 57.1 | Hurt |
| pretrain on CC3M rather than COCO Cap (ablation) | RegionCLIP · COCO open-vocabulary (Generalized All) | AP50: 47.5 -> 50.4 | Helped |
Closest papers by meaning
- FineCLIP: Self-distilled Region-based CLIP for Better Fine-grained Understanding (NeurIPS 2024 · similarity 0.90)
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.90)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.90)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.90)
- CLIM: Contrastive Language-Image Mosaic for Region Representation (AAAI 2024 · similarity 0.89)
- Contrastive Localized Language-Image Pre-Training (ICML 2025 · similarity 0.89)
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.