Research map / Detection and segmentation / Open-set and few-shot detection
DenseCLIP: Language-Guided Dense Prediction With Context-Aware Prompting
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 |
|---|---|---|---|
| add pixel-text matching and post-model context-aware prompting | CLIP ResNet-50 with Semantic FPN · ADE20K val | mIoU (SS): 38.6 -> 43.5 | Helped |
| add pixel-text matching and post-model context-aware prompting | CLIP ResNet-101 with Semantic FPN · ADE20K val | mIoU (SS): +4.7% | Helped |
| add pixel-text matching and post-model context-aware prompting | CLIP ViT-B with Semantic FPN · ADE20K val | mIoU (SS): +2.3% | Helped |
| add pixel-text matching and language guidance | RetinaNet with ResNet-50 CLIP · COCO val2017 | AP: +0.9% | Helped |
| add pixel-text matching and language guidance | Mask R-CNN with ResNet-50 CLIP · COCO val2017 | APᵐ: +0.8% | Helped |
| add learnable textual contexts (ablation) | CLIP ResNet-50 dense prediction · ADE20K val | mIoU: 39.6 -> 42.1 | Helped |
| add vision-to-language pre-model prompting (ablation) | CLIP ResNet-50 with language-domain prompting · ADE20K val | mIoU: 42.1 -> 42.9 | Helped |
| use post-model rather than pre-model vision-to-language prompting (ablation) | CLIP ResNet-50 with language-domain prompting · ADE20K val | mIoU: 42.9 -> 43.5 | Helped |
| set image encoder learning-rate multiplier to 0.1 rather than 1.0 (ablation) | CLIP ResNet-50 · ADE20K val | mIoU: 42.6 -> 43.5 | Helped |
| freeze the text encoder rather than use learning-rate multiplier 0.1 (ablation) | CLIP ResNet-50 · ADE20K val | mIoU: 42.2 -> 43.5 | Helped |
| optimize textual contexts with CoOp rather than use the fixed human-defined template (ablation) | CLIP ResNet-50 · ADE20K val | mIoU: 42.9 -> 43.5 | Helped |
| make γ learnable, initialized to 10^-4 (ablation) | post-model prompting · ADE20K val | mIoU: 42.6 -> 43.5 | Helped |
Closest papers by meaning
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.88)
- Demystifying CLIP Data (ICLR 2024 · similarity 0.88)
- DeCLIP: Decoupled Learning for Open-Vocabulary Dense Perception (CVPR 2025 · similarity 0.88)
- ResCLIP: Residual Attention for Training-free Dense Vision-language Inference (CVPR 2025 · similarity 0.88)
- TIPS: Text-Image Pretraining with Spatial awareness (ICLR 2025 · similarity 0.88)
- FineCLIP: Self-distilled Region-based CLIP for Better Fine-grained Understanding (NeurIPS 2024 · similarity 0.87)
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.