Research map / Vision-language and multimodal / CLIP and zero-shot
Unbiased Region-Language Alignment for Open-Vocabulary Dense Prediction
Yunheng Li, Yuxuan Li, Quan-Sheng Zeng, Wenhai Wang, Qibin Hou, Ming-Ming Cheng
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 |
|---|---|---|---|
| use DenseVLM with region proposals instead of CLIPSelf pseudo region-text pairs (stuff masks) (ablation) | region-language alignment fine-tuning · COCO panoptic stuff masks | Top1 accuracy: 35.7 -> 45.9 | Improved |
| add region proposals to CLIPSelf (ablation) | CLIPSelf · COCO panoptic stuff masks | Top1 accuracy: 41.7 -> 35.7 | Hurt |
| add region denoising threshold 0.3 (vs 0.0) (ablation) | DenseVLM · COCO panoptic boxes | Top1 accuracy: 72.1 -> 73.4 | Improved |
| raise denoising threshold from 0.3 to 0.6 (ablation) | DenseVLM · COCO panoptic boxes | Top1 accuracy: 73.4 -> 73.1 | Hurt |
| increase input resolution from 224 to 1024 (ablation) | DenseVLM · COCO panoptic boxes | Top1 accuracy: 60.1 -> 76.6 | Improved |
| increase input resolution from 512 to 1024 (ablation) | DenseVLM · COCO panoptic boxes | Top1 accuracy: 73.4 -> 76.6 | Improved |
| enlarge category set from 133 to 316 categories (ablation) | DenseVLM · COCO panoptic boxes | Top1 accuracy: 71.1 -> 73.4 | Improved |
| DenseVLM versus CLIPSelf on ViT-L/14 (ablation) | OpenCLIP-based region alignment · COCO panoptic boxes | Top1 accuracy: 68.3 -> 75.2 | Improved |
| DenseVLM versus CLIPSelf on ViT-B/16 (ablation) | region alignment · ADE20K panoptic boxes | Top1 accuracy: 43.4 -> 51.3 | Improved |
Closest papers by meaning
- CLIPSelf: Vision Transformer Distills Itself for Open-Vocabulary Dense Prediction (ICLR 2024 · similarity 0.90)
- LLMs Meet VLMs: Boost Open Vocabulary Object Detection with Fine-grained Descriptors (ICLR 2024 · similarity 0.90)
- Locality Alignment Improves Vision-Language Models (ICLR 2025 · similarity 0.89)
- DenseMLLM: Standard Multimodal LLMs for Dense Prediction (ICML 2026 · similarity 0.87)
- CogVLM: Visual Expert for Pretrained Language Models (NeurIPS 2024 · similarity 0.87)
- ROD-MLLM: Towards More Reliable Object Detection in Multimodal Large Language Models (CVPR 2025 · 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.
