Research map / Vision-language and multimodal / CLIP and zero-shot
CrowdCLIP: Unsupervised Crowd Counting via Vision-Language Model
Read the paperCheck 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 |
|---|---|---|---|
| replace CSS-CCNN with CrowdCLIP | unsupervised crowd counting · UCF-QNRF test | MAE: 437.0 -> 283.3; 35.2% | Helped |
| replace CSS-CCNN with CrowdCLIP | unsupervised crowd counting · ShanghaiTech Part A test | MAE: 197.3 -> 146.1; 26.0% | Helped |
| replace CSS-CCNN with CrowdCLIP | unsupervised crowd counting · ShanghaiTech Part B test | MAE | Helped |
| replace CSS-CCNN with CrowdCLIP | unsupervised crowd counting · JHU-Crowd++ test | MAE: 217.6 -> 213.7; 1.8% | Helped |
| replace CSS-CCNN with CrowdCLIP | unsupervised crowd counting · UCF-CC-50 | MAE: 564.9 -> 438.3; 22.4% | Helped |
| fine-tune the image encoder with ranking prompts (ablation) | CrowdCLIP · UCF-QNRF test | MAE: 414.7 -> 283.3 | Helped |
| fine-tune the text encoder instead of keeping it fixed (ablation) | CrowdCLIP · UCF-QNRF test | MAE: 414.7 -> 523.4 | Hurt |
| fine-tune both image and text encoders rather than keeping text fixed (ablation) | CrowdCLIP · UCF-QNRF test | MAE: 283.3 -> 416.6 | Hurt |
| add first-stage coarse patch filtering (ablation) | zero-shot CLIP · UCF-QNRF test | MAE: 528.7 -> 437.7 | Helped |
| add second-stage fine-grained head filtering (ablation) | zero-shot CLIP · UCF-QNRF test | MAE: 437.7 -> 414.7 | Helped |
| use the fine-tuned image encoder for final count mapping (ablation) | CrowdCLIP with first- and second-stage filtering · UCF-QNRF test | MAE: 414.7 -> 283.3 | Helped |
| set patch number P to 5 instead of 4 (ablation) | CrowdCLIP · UCF-QNRF test | MAE: 283.3 -> 305.2 | Hurt |
Closest papers by meaning
- Leveraging Unlabeled Data for Crowd Counting by Learning to Rank (CVPR 2018 · similarity 0.91)
- Almost Unsupervised Learning for Dense Crowd Counting (AAAI 2019 · similarity 0.90)
- Crowd Counting and Density Estimation by Trellis Encoder-Decoder Networks (CVPR 2019 · similarity 0.90)
- Switching Convolutional Neural Network for Crowd Counting (CVPR 2017 · similarity 0.89)
- Taste More, Taste Better: Diverse Data and Strong Model Boost Semi-Supervised Crowd Counting (CVPR 2025 · similarity 0.87)
- Free Lunch Enhancements for Multi-modal Crowd Counting (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.