# Counting Crowds in Bad Weather

**Authors:** Zhi-Kai Huang, Wei-Ting Chen, Yuan-Chun Chiang, Sy-Yen Kuo, Ming-Hsuan Yang  
**arXiv:** [2306.01209](https://arxiv.org/abs/2306.01209)  
**Date:** 2023-06-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2306.01209v1/

## Key points

- Computer vision has seen a surge in interest in crowd counting due to its wide range of applications
- Existing methods struggle to perform well under adverse weather conditions such as haze, rain, and snow
- The authors propose a method for robust crowd counting in adverse weather scenarios
- Their model learns effective features and adaptive queries to account for large appearance variations caused by adverse weather conditions
- The proposed model incorporates weather queries to learn the weather information based on the degradation of the input image while optimizing with the crowd counting module
- Experimental results demonstrate the effectiveness of the proposed algorithm in accurately counting crowds under different types of adverse weather conditions on benchmark datasets
- The authors plan to make the source code and trained models publicly available for further research and application.

## Summary

The field of computer vision has seen a surge in interest in crowd counting due to its wide range of applications in image understanding. While numerous methods have been proposed and achieved state-of-the-art performance for real-world tasks, they often struggle to perform well under adverse weather conditions such as haze, rain, and snow. This is because the visual appearances of crowds in these scenes are drastically different from those in clear weather typically found in datasets. To address this limitation, the authors propose a method for robust crowd counting in adverse weather scenarios. Unlike existing approaches that use a two-stage approach involving image restoration and crowd counting modules, their model learns effective features and adaptive queries to account for large appearance variations caused by adverse weather conditions. The proposed model incorporates weather queries that allow it to learn the weather information based on the degradation of the input image while simultaneously optimizing with the crowd counting module. Experimental results demonstrate that the proposed algorithm is effective in accurately counting crowds under different types of adverse weather conditions on benchmark datasets. To facilitate further research and application, the authors plan to make the source code and trained models publicly available. Overall, this paper presents an innovative approach to address the challenges of crowd counting in adverse weather scenarios by leveraging adaptive queries and learning effective features. The results highlight its effectiveness and potential impact on various real-world applications requiring accurate crowd counting under diverse environmental conditions.