Local Information Matters: A Rethink of Crowd Counting

TL;DR

This paper introduces LIMM, a crowd counting model emphasizing local information through window partitioning and contrastive learning, achieving state-of-the-art results by focusing on local density distinctions.

Contribution

The paper proposes a novel crowd counting approach that emphasizes local modeling via window partitioning and contrastive learning, improving accuracy without sacrificing large object detection.

Findings

Significant MAE reduction on JHU-Crowd++ high-density subset (8.7%).

Enhanced local density discrimination capability.

State-of-the-art crowd counting performance.

Abstract

The motivation of this paper originates from rethinking an essential characteristic of crowd counting: individuals (heads of humans) in the crowd counting task typically occupy a very small portion of the image. This characteristic has never been the focus of existing works: they typically use the same backbone as other visual tasks and pursue a large receptive field. This drives us to propose a new model design principle of crowd counting: emphasizing local modeling capability of the model. We follow the principle and design a crowd counting model named Local Information Matters Model (LIMM). The main innovation lies in two strategies: a window partitioning design that applies grid windows to the model input, and a window-wise contrastive learning design to enhance the model's ability to distinguish between local density levels. Moreover, a global attention module is applied to the end of the model to handle the occasionally occurring large-sized individuals. Extensive experiments on multiple public datasets illustrate that the proposed model shows a significant improvement in local modeling capability (8.7\% in MAE on the JHU-Crowd++ high-density subset for example), without compromising its ability to count large-sized ones, which achieves state-of-the-art performance. Code is available at: https://github.com/tianhangpan/LIMM.