Cross-head Supervision for Crowd Counting with Noisy Annotations

TL;DR

This paper introduces CHS-Net, a crowd counting model with cross-head supervision using a convolution and transformer head to mitigate noisy annotations, improving accuracy on standard datasets.

Contribution

The paper proposes a novel cross-head supervision mechanism with a progressive learning strategy to enhance crowd counting under noisy annotation conditions.

Findings

Outperforms state-of-the-art methods on ShanghaiTech and QNRF datasets.

Effectively mitigates the impact of noisy annotations in crowd counting.

Demonstrates robustness with a combined convolution and transformer architecture.

Abstract

Noisy annotations such as missing annotations and location shifts often exist in crowd counting datasets due to multi-scale head sizes, high occlusion, etc. These noisy annotations severely affect the model training, especially for density map-based methods. To alleviate the negative impact of noisy annotations, we propose a novel crowd counting model with one convolution head and one transformer head, in which these two heads can supervise each other in noisy areas, called Cross-Head Supervision. The resultant model, CHS-Net, can synergize different types of inductive biases for better counting. In addition, we develop a progressive cross-head supervision learning strategy to stabilize the training process and provide more reliable supervision. Extensive experimental results on ShanghaiTech and QNRF datasets demonstrate superior performance over state-of-the-art methods. Code is available at https://github.com/RaccoonDML/CHSNet.