Learning Error-Driven Curriculum for Crowd Counting

TL;DR

This paper introduces an error-driven curriculum learning approach for crowd counting that employs a tutor network to focus training on hard examples, significantly enhancing density map regression performance.

Contribution

A novel learning strategy using a tutor network to generate pixel-level weights, improving crowd counting accuracy by emphasizing difficult examples during training.

Findings

Achieved state-of-the-art results on benchmark datasets.

Effectively addresses pixel value imbalance in density maps.

Improves model focus on hard-to-predict regions.

Abstract

Density regression has been widely employed in crowd counting. However, the frequency imbalance of pixel values in the density map is still an obstacle to improve the performance. In this paper, we propose a novel learning strategy for learning error-driven curriculum, which uses an additional network to supervise the training of the main network. A tutoring network called TutorNet is proposed to repetitively indicate the critical errors of the main network. TutorNet generates pixel-level weights to formulate the curriculum for the main network during training, so that the main network will assign a higher weight to those hard examples than easy examples. Furthermore, we scale the density map by a factor to enlarge the distance among inter-examples, which is well known to improve the performance. Extensive experiments on two challenging benchmark datasets show that our method has achieved state-of-the-art performance.