Domain-General Crowd Counting in Unseen Scenarios

TL;DR

This paper proposes a novel domain generalization approach for crowd counting that trains on a single source domain and generalizes to unseen domains using dynamic sub-domain division and feature disentanglement.

Contribution

It introduces a dynamic sub-domain division scheme and domain-invariant/-specific feature disentanglement framework for crowd counting, addressing the domain shift problem.

Findings

Strong generalization on multiple benchmarks

Effective disentanglement of domain-invariant and -specific features

Outperforms existing domain adaptation methods

Abstract

Domain shift across crowd data severely hinders crowd counting models to generalize to unseen scenarios. Although domain adaptive crowd counting approaches close this gap to a certain extent, they are still dependent on the target domain data to adapt (e.g. finetune) their models to the specific domain. In this paper, we aim to train a model based on a single source domain which can generalize well on any unseen domain. This falls into the realm of domain generalization that remains unexplored in crowd counting. We first introduce a dynamic sub-domain division scheme which divides the source domain into multiple sub-domains such that we can initiate a meta-learning framework for domain generalization. The sub-domain division is dynamically refined during the meta-learning. Next, in order to disentangle domain-invariant information from domain-specific information in image features, we design the domain-invariant and -specific crowd memory modules to re-encode image features. Two types of losses, i.e. feature reconstruction and orthogonal losses, are devised to enable this disentanglement. Extensive experiments on several standard crowd counting benchmarks i.e. SHA, SHB, QNRF, and NWPU, show the strong generalizability of our method.