Semi-supervised crowd counting from unlabeled data

doi:10.1016/j.patcog.2025.112787

Semi-supervised crowd counting from unlabeled data

2025-11-21

0

PRE

AI

DOI：10.1016/j.patcog.2025.112787

Original

Original request for help

Save

Abstract

En 中文

• We present S4Crowd, a novel semi-supervised crowd counting framework incorporating variation modeling, pseudo labeling, and dynamic weighting for robust performance with limited annotations. • Two self-supervised regularization terms-Crowd Scale Equivariance (CSE) and Crowd Entropy Consistency (CEC)-enable unsupervised modeling of crowd variations. • Additionally, the Gated-Crowd-Recurrent-Unit (GCRU) encodes high-order crowd sequences and captures second-order statistics. • Extensive experiments demonstrate that out method consistently surpasses existing semi-supervised methods.

Semi-supervised crowd counting from unlabeled data

Abstract

Journal

Organization