Title: A Benchmark for Semi-supervised Multi-modal Crowd Counting

Abstract:

This study introduces the inaugural benchmark dedicated to semi-supervised multi-modal crowd counting. To establish a framework for this previously uncharted area, we define the specific parameters of the semi-supervised multi-modal environment alongside a standardized protocol that delineates the split between labeled and unlabeled data across varying labeling ratios. Furthermore, to create robust reference standards, we adapt a wide array of representative baseline models, incorporating both established fully supervised multi-modal techniques and semi-supervised single-modal approaches. We subsequently assess the performance of these models within the context of our newly proposed benchmark. The associated code and data partitions will be made publicly available at https://github.com/HenryCilence/Semi-supervised-Multimodal-Crowd-Counting.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC