Self-supervised Domain Adaptation in Crowd Counting

Abstract

Self-training crowd counting has not been attentively explored though it is one of the important challenges in computer vision. In practice, the fully supervised methods usually require an intensive resource of manual annotation. In order to address this challenge, this work introduces a new approach to utilize existing datasets with ground truth to produce more robust predictions on unlabeled datasets, named domain adaptation, in crowd counting. While the network is trained with labeled data, samples without labels from the target domain are also added to the training process. In this process, the entropy map is computed and minimized in addition to the adversarial training process designed in parallel. Experiments on Shanghaitech, UCF_CC_50, and UCF-QNRF datasets prove a more generalized improvement of our method over the other state-of-the-arts in the cross-domain setting.

Video- and Location-based Analysis of Cycling Routes for Safety Measures and Fan Engagement

“…For high density crowds, it is preferred to use the density map method to tackle the loss of information due to occlusion. A few examples are multi-scale models [35] [29], which use multi-column or pyramid-like networks to extract information on multiple scales (e.g., each column has a different kernel size), context-aware models [19], which add context information to the images, and domain adaptation models [21], which use synthetic images as datasets. Combination method both use regression and detection methods into a single model.…”

“…• Footage taken with a GoPro inside a car during the Druivenkoers Overijse 2021; • Footage taken with an earlier version of our proposed device during Kuurne-Brussel Kuurne 2022 and E3 Saxo Bank Classic 2022; • Several YouTube videos of the Tour de France [21], Ronde van Vlaanderen [31], Paris Roubaix [36] and the Vuelta [26].…”