CrowdFormer: Weakly-supervised crowd counting with improved generalizability

Convolutional neural networks (CNNs) have dominated the field of computer vision for nearly a decade. However, due to their limited receptive field, CNNs fail model global context. On other hand, transformers, an attention-based architecture, can context easily. Despite this, there are studies that investigate effectiveness transformers in crowd counting. In addition, majority existing crowd-counting methods based on regression density maps which requires point-level annotation each person present scene. This task is laborious and also error-prone. has led increased focus weakly-supervised methods, require only count-level annotations. this paper, we propose method counting using pyramid transformer. We conducted extensive evaluations validate proposed method. Our achieves state-of-the-art performance. More importantly, it shows remarkable generalizability.