High-density Image Object Counting Network Based on Spatial Context and Channel Attention: Take crowds and vehicles in a statistically dense scene as an example

Accurate object counting is a challenging task in image analysis, low-density image object counting can usually be achieved by object detection algorithms, and high-density object counting still has limited counting accuracy. We propose a high-density image object counting network based on spatial context and channel attention, abbreviated as HIOC-Net, which divides low-level features into multiple blocks of different scales through a spatial context-aware module to extract rich contextual features, and then uses the channel attention-aware module to process the interdependence of feature information in the channel dimension, so that the model can focus on useful features, suppressing the irrelevant background. This paper conducts extensive experiments on large-scale crowd and vehicle counting datasets, including ShanghaiTech, UCF_CC_50, WorldExpo'10, TRANCOS, and HBR_YD datasets. The results show that our method not only surpasses many state-of-the-art methods in counting accuracy but also achieves competitive localization accuracy, resulting in high-quality object density maps.