Spatial and Channel Similarity-Aware Attention-Enhanced Network for Object Counting

Precise object counting is crucial in practical applications, finding extensive utility across numerous societal domains. In the context of few-shot object counting, variations in object angles can significantly alter the distribution and distinguishability of object feature points, thereby increasing the difficulty of feature extraction. To address these challenges, a spatial and channel similarity-aware attention-enhancement network for few-shot scenarios is introduced. The network employs slice convolution and attention mechanisms within the spatial and channel dimensions, respectively. Through such means, it effectively extracts object features from diverse angles, adeptly surmounting the challenges in feature extraction engendered by angular diversity. Specifically, the network introduces a spatial similarity-aware module, which performs multi-angle matching of support and query features in the spatial dimension to enhance the model’s feature extraction capabilities. Additionally, a channel attention enhancement module is incorporated to weight and enhance critical channel features, improving the network’s ability to handle diverse object angles. Comprehensive experiments carried out on TRANCOS, CARPK, PUCPR+ and UCSD datasets demonstrate that the proposed network exhibits excellent counting performance and robust generalization capabilities. It offers an effective approach for precise object counting tasks.