Efficient Multi-Scale Attention Module with Cross-Spatial Learning

Remarkable effectiveness of the channel or spatial attention mechanisms for producing more discernible feature representation are illustrated in various computer vision tasks. However, modeling the cross-channel relationships with channel dimensionality reduction may bring side effect in extracting deep visual representations. In this paper, a novel efficient multi-scale attention (EMA) module is proposed. Focusing on retaining the information on per channel and decreasing the computational overhead, EMA groups the channel dimensions into multiple sub-features and makes the spatial semantic features well-distributed inside each feature group. Specifically, apart from encoding the global information to re-calibrate the channel-wise weight in each parallel branch, the output features of the two parallel branches are further aggregated by a cross-dimension interaction method. The extensive experiments on common-used benchmarks, such as CIFAR100 for image classification, and object detection on MS COCO and VisDrone2019 datasets, are conducted which indicate that EMA outperforms several recent attention mechanisms significantly without changing networks depth.