Research Article
Local-Channel Large Separable Kernel Attention
@INPROCEEDINGS{10.1007/978-3-031-96146-5_20, author={Zengjie Deng and Tao Lei and Haixia Cui}, title={Local-Channel Large Separable Kernel Attention}, proceedings={Smart Grid and Innovative Frontiers in Telecommunications. 8th EAI International Conference, EAI SmartGIFT 2024a, Santa Clara, United States, March 23-24, 2024, Proceedings}, proceedings_a={SMARTGIFT}, year={2026}, month={9}, keywords={CNN local-channel information Large Kernel Attention (LKA)}, doi={10.1007/978-3-031-96146-5_20} }- Zengjie Deng
Tao Lei
Haixia Cui
Year: 2026
Local-Channel Large Separable Kernel Attention
SMARTGIFT
Springer
DOI: 10.1007/978-3-031-96146-5_20
Abstract
Large Kernel Attention (LKA) has demonstrated superior performance in image classification/segmentation and surpassing transformer-based visual tasks. However, using 1 1 Conv for full-channel interaction in Large Kernel Separable Attention (LSKA) results in quadratic growth in computational complexity and memory usage as the number of channels increases. To alleviate this issue, this study proposes a Local-channel Large Separable Kernel Attention (LLSKA) module where the group convolution is employed to focus module solely on local-channel information to reduce the model complexity while maintaining system performance. The number of groups is determined by an adaptive strategy to determine the local-channel receptive ranges.