KAConv: Kernel attention convolutions

Most of the previous network optimization works applied attention mechanism to feature maps, but neglected to embed attention into convolution kernel of the end-to-end network that is convenient to deploy. To address this issue, we present a novel attention convolution method named Kernel Attention Convolution (KAConv) to enhance the flexibility of convolution. The proposed KAConv gener-ates different attention weights for different spatial positions of convolution kernels based on the input features, so as to dynamically adjust the parameters of convolution kernels during the forward propaga-tion to enhance the flexibility of convolution. We decompose the convolution kernels into subkernels spatially, and generate the corresponding feature maps through which attention weights are obtained. The final refined feature maps are aggregated by the attention weighted feature maps corresponding to each subkernel. KAConv is a computationally lightweight convolution method, which not only incor-porates attention into kernels but also enhances informative representations. By replacing the standard convolution with the proposed KAConv in convolutional neural networks (CNNs), the networks yield sig-nificant performance improvement. Extensive experiments on the ImageNet-1K benchmark demonstrate that KAConv outperforms existing attention mechanism-based methods. We also carry out experiments on the MS COCO and PASCAL VOC datasets to show the generalization ability of our method.(c) 2022 Elsevier B.V. All rights reserved.

Automated summary

The study introduces a novel attention convolution method called Kernel Attention Convolution (KAConv) to enhance the flexibility of convolution. By embedding attention into the convolution kernel, KAConv generates different attention weights to dynamically adjust the parameters of convolution kernels, improving the flexibility of convolution. Experiment results demonstrate that KAConv outperforms existing attention mechanism-based methods in the ImageNet-1K benchmark.