Abstract

Most of the current attention algorithms focus on one-dimensional channel attention or two-dimensional positional attention, and the processed images are three-dimensional, so these attention modules often cannot focus on all the areas that need attention, resulting in some key information missing. The three-dimensional attention module is designed in this paper. it can obtain a three-dimensional image attention weight matrix by combining one-dimensional channel attention and two-dimensional position attention module, and can obtain a new image with attention allocation after calculation. this paper uses deep learning technology, combines the channel attention module and the position attention module, and designs a three-dimensional attention module. The three-dimensional attention module has good results in a variety of visual tasks. Compared with SENet, in Cifar100 dataset, ResNet50 as the main network added attention has a 1.12% improvement. On the PKU VehicleID dataset, it has an average 2% improvement over SENet.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call