Weak supervision semantic segmentation method based on attention directing inference network

Abstract

The invention discloses a weak supervision semantic segmentation method based on an attention directing inference network. The main content comprises the self-directing on the network attention and the supervision on the additional integration, wherein the supervision process is as follows: the attention directing inference network has two network streams of the classification stream and the attention mining; the classification stream is conductive to identifying the region of the class, the attention mining ensures that all regions possibly conductive to the classification decision making canbe included into the attention of the network, so that an attention graph becomes more complete and accurate, the attention graph can be jointly generated and trained through two loss functions; theextension of the attention directing inference network is imported to seamlessly integrate the additional supervision in the weak-supervision learning framework, thereby controlling the attention graph learning process. Based on the end-to-end framework, the supervision of the specific task can be directly applied to the attention graph at the training stage, and the difference between the weak supervision and the additional supervision can be reduced, and the generalization performance is improved.

Description

technical field [0001] The invention relates to the field of semantic segmentation, in particular to a weakly supervised semantic segmentation method based on a guided attention reasoning network. Background technique [0002] With the popularization of the Internet and the rapid development of multimedia technology, multimedia information represented by images is showing an explosive growth trend, which brings great challenges to the storage, management and retrieval of images. Therefore, how to effectively identify, classify and manage massive image data has become an urgent problem to be solved. Image semantic segmentation is a key link in image processing and analysis, and it is also a classic research branch in the field of computer vision. Through the image semantic segmentation technology, the main target in the image can be segmented and identified, so as to realize the processing and analysis of image information. In the medical field, the computer can automatical...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem of fuzzy and inaccurate target segmentation, the object of the present invention is to provide a weakly supervised semantic segmentation method based on guided attention reasoning network, which has two network streams of classification flow and attention mining, classification flow Areas that help to identify classes, attention mining ensures that all areas that may be helpful for classification decisions will be included in the attention of the network, making the attention map more complete and accurate, through these two loss functions can be combined Generate and train attention maps; introduce an extension of the guided attention inference network to seamlessly integrate additional supervision in a weakly supervised learning framework to control the attention map learning process

Images