Cross-domain semantic segmentation method based on graph convolution and contrast learning

Abstract

The invention provides a cross-domain semantic segmentation method based on graph convolution and contrast learning, and belongs to the field of transfer learning and computer vision. The invention designs a method for solving an adjacent matrix in different domains, and provides a new thought for establishing a long-distance context relationship between the domains. In order to solve the problem of unbalanced distribution of different categories, learning loss is grouped and compared. Secondly, in order to extract domain-invariant information, a newly proposed double-domain adjacent matrix is utilized to perform graph convolution operation; in the process of graph convolution operation, a graph structure is constructed on a feature graph, and in order to establish a long-distance context relationship between domains, the graph convolution operation is creatively completed in the proposed graph structure by using the proposed double-domain adjacent matrix. According to the method provided by the invention, a long-distance context relationship between domains is established, domain-invariant information can be extracted more effectively, and a better evaluation result is obtained in subjective and objective evaluation.

Description

technical field [0001] The invention designs a cross-domain semantic segmentation method based on graph convolution and comparative learning, which belongs to the field of pattern recognition and computer vision. Background technique [0002] The basic task of semantic segmentation is to correctly classify each pixel in the image. Since semantic segmentation is a classification task at the pixel level, it is very helpful for scene understanding tasks, so semantic segmentation tasks have broad development prospects in tasks such as autonomous driving, medical image segmentation, and scene recognition. [0003] Since the advent of fully convolutional networks, convolutional neural networks have made great progress in the field of semantic segmentation. However, since the semantic segmentation task requires pixel-level labels during the training process, and labeling pixel-level labels requires a lot of manpower and material resources, people consider using synthetic datasets ...

AI Technical Summary

Problems solved by technology

[0005] The present invention provides a cross-domain semantic segmentation method based on graph convolution and comparative learning, which is used to solve the problem of unbalanced distribution of different categories in the existing segmentation methods, the distribution of categories with a small proportion is difficult to migrate, and the cross-domain effect is relatively poor problem, the present invention realizes the goal of transfer learning

Images