Semantic edge detection method, device, system and medium based on joint learning

Abstract

The invention discloses a semantic edge detection method, device, system, and storage medium based on joint learning. The method includes: pre-constructing a generative confrontation network and an edge detection network; acquiring edge noise samples, and generating a confrontation network according to the edge noise samples Carry out the joint learning training of noise learning and edge detection with the edge detection network, when the edge detection result reaches the expected value, the generation confrontation network and the edge detection network that complete the joint learning training are obtained; the image to be detected is input to the edge that completes the joint learning training After the detection network outputs the semantic edge detection results. In the embodiment of the present invention, by constructing a generative confrontation network and an edge detection network in a unified framework for joint learning, and simultaneously performing noise learning tasks and edge detection tasks, a closed-loop training structure is formed, which can effectively improve the performance of each task, so that after joint learning, The edge detection network has a more precise and clear semantic edge detection effect.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a semantic edge detection method, device, system and storage medium based on joint learning. Background technique [0002] Image edges are important clues to identify an object, such as in the absence of a large part of the boundary, it is still possible to identify objects only from sketches, and edges have also proved to be useful for 3D reconstruction, localization and image generation. [0003] However, accurate labeling of image boundaries is currently very slow, otherwise the accuracy requirements will not be met, therefore, most existing datasets contain significant labeling noise, which may be the root cause of most detectors outputting thick boundary predictions , while thick boundary predictions are undesirable for downstream tasks. [0004] Therefore, the prior art still needs to be improved and developed. Contents of the invention [0005] In view...

AI Technical Summary

Problems solved by technology

[0005] In view of the above deficiencies in the prior art, the purpose of the present invention is to provide a semantic edge detection method, device, system and storage medium based on joint learning, aiming to solve the problem of excessive semantic edge prediction in the prior art

Images