Image processing method and training method of image processing network

Abstract

The present disclosure relates to the technical field of computer vision, in particular to an image processing method and a training method for an image processing network. A training method for an image processing network, comprising: obtaining a first sample set, the first sample set including a plurality of sample data; for each sample data in the first sample set, processing the sample data to obtain a global image information The first image of the first image and the first label corresponding to the first image, and at least one second image containing partial image information and the second label corresponding to the second image; the first sub-network to be trained based on the first image and the first label Training is performed, and the second sub-network to be trained is trained based on the second image, the output result output by the first sub-network and the second label. The implementation mode of the present disclosure improves the network joint training effect and reduces the risk of network effect degradation.

Description

technical field [0001] The present disclosure relates to the technical field of computer vision, and in particular to an image processing method and an image processing network training method, device, electronic equipment and storage medium. Background technique [0002] Image processing algorithms based on deep neural networks often include multiple tasks. Taking the image matting algorithm as an example, the image matting algorithm mainly includes two tasks of image semantic segmentation and image matting, and it is difficult to use an independent network to cover both tasks at the same time. Therefore, in the related art, the system uses two independent networks, that is, a semantic segmentation network and a matting network. [0003] When performing joint network training on multiple networks, because different networks pay attention to different image information, it is difficult for existing training methods to meet the training requirements of each network at the sa...

AI Technical Summary

Problems solved by technology

[0003] When performing joint network training on multiple networks, because different networks focus on different image information, it is difficult for existing training methods to meet the training requirements of each network at the same time, and it is easy for one of them to degenerate.

Images