Image semantic segmentation model training method and device, image semantic segmentation method and device and storage medium

Abstract

The invention provides an image semantic segmentation model training method and device, an image semantic segmentation method and device, and a storage medium, and relates to the technical field of computers, and the method comprises the steps: carrying out the judgment of a semantic segmentation image generated by a semantic segmentation model through employing a discriminator model; constructinga loss function corresponding to the discriminator model, wherein the loss function comprises a target domain loss function generated based on the target domain image, the target domain loss functioncomprises at least one of a first semantic loss function generated based on semantic consistency of the image blocks, a second semantic loss function generated based on semantic consistency of the clustering clusters and a third semantic loss function generated based on image space logic construction. According to the method and device and the storage medium, the semantic segmentation model reasoning result of the model on the target domain image is constrained in the form of the regularization item in the training process, cross-domain migration is performed on the image semantic segmentation model, and the efficiency and accuracy of image semantic segmentation model training are improved.

Description

technical field [0001] The present disclosure relates to the technical field of image processing, and in particular to an image semantic segmentation model training, an image semantic segmentation method, device and storage medium. Background technique [0002] The goal of image semantic segmentation is to discern the semantic category of each pixel in the image. A neural network can be used to build a deep semantic segmentation model. The training of a deep semantic segmentation model requires a large amount of training data with pixel-level annotations. However, obtaining these training data is very expensive and slow. Currently, computer-synthesized images are used in model training, but there is a large difference between computer-synthesized images and real images, which leads to poor performance of semantic segmentation models trained on synthetic images on real images, that is, The phenomenon of "domain mismatch". Contents of the invention [0003] In view of this...

AI Technical Summary

Problems solved by technology

A neural network can be used to build a deep semantic segmentation model. The training of a deep semantic segmentation model requires a large amount of training data with pixel-level annotations. However, obtaining these training data is very expensive and slow

Currently, computer-synthesized images are used in model training, but there is a large difference between computer-synthesized images and real images, which leads to poor performance of semantic segmentation models trained on synthetic images on real images, that is, The phenomenon of "domain mismatch"

Images