Training method, training device, server and storage medium for semantic segmentation network

Abstract

The embodiment of the present invention relates to the field of computer vision, and discloses a training method, training device, server and storage medium of a semantic segmentation network. The training method of the semantic segmentation network in the present invention comprises: obtaining the real image and the simulated image corresponding to the real image, the simulated image is determined by the image generation network according to the feature image of the real image and the first semantic segmentation image, the first semantic segmentation image is The first semantic segmentation network is determined according to the real image, and the image generation network is obtained according to the feature image of the real training image, the semantic segmentation training image of the real training image, and the training of the real training image; according to the difference information between the real image and the simulated image, iteratively Adjust the network parameters of the first semantic segmentation network to obtain the target semantic segmentation network. In this embodiment, the semantic segmentation network can accurately perform semantic segmentation on the image, and the prediction accuracy of the semantic segmentation network is improved.

Description

technical field [0001] The embodiments of the present invention relate to the field of computer vision, and in particular to a training method, training device, server and storage medium of a semantic segmentation network. Background technique [0002] Semantic segmentation technology is an understanding of the pixel level of the image, and it is to classify the objects on the image at the pixel level, that is, to classify the pixels belonging to the same type of object into one category, and use the specified label (label) to mark. At present, this technology is widely used in medical image analysis, unmanned driving, geographic information system, robotics and other fields. In the medical field, it is mainly used for tumor image segmentation, dental caries diagnosis, etc. In the field of unmanned driving, semantic segmentation technology plays an important role as its core technology. Segmenting the road can determine the driving area of ​​the car; accurate segmentation o...

AI Technical Summary

Problems solved by technology

[0003] The inventors found that there are at least the following problems in related technologies: the training of the semantic segmentation network depends on the image data used for training, and the training image data is an image with image annotation information. The more training image data, the more difficult the semantic segmentation network obtained after training The semantic segmentation of the image is more accurate, but the training image data currently used is obtained by manual annotation, and the training image data of manual annotation is very limited, which leads to the semantic segmentation of the image by the semantic segmentation network obtained by training the labeled image data. Inaccurate

Images