SEMANTIC SEGMENTATION MODEL TRAINING METHOD AND APPARATUS, AND IMAGE SEMANTIC SEGMENTATION METHOD AND APPARATUS

摘要

A semantic segmentation model training method, an image semantic segmentation method, a semantic segmentation model training apparatus, an image semantic segmentation apparatus, an electronic device, and a computer-readable storage medium. The semantic segmentation model training method comprises: acquiring a training set, the training set comprising multiple images as well as annotation information corresponding thereto, and the annotation information corresponding to any image comprising a segmentation box annotation and/or a semantic segmentation annotation (S11); extracting features of the images, and obtaining feature data of the images (S12); on the basis of the feature data, obtaining first segmentation frame information and first semantic segmentation information (S13); on the basis of the feature data, the first segmentation frame information, and the first semantic segmentation information, obtaining second segmentation frame information and second semantic segmentation information of the images (S14); determining a loss value on the basis of the second segmentation frame information and annotation information and/or on the basis of the second semantic segmentation information and annotation information (S15); and on the basis of the loss value, adjusting parameters of a semantic segmentation model (S16). In the semantic segmentation model training method, training data can be conveniently acquired, thus the training effect can be improved by means of massive, high-quality data.