Text supervision semantic segmentation method and device, electronic equipment and storage medium

• Main Authors: CAI KAIXIN, LIANG XIAODAN, REN PENGZHEN
• Format: Patent
• Language: Chinese; English
• Published: 28.11.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; PHYSICS

The invention discloses a text supervision semantic segmentation method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining a to-be-segmented target image; performing image encoding on the target image through the aggregation token by using an image encoder to obtain aggregation features; wherein the image encoder is generated by training based on a plurality of mixed images marked with semantic segmentation labels, the mixed images are obtained by randomly mixing patches of different original images with known segmentation masks, and the semantic segmentation label of each patch of the mixed images corresponds to the segmentation mask of the original image of each patch source; aggregation features correspond to semantic segmentation masks; and performing text classification on the aggregated features to obtain a segmentation category, and further obtaining a semantic segmentation result in combination with a semantic segmentation mask. According to the metho