Semantic segmentation model training method, device, computer equipment and storage medium

Abstract

The present application provides a training method, device, computer equipment and storage medium for a semantic segmentation model, wherein the method includes: constructing a training sample set, the training sample set includes objects of the first category and objects of the second category, wherein the first category The object is marked with a bounding box and a segmentation mask, and the second category object is marked with a bounding box; the training sample set is input into the deep network model for training, and the first bounding box parameters of the first category object are trained , the first mask parameter and the second bounding box parameter of the second category object; the first bounding box parameter and the first mask parameter are input into the weight transfer function for training, and the bounding box prediction mask parameter is trained; A semantic segmentation model is constructed based on the parameters trained above. This application greatly reduces the marking cost of the training sample set, reduces the pressure on subsequent training samples and training calculations, and improves training efficiency.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a training method, device, computer equipment and storage medium for a semantic segmentation model. Background technique [0002] Image semantic segmentation means that the machine automatically segments the image and recognizes the content in the image. For example, given a photo of a person riding a motorcycle, the motorcycle and the person are separated from the photo. When performing semantic segmentation on an image, a segmentation mask (segmentation mask) needs to be obtained to segment the image; at present, the segmentation mask cannot be derived from the bounding box. Therefore, during training, if the image is to be segmented, it is necessary to All instances are labeled with a segmentation mask. [0003] Existing semantic segmentation methods require that all training instances must be labeled with a segmentation mask, i.e., label every pixel, making annot...

AI Technical Summary

Problems solved by technology

[0004] The main purpose of this application is to provide a training method, device, computer equipment and storage medium for a semantic segmentation model, to overcome the high cost defect of training the semantic segmentation model

Images