Image segmentation method, model training method, system, device and storage medium

Abstract

This invention discloses an image segmentation method, a model training method, a system, an apparatus, and a storage medium. The model training method includes: acquiring a prior probability map of a training image set and adding the prior probability map to the training image set to obtain input data; constructing the image segmentation model based on a convolutional neural network; a self-attention module being used to assign weights to the feature maps of the input data during the sampling process of the training data by the sampling convolution module; and inputting the input data into the image segmentation model for model training to obtain a target image segmentation model. This model training method, by utilizing prior knowledge to guide image segmentation, makes the image segmentation model focus more on the segmented region of the image to be segmented. Assigning weights to the feature maps of the input data during the sampling process helps the model better distinguish between background and foreground, while reducing the impact of noise on segmentation accuracy.

Images