Image segmentation method, device, equipment, storage medium

Abstract

The present application discloses an image segmentation method, device, device, and storage medium. The method includes: an encoder encodes the acquired image to be processed to obtain a multi-layer first feature map, and the multi-layer first feature map Including first features of different scales; the encoder includes at least two coding layers connected in sequence, each of the coding layers outputs a layer of first feature maps, and the input of each of the coding layers includes the output of the previous layer Or the image to be processed; the decoder performs decoding processing on the multi-layer first feature map to obtain a fusion feature map; wherein, the decoder includes at least two decoding layers connected in sequence, and each of the decoding layers uses The content attention operator and the spatial attention operator are implemented, and the input of each decoding layer includes the output result of the previous layer and the first feature map with the same scale; the classifier performs a pixel-by-pixel process on the fusion feature map Classify to obtain the image segmentation result of the image to be processed.

Description

technical field [0001] This application relates to computer vision technology, involving but not limited to an image segmentation method, device, equipment, and storage medium. Background technique [0002] Image segmentation mainly uses the color, intensity, texture and other information of image pixels to perform pixel-level semantic classification of images. In related technologies, image semantic segmentation is generally performed through the encoding and decoding framework in the deep convolutional network, and the image is input to the encoder Downsampling is performed to obtain a high-level feature map containing high-level features with low resolution but rich semantic information. Afterwards, the high-level feature map is input into the decoder and gradually restored to the resolution of the original image through multiple upsampling. Fusion to obtain image segmentation results. [0003] However, since the fusion of the high-level feature map and the low-level fe...

AI Technical Summary

Problems solved by technology

[0003] However, since the fusion of the high-level feature map and the low-level feature map by the deep convolutional network is only the sum of the transformed features in the receptive field, there may be a feature misalignment problem between the high-level features and the low-level features.

Images