Semantic segmentation method, network, equipment and computer storage medium

Abstract

The invention discloses a semantic segmentation method, network, equipment and computer storage medium, comprising: extracting deep features of an input image, performing rough segmentation based on the deep features, and obtaining rough segmentation results; based on the deep features, using a multi-task loss function to obtain regions Existence prediction result; extract the shallow features of the input image, use the region existence prediction result as input, obtain the region existence prediction probability mapping result, combine the shallow layer features and the region existence prediction probability mapping result, and extract the local part of the region existence guidance Features; combine the rough segmentation results and the local features guided by the existence of the region to perform segmentation correction, and obtain the segmentation correction results; calculate the pixel-level semantic segmentation results based on the segmentation correction results. Solve the problem of precise semantic segmentation. Achieve efficient multi-layer feature fusion, reduce computational overhead and dependence on the original rough segmentation results and bilinear interpolation, and achieve efficient and accurate pixel-level semantic segmentation.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a semantic segmentation method, network, equipment and computer storage medium. Background technique [0002] Semantic segmentation is a pixel-level classification task that assigns a semantic label to each pixel of an input image, and is widely used in applications such as autonomous driving and augmented reality. At present, the convolutional neural network is the mainstream method of semantic segmentation. It expands the receptive field by stacking convolution kernels and down-sampling operations, and extracts information at different levels from shallow to deep. Generally speaking, low-level features extract local and texture information, which is conducive to fine boundary segmentation; deep features extract global and semantic information, so as to judge object categories more accurately. However, deep features are down-sampled multiple times, and semantic segmentation requi...

AI Technical Summary

Problems solved by technology

[0004] In view of this, a semantic segmentation method is provided to solve the problems of precise semantic segmentation

Images