A fast segmentation method for image semantics based on multi-level network

Abstract

The invention discloses a fast image semantic segmentation method based on a multi-level network, which relates to the field of image semantic segmentation and deep learning. The multi-level semantic segmentation model constructed in the method includes a first-level network, a second-level network and a third-level network , the first-level network includes a densely connected block containing a layer and a convolutional layer, the second-level network includes a densely connected block containing a layer, a cascade layer, and a convolutional layer, and the third-level network includes a densely connected block containing a layer. Concatenated layer and convolutional layer, the cascaded layer of the second-level network is connected to the convolutional layer of the first-level network through the upsampling layer, and the cascaded layer of the third-level network is connected to the cascaded layer of the second-level network through the upsampling layer ; Each densely connected block of the first-level network, the second-level network and the third-level network is also connected to INPLACE‑ABN; the input of the layer in each densely connected block contains the cascaded output of all previous layers The feature map; the invention solves the problem that the image semantic segmentation speed and accuracy cannot be balanced in the prior art.

Description

technical field [0001] The invention relates to the fields of image semantic segmentation and deep learning, in particular to a multi-level network-based fast image semantic segmentation method. Background technique [0002] Image semantic segmentation is a basic task in the field of computer vision. The purpose is to predict the labels of all pixels in the image, so it is considered to be an important task to help obtain a deep understanding of scenes, objects and characters. In recent years, the development of deep convolutional neural networks has made image semantic segmentation a great success. [0003] Most of the current best image semantic segmentation methods are based on fully convolutional neural networks. Fully convolutional neural networks have designed state-of-the-art image semantic segmentation algorithms for a large number of applications, but the effectiveness of these networks depends largely on the design of deep and wide models, which involve many opera...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to design a fast image semantic segmentation method based on a multi-level network to solve the problem that the image semantic segmentation speed and accuracy in the prior art cannot be balanced

Images