Real-time image semantic segmentation method and system, readable storage medium and terminal

Abstract

The invention discloses a real-time image semantic segmentation method and system, a readable storage medium and a terminal. The method comprises the steps that an encoder carries out convolution operation on an input original feature image by utilizing a convolution unit based on feature channel splitting and rearrangement so as to extract image features and output the image features to a decoder; and the decoder adopts an attention pyramid network model to carry out intensive feature extraction on the output feature image of the encoder, maps the extracted feature to a segmentation category,and finally samples a segmentation result to an input original feature image resolution. According to the scheme, when the real-time image semantic segmentation is carried out by using limited computing resources, the segmentation accuracy is improved.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a real-time image semantic segmentation method and system, a readable storage medium and a terminal. Background technique [0002] Recently, building larger and deeper convolutional neural networks (CNNs, Convolutional Neural Networks) has become a major trend in solving scene understanding tasks. The most accurate CNNs typically have hundreds of convolutional layers and thousands of feature channels, and despite achieving higher performance, these advances come at the expense of model runtime and speed. Especially in the context of many real-world scenarios such as augmented reality, robotics, and self-driving cars, real-time prediction and evaluation of networks under resource-constrained conditions is often required. Therefore, those precise networks that require a lot of resources are not suitable for mobile platforms (e.g., drones, robots, and smartphon...

AI Technical Summary

Problems solved by technology

Therefore, those precise networks that require a lot of resources are not suitable for mobile platforms (e.g., drones, robots, and smartphones) with limited resources such as computing power, storage capacity, and energy overhead.

Images