A 3D Semantic Predictive Segmentation Method for Object Image Based on Asymmetric Coding Network

Abstract

The invention discloses an object image 3D semantic prediction and segmentation method of an asymmetric coding network. Select RGB images, depth maps and corresponding semantic labels to form a training set, construct a convolutional neural network of an asymmetric encoding network, including two input layers, a hidden layer and three output layers, and input the training set to the convolutional neural network The training is carried out in the training set to obtain the predicted segmented image corresponding to each RGB image in the training set, and the loss function value between the predicted segmented image corresponding to each RGB image in the training set and the corresponding real depth image is calculated, and the minimum value is continuously trained The weight vector and bias item corresponding to the loss function value; the RGB image and depth image to be predicted are input into the trained deep separable convolutional neural network training model to obtain the predicted segmented image. The invention constructs a network structure of an asymmetric coding network, realizes 3D semantic prediction segmentation of an object image, and has high precision of segmentation results.

Description

technical field [0001] The invention relates to a semantic segmentation technology, in particular to a 3D semantic prediction and segmentation method of an object image using an asymmetric coding network. Background technique [0002] The economic take-off has prompted the vigorous development of technology, and artificial intelligence has emerged in response to the needs of the times, and has continuously improved people's quality of life. Various technologies of artificial intelligence have also attracted more and more people's attention. As one of the representatives of artificial intelligence, computer vision tasks have also received increasing attention. 3D semantic segmentation technology is one of the basic computer vision tasks. The importance of visual scene understanding is becoming more and more important. [0003] The goal of indoor 3D semantic segmentation is to predict the class label for each pixel in the input image. It is a fundamental task of computer vis...

AI Technical Summary

Problems solved by technology

However, the low precision of manual features prevents the technology from making substantial progress, which leads to the unsatisfactory results of traditional methods in the field of semantic segmentation.

Images