Weakly-Supervised Image Semantic Segmentation Method Based on Candidate Regions and Neighborhood Classifiers

Abstract

The present invention relates to the field of computer vision, a weakly supervised image semantic segmentation method based on candidate regions and neighborhood classifiers, using candidate regions as the basic processing unit, and applying neighborhood classifiers to image semantic segmentation to obtain high-precision, less susceptible Weakly Supervised Image Semantic Segmentation Confused by Noisy Labels. As a key technology urgently needed in the current field of computer vision, the method of the invention can realize the label prediction of each pixel in the test image, and can obtain relatively high semantic segmentation accuracy of the weakly supervised image.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a weakly supervised image semantic segmentation method based on high-precision candidate regions and neighborhood classifiers. Background technique [0002] The application and demand of image semantic segmentation in the fields of autonomous driving, video surveillance, augmented reality, UAV application and medical image analysis are more and more extensive and urgent. Although fully supervised image semantic segmentation has made great progress in the fields of image classification and object recognition with the help of deep convolutional neural networks. However, fully supervised image semantic segmentation is a data-hungry task that requires a large amount of training data with pixel-level annotations. At the same time, annotating large amounts of pixel-level data is very time-consuming and labor-intensive. [0003] However, image-level weakly supervised annotations are not...

AI Technical Summary

Problems solved by technology

However, the classification network can only identify some small discriminative object regions, and cannot sufficiently realize the inference of semantic labels.

Images