Semi-supervised semantic segmentation method based on maximum confidence

Abstract

The invention discloses a semi-supervised semantic segmentation method based on maximum confidence, and the method comprises the steps: selecting a part of images from an existing training data set asmarked images, and taking the remaining images as unmarked images; constructing a network model, and predicting prediction class probability graphs of the marked images and the unmarked images through a segmentation network in the network model; maximizing the confidence coefficient of the marked image prediction class probability graph by adopting supervised learning and adversarial generation modes; predicting a segmentation error region in the unmarked image prediction class probability graph by adopting an unsupervised learning mode; training the network model by combining loss of supervised learning and loss of unsupervised learning; and in a test stage, inputting an unmarked image to be segmented into the trained network model to obtain a segmented semantic image. According to the scheme provided by the embodiment of the invention, semantic segmentation can be accurately carried out on the unmarked image.

Description

technical field [0001] The invention relates to the field of image semantic segmentation, in particular to a semi-supervised semantic segmentation method based on maximum confidence. Background technique [0002] Image segmentation refers to dividing an image into several disjoint regions based on features such as gray scale, color, spatial texture, geometric shape, etc., so that these features show consistency or similarity in the same region, but appear in different regions. obvious difference. Simply put, in an image, different targets are separated from the background, and it is clear from the segmentation result what object is segmented. Overall, semantic segmentation is a difficult task aimed at scene understanding. Scene understanding, as the core problem of computer vision, has been widely used in today's information society. These applications include: autonomous driving, human-computer interaction, computer photography, image search engines, and augmented realit...

AI Technical Summary

Problems solved by technology

[0004] 1) Due to the lack of detailed boundary location information, the performance of weakly supervised methods is far inferior to that of fully supervised methods

[0005] 2) Some semi-supervised learning methods are very inefficient in using unlabeled data because they ignore a large amount of available misclassification information

Images