Semi-supervised learning method and system based on target segmentation field self-learning

Abstract

The invention provides a semi-supervised learning method based on target segmentation field self-learning. The method comprises the following steps: training an initial segmentation network by using marked data in a training data set; generating a pseudo label from unmarked data in the training data set through the trained initial segmentation network; performing shape quality evaluation and semantic quality evaluation on the generated pseudo label; fusing the shape quality and the semantic quality to obtain pseudo label quality; estimating the distribution of the real labels and the pseudo labels, and optimizing the distribution of the pseudo labels; adding data with relatively high pseudo label quality into the training data set to expand the training data set; optimizing the trained initial segmentation network by using the expanded training data set; and iteratively repeating the above steps until the performance of the segmentation network is saturated. The invention further provides a corresponding system, a terminal and a medium. The problem of low segmentation precision in the target segmentation field under the condition of a small number of sample annotations is solved, and good performance is realized.

Description

technical field [0001] The invention relates to a method in the field of computer vision and image processing, in particular to a semi-supervised learning method and system based on self-learning in the field of target segmentation. Background technique [0002] In recent years, with the rapid development of computer vision technology, semantic segmentation task, as an important branch of vision tasks, has been widely studied. Using a deep neural network to train a semantic segmentation task often requires manual labeling of images in the dataset as supervision, but manual labeling for each dataset is extremely costly in manpower, material and financial resources. Therefore, there are usually a lot of unlabeled data in the dataset, and only a small part of the labeled data. How to effectively use unlabeled data to solve the problem of less labeling is the focus of research. In order to solve this problem, semi-supervised learning techniques have been widely studied in recen...

AI Technical Summary

Problems solved by technology

There is still a problem with this method. When generating the confidence map, the confidence map will tend to divide the target areas segmented in the segmentation map into areas with high confidence, and cannot pay attention to whether the segmentation target is accurate.

Because the segmentation map is a binary map, there is not enough image information, so the obtained confidence map is also inaccurate.

[0005] It can be seen that the existing technology still has the problem that the performance gain of the model is unstable or even reduces the performance of the model. At present, there is no description or report of a technology similar to the present invention, and no similar information has been collected at home and abroad.

Images