Digital image automatic annotation method using cross-task information

Abstract

The invention discloses a digital image automatic annotation method utilizing cross-task information. Transfer learning has been successfully applied to a scene with insufficient annotation samples in an image annotation task. The transfer learning can improve the effect of the target task model by utilizing knowledge of related fields. However, in many image annotation scenes, it is difficult to ensure that an auxiliary field has high correlation with a target task, so that the migration algorithm is often difficult to stably achieve effect improvement. According to the method, common knowledge among the domains is extracted from the auxiliary domain model through a matrix decomposition technology, and knowledge migration is realized through biased regularization of the common knowledge on the target model. The stable effect of a target task model is improved under the condition that the correlation between the fields is uncertain, and an efficient model is learned under the condition that the number of labeled samples is insufficient.

Description

technical field [0001] The invention belongs to the technical field of digital image labeling, and in particular relates to an automatic digital image labeling method utilizing cross-task information. Background technique [0002] Digital image annotation is a common task in artificial intelligence applications. The training of existing image annotation models often requires a large number of labeled samples, which is expensive. However, due to the long-tail distribution of the appearance probability distribution of objects in the real world, there are a large number of object categories and only a small number of labeled samples can be collected. At the same time, some task domains have the property that it is difficult to obtain the labeled data corresponding to the target task. For example, in the fields of medical health and bioinformatics, data annotation needs to be given by experts, and the annotation cost is extremely high. In addition, there are some tasks where ...

AI Technical Summary

Problems solved by technology

[0003] Purpose of the invention: In order to overcome the problem that it is difficult to use the auxiliary domain model to improve the performance of the target model when the domain correlation is low, the present invention provides a digital image automatic labeling method using cross-task information,

Images