Target image data set generation method and device thereof

Abstract

The invention provides a target image data set generation method and a device thereof. The method comprises the following steps: acquiring a plurality of first images, and constructing a first image library; automatically identifying and understanding a plurality of first object model images in the first image library; acquiring a plurality of second images, and constructing a second image library; automatically identifying and understanding a plurality of second object model images in the second image library; and in accordance with at least one of a plurality of image visual effects regarding the plurality of first object model images and the plurality of second object model images, obtaining a target image data set under different environmental models and / or under different motion conditions. By adoption of the target image data set generation method and the device thereof provided by the invention, mass customized learning samples can be generated quickly and automatically.

Description

technical field [0001] The present invention relates to a kind of image processing method and its device, more specifically, the present invention relates to a kind of target image data set generating method and its device. Background technique [0002] At present, due to the rapid development of the artificial intelligence industry, deep learning has made breakthroughs in the fields of computer vision, speech recognition, and natural language processing. The great success of deep learning research is largely due to the availability of a large number of excellent training sample data sets. [0003] There are already a large number of publicly available datasets available worldwide for research evaluation. For example: ImageNet, MINIST, COCO, PASCAL VOC, etc. In addition, for the labeling of large datasets, the form of crowdsourcing is basically adopted worldwide, and some dataset labeling tools are used for manual labeling to ensure the high accuracy of the sample dataset....

AI Technical Summary

Problems solved by technology

It should be pointed out that for a specific field, it is difficult to obtain a large number of training samples (especially positive samples)

For example, due to the particularity of military applications, it is difficult to obtain a large number of training sample data sets in the military field

For specific areas where it is difficult to obtain public sample data, Zhang Xiaoguo et al. did not propose an effective solution

Images