Image data set automatic acquisition and annotation framework of land confrontation intelligent agent

Abstract

The invention discloses image data set automatic acquisition and annotation framework for a land confrontation intelligent agent, and the framework mainly comprises a parameter initialization part which is responsible for reading a configuration file and setting the basic parameters of the framework, an image annotation component being composed of three small components including depth informationannotation, semantic segmentation annotation and object bounding box annotation that are respectively responsible for annotating depth information and semantic segmentation information of the image and bounding box information of an object in the image, and a data storage component responsible for storing the annotation information of the image annotation component into a disk. According to the framework provided by the invention, an appropriate virtual scene can be established for different environments of land confrontation, then the image data set under the land confrontation scene is accurately obtained in real time, and the obtained data set can be used for training a land confrontation intelligent agent, so that the confrontation quality and the confrontation capability of the landconfrontation intelligent agent are improved.

Description

technical field [0001] The present invention relates to the field of machine learning and deep learning, in particular to a framework for automatic acquisition and labeling of image data sets of land confrontation agents. This framework can provide synthetic adversarial agent image datasets for various deep learning tasks such as target detection, semantic segmentation, and depth estimation in terrestrial adversarial agent scenarios. Background technique [0002] The advancement of neural networks has greatly promoted the development of computer vision and changed the research landscape and way of thinking in the field of computer vision. The mindset of feature engineering has almost been replaced by the use of neural networks to automatically extract features from large numbers of images. This also means that image data has become the most important resource in the field of computer vision. Some important tasks in the field of computer vision, such as: face recognition, o...

AI Technical Summary

Problems solved by technology

[0003] At present, there are few image datasets specifically for agent confrontation, and directly applying existing datasets to the field of agent confrontation will result in inconsistent data distribution, slow convergence of deep models, low model generalization ability and deep learning. Algorithms are not robust and other problems

Directly re-collecting specialized agent-against datasets would be time-consuming, manpower-consuming, and financially expensive, and would be difficult to progress due to various specific issues

These problems limit the development of artificial intelligence in the field of agent confrontation

Images