Data generation method, device, server and storage medium

Abstract

The embodiment of the invention discloses a data generation method, device, server and storage medium. The method comprises the following steps of: constructing a real three-dimensional model according to acquired real scene data; constructing a virtual three-dimensional model, and fusing the real three-dimensional model and the virtual three-dimensional model to obtain a scene three-dimensional model; and acquiring the data of the scene three-dimensional model as scene data. By acquiring the acquired real scene data and constructing the real three-dimensional model according to the real scenedata, the model data of the real scene can be obtained; by constructing the virtual three-dimensional model, a plurality of scene models can be obtained, so that the three-dimensional model is more diversified; and by fusing the real three-dimensional model and the virtual three-dimensional model, the obtained scene three-dimensional model has characteristics of real color, texture and the like,and the diversified scene three-dimensional model is more authentic, so that the obtained data is close to the real data, and the accuracy is higher.

Description

technical field [0001] Embodiments of the present invention relate to the field of data generation, and in particular, to a data generation method, device, server, and storage medium. Background technique [0002] At present, in the training of self-driving vehicles, a large amount of training data is required, so that the self-driving vehicles can realize the recognition and target positioning of the surrounding traffic environment according to the training data and deep learning. [0003] At present, there are mainly two methods for obtaining training data, real data collection and virtual data generation. Obtaining training data through real data collection requires a lot of manpower, hardware and energy costs, and it is difficult to obtain data of vehicles in dangerous situations in extreme weather. , such as data of scenarios such as sudden braking, sharp turns, and car crashes, which makes the training data not comprehensive and accurate enough. Although the method of...

AI Technical Summary

Problems solved by technology

[0003] At present, there are mainly two methods for obtaining training data, real data collection and virtual data generation. Obtaining training data through real data collection requires a lot of manpower, hardware and energy costs, and it is difficult to obtain data of vehicles in dangerous situations in extreme weather. , such as data of scenarios such as sudden braking, sharp turns, and car crashes, which makes the training data not comprehensive and accurate enough

Although the method of generating training data through virtual data can obtain data of dangerous scenarios, the data scenarios are not realistic and comprehensive enough, and it is difficult to achieve the fidelity of real data, thus affecting the accuracy of training

Images