A method for constructing a sorting scene parallel data set based on sparse actual acquisition data

Abstract

The invention discloses a method for constructing a sorting scene parallel data set based on sparse actual acquisition data, which comprises the following steps of first generating a real data set; 2importing a real data set into a 3D processing tool and construct a manual sorting and stacking scene; 3 rendering that graphics to generate an artificial data set; 4 mixing the real data set and themanual data set according to the scene and influencing factors, and placing the same category in the same folder to form a parallel data set which combines the virtual and the real. The construction method of the parallel data set disclosed by the invention solves the problems of low collection efficiency, time-consuming and labor-consuming of the real data set and single content, and makes the data set diverse, reliable, good applicability and more suitable for training of the neural network. At the same time, the use of automatic annotation technology can effectively reduce the time and costof data set annotation, and improve the accuracy of data annotation.

Description

technical field [0001] The invention relates to a data set construction method, in particular to a sorting scene parallel data set construction method based on sparse actual data. Background technique [0002] With the development of the digital age, data is becoming more and more important. But data collection is a huge challenge. Most of the existing dataset collection methods are manually collected and labeled, which makes collecting and labeling datasets time-consuming and laborious, and the quality of the collected datasets is not high, which has limitations in all aspects. Such as the perspective of collection, the content included, etc. However, the virtual data sets collected by tool software (hereinafter collectively referred to as artificial data sets) are efficient but divorced from reality, and their authenticity and reliability are not high. Therefore, the prior art has the following problems: [0003] (1) The collection efficiency of real data sets is low, ...

AI Technical Summary

Problems solved by technology

[0003] (1) The collection efficiency of real data sets is low, time-consuming and labor-intensive, and the content is single

[0004] (2) In the artificial data set, there is no data collection and labeling based on the real data set, so that the data set collected only by tool software is out of reality, and the authenticity and reliability are not high

[0005] (3) The real data set and the artificial data set are not mixed and used according to certain rules, which makes the data set have problems such as large limitations and poor applicability

Images