Method for generating training data and electronic device

Abstract

The invention provides a method for generating training data and an electronic device. The method comprises the following steps: obtaining an object model of a specific object; obtaining a first image when the object model is at the first angle and a first silhouette corresponding to the first image; extracting a first object image from the first image based on the first silhouette; embedding the first object image into the first background image to generate a first training image; generating first mark data of the first object image in the first training image; and taking the first training image and the first mark data as first training data of the specific object. Therefore, the training data can be automatically, quickly and correctly generated, so that the related time and money cost is reduced.

Description

technical field [0001] The invention relates to a deep learning technology, in particular to a method and an electronic device for generating training data. Background technique [0002] At present, deep learning requires huge training data for training to get good results. However, in the process of product development, some training data is difficult to obtain, and it is very time-consuming and expensive for manufacturers. [0003] Generally speaking, at present, the generation and collection of training data usually use manual labeling, but because it is very difficult for humans to label human body poses or object poses, it takes a lot of time to label each image. time, and the error rate is very high. [0004] In addition, for some specific objects (such as a product) unique to some manufacturers, since the existing data sets on the market may not be available, the actual data may have to be collected by the manufacturer itself, which consumes unnecessary time and mo...

AI Technical Summary

Problems solved by technology

However, in the process of product development, some training data is difficult to obtain, and it is very time-consuming and expensive for manufacturers

[0003] Generally speaking, at present, the generation and collection of training data usually use manual labeling, but because it is very difficult for humans to label human body poses or object poses, it takes a lot of time to label each image. time, and the error rate is very high

[0004] In addition, for some specific objects (such as a product) unique to some manufacturers, since the existing data sets on the market may not be available, the actual data may have to be collected by the manufacturer itself, which consumes unnecessary time and money

Images