Training sample generation method and device, storage medium and electronic equipment

Abstract

The invention discloses a training sample generation method and device, a storage medium and electronic equipment. The training sample generation method comprises: acquiring an object identifier of ato-be-trained object and a model file corresponding to the object identifier, wherein the model file comprises a three-dimensional model and a skin map set, and the skin map set comprises a pluralityof pure-color skin maps and a plurality of colored skin maps; determining corresponding transformation models of the three-dimensional model under different orientation angles, wherein each orientation angle corresponds to one transformation model; generating a plurality of sample image groups according to the plurality of color skin maps and the transformation model; generating a labeling graph of each sample graph group according to the object identifier, the plurality of pure-color skin chartlets and the transformation model; and generating the training sample of the to-be-trained object according to the annotation image and the sample image group, so that the sample image and the annotation image can be automatically generated, manual screenshot and annotation are not needed, and the sample generation efficiency is high.

Description

technical field [0001] The present application relates to the field of computers, and in particular to a method, device, storage medium and electronic equipment for generating training samples. Background technique [0002] Existing recommendation systems based on game content usually generate a promotional poster about the game when recommending a game to users. Basic information about the game. [0003] At present, the character image can be automatically recognized from the game scene with the help of a pre-trained deep learning model. The training samples of the deep learning model are usually prepared manually. Take the glory of the king as an example. To better identify a certain hero, you need to prepare about 1000 training samples. Each training sample is manually obtained from screenshots of the game scene for the hero under different game skins and different map positions, and needs to be manually analyzed. Annotate the sample image. Assuming that King of Glory ...

AI Technical Summary

Problems solved by technology

Assuming that King of Glory has 100 heroes, it is necessary to manually annotate 100,000 training samples with screenshots. According to one person hour, 150 samples can be annotated. Calculated for 8 hours a day, one person needs more than 83 days of work. Obviously, this kind of training Samples are generated in an extremely inefficient and costly manner

Images