Research On The Generation Of E-sports Cards Based On Deep Learning

In recent years,with the increasing enrichment of entertainment resources,e-sports games have sprung up in people's field of vision.The production of e--sports games involves a lot of content creation,and the game content creation contains a large number of repetitive,randomly generated parts,which can often be automatically generated in some certain ways.In all kinds of e-sports games,the e-sports card game is very popular among players because of its fast pace and easy to play.E-sports card game company need to maintain a fast and large number of card design speed to maintain the popularity of e-sports card game.Some players will also have their own ideas for designing cards while playing games.Both will be hard to create interesting cards because of lack of inspiration.Based on the above situation,this thesis first introduces the research and the status quo of deep learning in the field of e-sports games,and elaborates on related technologies.Then,two kinds of e-sports card generation schemes based on deep learning are proposed,which are:(1)Using the image caption method in the field of deep learning to generate esports cards.(2)After analyzed the characteristics of Magic The Gathering,this thesis proposes to use the Recurrent Neural Network to generate card text,and to generate the e-sports card by matching the input image with the generated text through probability vector matching.Based on this,a joint training is proposed to improve the rationality of matching.This thesis collects the e-sports card datasets of the two e-sports card games named Hearthstone and Magic The Gathering,and correspondingly pre-processed for the specificity of the e-sports card text.Then the above-mentioned e-sports card generation experiment was carried out on two esports card datasets.The feasibility and generalization of the e-sports card generation method and the improved method to improve the matching rationality were verified by experiments.