Object detection with one-shot Convolutional Neural Networks for playing the game Cuphead

Abstract

Video games and the video game industry have considerably grown in the last years. Taking into consideration the importance of video games and artificial intelligence nowadays, and among all the applications that artificial intelligence has these days, the goal of this project has been to develop an AI-based approach for playing the videogame Cuphead using Convolutional Neural Networks, a decision making system and automatic keyboard interaction. Different object detection models were tested and after choosing one of them, it was fine-tuned using data that was labeled automatically using a method created for this purpose.