Semantic Segmentation to Identify Enemies in First-Person Shooters

Video games are used as simulations in training for activities of real life, with that as a base, this work presents a comparative study between seven different architectures of convolutional neural networks that perform semantic segmentation. These networks are applied to segment individuals in first-person shooters, four of them aim to return a result as fast as possible and the other three are short versions of networks focused on the quality of the segmentation.