설계 요구사항이 반영된 선형 생성을 위한 생성적 적대 신경망 활용

Abstract

A generative adversarial network(GAN) can make high-quality output through adversarial games between a generator and a discriminator. However, the vanilla GAN cannot consider requirements in generation because the input of the GAN comprises only a random vector. This paper presents a novel GAN for generating ship hulls in the early design stages. The GAN uses a new loss function to generate a hull that reflects the design requirements defined by users. The generated hull is represented using voxels. The argues on the conversion from the voxels to the CAD format is given in conclusion.