Study of GANs using a Few Images for Sealer Inspection Systems

EasyChair Preprint 2549

Abstract

This paper describes a comparative study of the performance of Generative Adversarial Networks (GANs) through the quality of the generated images by using a few samples. In the deep learning-based systems, the amount and quality of data are important. However, in industrial sites, data acquisition is difficult or limited for some reasons such as security and industrial specificity, etc. Therefore, it is necessary to increase small-scale data to large-scale data for the training model. GANs is one of the representative image generation models using deep learning. Three GANs such as DCGAN, BEGAN, and SinGAN are used to compare the quality of the generated image samples. The comparison is carried out based on the score with different measuring methods.

Keyphrases: Generative Adversarial Networks, Sealer, Vision Inspection Systems

@booklet{EasyChair:2549,
  author    = {Dongwook Seo and Yejin Ha and Seungbo Ha and Kang-Hyun Jo and Hyun-Deok Kang},
  title     = {Study of GANs using a Few Images for Sealer Inspection Systems},
  howpublished = {EasyChair Preprint 2549},
  year      = {EasyChair, 2020}}