Summary
International Workshop on Smart Info-Media Systems in Asia
2022
Session Number:SS2
Session:
Number:SS2-3
Pseudo-Realistic Food Datasets Generation for Robotic Tasks
Obada Al aama, Yuma Yoshimoto, Hakaru Tamukoh,
pp.73-78
Publication Date:2022/9/15
Online ISSN:2188-5079
PDF download
Summary:
Improving the neural networks recognition performance needs to train it using large dataset, and constructing a food dataset is time and effort consuming. In this paper, we introduce a Cycle Generative adversarial network (Cycle-GAN) to generate a large pseudo-realistic food dataset based on a large number of simulated images and a small number of real images in comparison to traditional techniques. RGB-D camera in two different angles and a turntable were used to capture real RGB-D images of different food samples. 3D modeling software was used to generate simulated images for the 3D food models, which were created by a 3D scanner, using the same configuration of captured real images. A VGG-16 was trained and tested using the generated RGB dataset. Results showed that Cycle-GAN was an effective tool for the generation of near to real images, and that it can be an efficient tool to minimize real image capturing efforts.