Improvement of Deep Learning Technology to Create 3D Model of Fluid Art

Abstract

Art is an essential part of the entertainment. As 3D entertainment such as 3D games is a trend, it is an exciting topic how to create 3D artworks from 2D artworks. In this work, we investigate the 3D reconstruction problem of the artwork called “Sound of Ikebana, ” which is created by shooting fluid phenomena using a high-speed camera and can create organic, sophisticated, and complex forms. Firstly, we used the Phase Only Correlation method to capture the artwork’s point cloud based on the images captured by multiple high-speed cameras. Then we create a 3D model by a deep learning-based approach from the 2D Sound of Ikebana images. Our result shows that we can apply deep learning techniques to improve the reconstruction of 3D modeling from 2D images with highly complicated forms.