ESTIMATING ROBUSTNESS OF DEEP LEARNING MODELS BY THREE ADVERSARIAL ATTACKS

Abstract

The three-dimensional (3D) models have been widely used in many fields such as education, healthcare, and entertainment because they provide a comprehensive appearance of objects in 3D coordinate system. In the healthcare field, for example, 3D models are often used to simulate internal organs, as well as surgical planning or for training purposes new technologies. However, the 3D structure of organs is mostly complex, and processing requires a lot of skills and experience from experts. This paper presents a method for representing 3D modes based on conformal map and spherical harmonics. Firstly, the conformal map is used to transform the source model in sphere coordinate. Then the spherical harmonics is applied to represent the mapped model in different resolution and details. In the experimental section, we simulated the human liver using a public dataset, Medical Image Computing and Computer Assisted Intervention (MICCAI) SLIVER07. The results showed that the proposed method is effective in representing liver model.