[17a-A37-1~8] CS.2 Code-sharing Session of 3.2 & 4.4

9:00 AM - 9:30 AM JST(12:00 AM - 12:30 AM UTC)English Presentation

[17a-A37-1] [JSAP-Optica Joint Symposia Invited Talk] Deep Neural Network 3D Reconstruction Using One-Shot Color Mapping of Reflectance Direction Fields

〇Hiroshi Ohno¹ (1.Toshiba RDC)

Keywords:

3D reconstruction,color-mapping of reflectance direction field,Deep neural network

In many manufacturing processes, real-time inspection of microscale three-dimensional (3D) surfaces is crucial. Therefore, a method integrating deep neural networks (DNNs) has been proposed for obtaining a microscale 3D surface from a single image, or two images, captured by an imaging system referred to as the one-shot BRDF (Bidirectional Reflectance Direction Function) system, equipped with a multicolor filter. This system can acquire reflectance direction fields using one-shot color mapping that assigns light directions to specific colors. Assuming a smooth and continuous surface, the 3D shape can be reconstructed from either a single image or two images captured by the one-shot BRDF system. This process utilizes DNNs that function as a gradient descent method to solve nonlinear partial differential equations, without the need for training data. This DNN-incorporated method is referred to as OneShot3DNet. The effectiveness of this method has been validated numerically and experimentally.