Mesh R-CNN++ for 3D Mesh generation: from single to multiple views

• Main Author: Zhang, HengKai
• Other Authors: Qian Kemao
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/156477
• Institution: Nanyang Technological University

Inferring the 3-dimensional structure and geometry of scenes and objects from one or multiple 2-dimensional images has been one of the primary goals of image-based 3D reconstruction. In recent years, with the improved progress of deep learning techniques, and the increasing availability of large 3D training datasets, led to significant advances in 3D shape understanding using deep learning. Inspired by traditional multiple view geometry methods, this project proposed, Mesh R-CNN++, a multi-view deep learning shape predictor. Extensive experiments against current state-of-the-art single and multi-view deep learning shape predictors showed that Mesh R-CNN++ produces 3D models with accurate thin structures and surface details using multiple images.