3D human modelling

• Main Author: Tan, Jia Le
• Other Authors: Chen Change Loy
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/147910
• Institution: Nanyang Technological University

In recent years, applications such as virtual reality or augmented reality have been gaining popularity as technology progresses rapidly. In particular, the use of 3D objects within these applications has encouraged the acquisition of 3D models to become easier and more accessible to the general public. In view of this, the developer community explored the possibility of modelling a 3D human model from a single image using deep neural networks. While geometry reconstruction of 3D human model has achieved substantial progress, texture reconstruction has received little attention from the community in this regard. In this study, we identify the limitations in the current texture reconstruction framework: (i) texture around unseen regions are featureless (ii) single image reconstruction does not provide sufficient texture information. To address this, we propose Simulated Multi-View Texture Inference Network, an architecture simulating a multi-view texture inference, which circumvents the need for an implicit function to hallucinate the texture around unseen regions. We show that our approach can produce human mesh with consistent texture and perform better than other approaches in some cases. We also conducted experiments to evaluate our approach quantitatively and qualitatively in our study.