Inferring Multi-view Images from a Single Image Through Machine Learning

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Inferring Multi-view Images from a Single Image Through Machine 
        Learning"

by

TSUI Yuk Hang

Abstract:

Multi-view images are essential information in pose recognition for 3D 
space and 3D mesh reconstruction. Compared with collecting multi-view 
images by users or producing them from 3D models, constructing multi-view 
images from a single-view image through Machine Learning facilitates the 
process of collecting data and reduces technological requirements. 
Inferring other faces of an object from a single 2D image requires the 
model to derive the shape of other views and transfer the corresponding 
pattern to other sides. Existing models including ShaRF or SDF-SRN will 
relate the image to a 3D object first, then generate information under 
different scenarios. In this project, we propose a network that generates 
multi-view images without inferring the shape of the object. This helps 
reduce the model size and running time which is critical in some 
applications. The experiment with ShapeNet data and real images 
demonstrated the effectiveness of the approach.


Date            : 5 May 2023 (Friday)

Time            : 14:30 - 15:10

Venue           : Room 4582 (near lifts 29/30), HKUST

Advisor         : Dr. BRAUD Tristan

2nd Reader      : Prof. SANDER Pedro