3D reconstruction from depth images using machine learning

Alhamazani, Fahd 2023. 3D reconstruction from depth images using machine learning. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (12MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (97kB)

Abstract

This thesis investigates 3D reconstruction from depth images, focusing on three related tasks. First, we present our work on reconstructing complete volumetric shapes from a single depth image. Our model proposes to incorporate a dynamic latent code, allowing the model to determine the appropriate code for the estimation. We further develop a multi-stage approach to iteratively improve completion, and employ a classifier as an auxiliary task to enhance estimation. Second, we advance the quality metric for 3D shapes by leveraging rendering them using various styles and from different views. We also improve the SSIM metric by introducing a mask to ensure it is stable with different rendering canvas sizes. Subsequently, we develop a neural network to mimic human visual judgment. Lastly, traditional reconstructions primarily target rigid bodies due to the straightforwardness of their shape formation. So we further develop a method for predicting canonical form, i.e., returning shapes to their original pose, which can significantly simplify shape completion for deformable objects.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date of First Compliant Deposit:	23 May 2024
Last Modified:	24 May 2024 07:48
URI:	https://orca.cardiff.ac.uk/id/eprint/169137

Actions (repository staff only)

Edit Item