Cretu, Stefan (2018) A parallel approach for the Shape, Illumination and Reflectance from Shading algorithm. Master's Thesis / Essay, Computing Science.
|
Text
master-thesis-StefanCretu-RUG2018.pdf Download (945kB) | Preview |
|
Text
toestemming.pdf Restricted to Registered users only Download (94kB) |
Abstract
Shape from shading (SfS) is a generic term for a family of algorithms that aims to reconstruct 3D geometric information from one or multiple views (static 2D images) of a 3D scene. A recent method, called Shape, Illumination and Reflectance from Shading (SIRFS) proposes a high-quality solution to the SfS problem. However, SIRFS is computationally expensive, which makes its trial-and-error usage for parameter setting experimentation prohibitive. In this project, we examine the end-to-end problem of parallelizing SIRFS. For this, we first reverse-engineer the current MATLAB implementation and propose a porting plan to C/C++. Next, we examine the performance bottlenecks of such a C/C++ solution, and propose parallelization solutions to it both on the CPU and GPU. We examine the pro’s and con’s of these solutions in detail. We conclude by discussing the obtained insights, stressing on the techniques and designs which have proven successful for the acceleration of SIRFS, but also outlining still existing hard bottlenecks that were discovered but not solved.
Item Type: | Thesis (Master's Thesis / Essay) |
---|---|
Supervisor name: | Telea, A.C. |
Degree programme: | Computing Science |
Thesis type: | Master's Thesis / Essay |
Language: | English |
Date Deposited: | 27 Jul 2018 |
Last Modified: | 30 Jul 2018 14:17 |
URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/18094 |
Actions (login required)
View Item |