This project aims at the creation of an off-screen rendering toolbox that enables the user to create a collection of predefined renderings for one (or many) dataset(s).
The ability to extract renderings of a 3D model is crucial in many 3D object retrieval and recognition tasks. Furthermore, the increased use of Convolutional Neural Networks in such tasks makes the aforementioned problem even more important. The renderings can vary from simple planar to panoramic views to more complex examples of angular distance rendering and/or filtered outputs.
The technical requirements of the project are the following:
• An API which is able to extract at least planar and panoramic views of a 3D model in both distance and angular difference mode
• The API should be able to perform the above tasks on a series of 3D models (i.e., a dataset).
• A CLI tool that enables the use of the API
• A mex interface function that enables the use of the API through MATLAB
• [OPTIONAL] a GUI that allows the user to call the API and set different parameters
Initial code for off-screen rendering is already available and will be provided, as well as code for the extraction of panoramic views in an efficient manner.
PANORAMA: A 3D Shape Descriptor Based on Panoramic Views for Unsupervised 3D Object Retrieval, International Journal of Computer Vision, September 2010, Volume 89, Issue 2, pp 177–192.
[Necessary] Good programming skills in C/C++ AND MATLAB, TDT4195 (Visual Computing Fundamentals), TDT4230 (Grafikk/Visualisering). Background knowledge in using computer graphics and vision libraries and tools (OpenGL, FreeGLUT, MESA, WGL/GLX). Vertex Buffer Object and Framebuffer programming (sample code will be provided).
[Desirable] TDT4265 (Datasyn). Some GPU programming and the use of shaders.
Dataset to be used: Various 3D Object datasets will be provided.
Advisors: Konstantinos Sfikas, Theo Theoharis
If interested, please send email to Konstantinos (firstname.lastname@example.org) by May 7th, 2017.