Class time/location: Tuesdays 6:15-8:45pm, Lieb Building 2nd floor conference room (just ring
the bell to be let in)
Prof. Quynh Dinh
Office hours: Tuesdays 2-3pm, Lieb Building rm 302. All other times by appointment.
Most of the reading will be technical papers downloaded from below.
No exams or quizzes are planned for this course. However, if participation in paper discussions should drop significantly, short quizzes at the beginning of class time may be given to ensure that the material is being read.
The following are lists of papers that should be covered in class. Please choose at least 2 papers from the Long Papers list and no more than 2 papers from the Short Papers list. Additional papers of interest are listed in the next section categorized by topic. After choosing papers here, choose another paper from the additional papers or a paper from the Long Papers list.
You should choose a total of 5 papers to cover (lead discussions) during the semester.
Links below are to original papers on the specified topic. They form part of the reading assignments for this class.
Implement the algorithms described in any of the papers above or extensions to them (low risk). Please avoid repeating projects from past years. Some algorithms that are particularly amenable to a semester project are:
Survey of Simplification Algorithms. There are many simplification algorithms implemented and available on the internet. In this project, you would download and make necessary modifications so that the programs run. Then experiment on the various algorithms using 10 different models. You will need to analyze and write-up how these algorithms compare with each other. Under what circumstances (what types of models) does each perform well or not well. (low risk)
Use spin-images (developed by Johnson etal.) to match features of 3D shapes. In this algorithm, each point on the surface has an associated spin-image depicting the relative positions of all other surface points. Spin-images can be used to compare the similiarity of points on surfaces. This method may be extended to match surface regions rather than just points. (low risk)
Implement 4 different ways to warp an image. Image warping is used to morph or transition from one image to another. Warps are often defined by a vector field which describes how the pixels of an image moves in changing from a source image to a target. The vector field can be automatically generated from a sparse set of vectors defined by the user. The user defines this sparse set by specifying pairs of corresponding feature points between source and target. In this project, you will implement 2 standard warping algorithms, and 2 experimental algorithms. (medium risk)
Synthesizing features on evolving geometry using texture synthesis applied to 3D geometry. This project involves classifying and identifying features geometrically, then replicating, and blending features to existing geometry. (medium risk)
Using implicit functions to represent volumetric data. In an implicit function, the surface exists where the function evaluates to 0 (or some other constant). Implicit functions have traditionally been used to represent surfaces, but they are actually a volumetric representation. In this project, you will develop ways to use implicit functions to represent volumetric data from medical scans - first in 2D, then in 3D if time permits. (high risk)
Improving the parameterization of an implicit surface through attractive and repulsive forces. In this project, you would use an existing method (and code) to texture map implicit surfaces, and improve the texture mapping by creating repulsive forces in regions where the texture is compressed and attractive forces in regions where the texture is stretched. (medium risk)