Projects

If you are unsure what would make a good project feel free to email me and the TA with your thoughts. You could briefly outline 1, 2 or 3 project in a few sentences each, and ideally provide links to relevant paper(s) or web pages. We will respond with feedback on each of the ideas asap.

Deliverables

Proposal For the proposal you are expected to have researched the available sources (books, papers, the web), and explored available software systems. Summarize your findings in the literature in 2-4 pages, appropriately citing your sources. Write a clear description of your proposed project including motivations (why is it useful), a succinct description of the theory you plan to draw on (1-2 pages) and a plan and schedule for your proposed work. You need to break down your project into several steps and describe how you will demonstrate and verify these steps. (This is to avoid the risk that the final program and demo does not work at end of classes. With steps at least you have something working) Final project report In addition to above, describe in detail the work you did, describe the results (experimental and/or theoretical), and discuss their relevance and applicability. The format and quality of the report should be similar to a short (6-10 pages) research paper. Study the style and conventions in the papers you read for the proposal. If you are involved in a group project note that there must be individually identifiable content. That is you write individual parts/sections that combine into one report. In the the first case the report should be focussing on what that individual did in the project (but briefly referring to other member parts for completeness). In the second case each person would write a technical chapter on their own (on their part. About 4-6 pages -- this is the "beef" of your contribution), and all the group members would collaborate on writing an introduction experiments and conclusions sections.

Final report submission details

Suggested topics

Geometry

1. Analysis, editing and improvements of SFS geometry. This project is suitable for someone with some prior exposure to either computational geometry or graphics. You will study the particularities of geometries generated from SFS and how to improve these.
2. Scene modeling from SFM: The 3D points obtained from a SFM and visual stereo system needs analysis, correcting and triangulation before it can be used in e.g. rendering. One can do this using ARC 3D for 3D point generation, meshlab for editing of the point cloud, and surface generation, and capgui for texturing. The project consists of analyzing what are cumbersome steps and algorithmic improvements in this chain. Try to improve those steps by either implementing better functionality or considering and analysing the performance of available software packages.
3. Study manual agmentation of captured point sets. In many cases a model in terms of simple geometric primitives is desired (e.g. a house could be modeled using 3D solid geometry primities parallelepipedes for the bottom part and a prism for the roof). A task would be how to easily with some (but minimal) user interaction go from a 3D point set of a scene to a model represented by the composition (union) of 3D solid geometry primitives (spheres, cylinders, parallel epipedes, prisms etc). Alternatively one could use lines, planes and curves for surfaces.

Texturing and appearance:

1. One could consider what is the best subset of view angles to generate these textures and partition the view sphere (or circle) into e.g. quadrants. Figure out how to handle overlap and the boundaries. Study references from graphics literature, e.g. "A Theory of Locally Low Dimensional Light Transport" by Dhruv Mahajan, Ira Kemelmacher Shlizerman, Ravi Ramamoorthi, and Peter Belhumeur
2. In some cases (e.g. geometry very inaccurate) other (e.g. non-linear) subspace methods may perform better (e.g. isomap, kernel pca, etc). Try to find the properties of the texture manifold and matc h that to a suitable method.
3. Write a texturing plugin using some method (e.g. original or one of the above) for Blender so captured objects can be integrated and rendered in blender scenes.
4. Texture decomposition into low/high frequency.

   Single texturing:

5. In some cases when using models in legacy SW that can only handle a single texture, one has to distill the in some measure "best" color w.r.t. all the input data for a particular object point. There are both heuristic methods for this ( Blending images for texturing 3D Models, Adam Baumberg BMVC 2002) and globally optimal (Spatio-temporal Image-based Texture Atlases for Dynamic 3-D Models Janko and Jean-Philippe Pons 3DIM2009, Seamless image-based texture atlases using multi-band blending, Allène, C. and Pons, J-P. and Keriven, R. ICPR 2008, Seamless Mosaicing of Image-Based Texture Maps Victor Lempitsky and Denis Ivanov CVPR.2007)
   Study these methods (we have some example code that can get you started) and see how they compare. See if you can integrate one into the CapGui

1. Study a different tracking modality, e.g. hyperplanes (Juri and Dhome), Meanshift (Dorin Comaniciu, Peter Meer) integrate this tracking into XVison and compare performance with existing XVision trackers.
2. Implement compositional tracking similar to the IFA system.

How to go about it

Then find something to research and implement yourself. This could be a subtopic such as geometric refinement or an applicatino where you capture and use vision based models for some effects in e.g. an animation or computer game.

In a tracking project you can similarly first try our XVision system and or other trackers you can download, then propose some extension or additional tracking modality to implement.