For the Spring 2005 semester, our seminar will be meeting every Wednesday at 4:00 in 3401 Siebel (except on Feb 2).

If you haven't already subscribed to our
mailing list, you should do that now.

• Feb 2 – Organizational meeting in 4407 Siebel

• Feb 9 – ZP+: Correct Z-Pass Stencil Shadow Volumes slides (Jared Hoberock)
  We present a novel algorithm for the rendering of hard shadows cast by a point light source. The well-known Z-pass method for rasterizing shadow volumes is not always correct. Our algorithm, which we call ZP+, elegantly corrects Z-pass defects. ZP+ takes advantage of triangle strips and fast culling capabilities of graphics hardware not available to conventional robust methods like Z-fail. While Z-fail can be up to 80% slower than Z-pass, our new method ZP+ is typically less than 10% slower than Z-pass. Finally, we give a comparison of the three methods. When a scene is geometry-bound, ZP+ is always faster than Z-fail. We also explain why, in some situations, Z-pass (hence ZP+) is surprisingly slower than Z-fail on more recent graphics hardware.

• Feb 16 – Spectral Surface Reconstruction From Noisy Point Clouds slides (Gautam Kumar)
  We introduce a noise-resistant algorithm for reconstructing a watertight surface from point cloud data. It forms a Delaunay tetrahedralization, then uses a variant of spectral graph partitioning to decide whether each tetrahedron is inside or outside the original object. The reconstructed surface triangulation is the set of triangular faces where inside and outside tetrahedra meet. Because the spectral partitioner makes local decisions based on a global view of the model, it can ignore outliers, patch holes and undersampled regions, and surmount ambiguity due to measurement errors. Our algorithm can optionally produce a manifold surface. We present empirical evidence that our implementation is substantially more robust than several closely related surface reconstruction programs.

• Feb 23 – Volume Interval Segmentation and Rendering (Yuan Zhou)
  In this paper, we segment the volume into geometrically disjoint regions that can be rendered to provide a more effective and interactive volume rendering of structured and unstructured grids. Our segmentation is based upon intervals within the scalar field, producing a set of geometrically defined interval volumes. We present many advantageous properties in using interval volumes, and provide several new rendering operations or shaders to provide effective visualizations of the 3D scalar field. In particular, we demonstrate new technologies that allow interval volumes to be rendered interactively and/or used to reduce the amount of rasterization or rendering primitives in a volume renderer. We illustrate the use of interval volumes to highlight contour boundaries or material interfaces. Several surface shaders that can easily be integrated in the volume renderer are presented. To construct the interval volumes, we cast the problem one dimension higher, using a higher-dimensional isosurface construction for interactive computation or segmentation. The algorithm is independent of the dimension and topology of the polyhedral cells comprising the grid, and thus offers an excellent enhancement to the volume rendering of unstructured grids. We present examples using hexahedral and tetrahedral cells from time-varying and multi-attribute datasets.

• Mar 2 – Surfacing by Numbers (Steve Zelinka)
  We present a novel technique for surface modelling by example called surfacing by numbers. Our system enables effortless reuse of fine-scale detail from existing 3D models or images. The user simply selects a source region and a target region, and our system transfers detail from the source to the target. The source may be elsewhere on the target surface, on another surface altogether, or even part of an image. As the transfer is formulated as a synthesis problem, solved using a novel surface-based adaptation of graph cut synthesis, the source and target regions need not match in size or shape, and details to be transferred can be geometric, textural or even user-defined in nature. A major contribution of our work is our fast, graph cut-based interactive surface segmentation algorithm. Unlike approaches based on scissoring, the user loosely strokes within the body of each desired region, and the system computes optimal boundaries between regions via minimum-cost graph cut. Thus, less precision is required, the amount of interaction is unrelated to the complexity of the boundary, and users do not need to search for a view of the model in which a cut can be made. (Contact me for copies of slides/paper)

• Mar 9 – Ko's paper on stable but non-disappative water by some student that didn't post the title and abstract.
  I did manage to find time to put together a page on the SemiLagrangian method. Let's try to make more of these. Feel free to create a page on some fancy technique or edit one that's already there.

• Mar 16 – Interactive modeling of topologically complex geometric detail slides (Wei-Wen Feng)
  Volume textures aligned with a surface can be used to add topologically complex geometric detail to objects in an efficient way, while retaining an underlying simple surface structure.Adding a volume texture to a surface requires more than a conventional two-dimensional parameterization: a part of the space surrounding the surface has to be parameterized. Another problem with using volume textures for adding geometric detail is the difficulty in rendering implicitly represented surfaces, especially when they are changed interactively.In this paper we present algorithms for constructing and rendering volume-textured surfaces. We demonstrate a number of interactive operations that these algorithms enable.

• Apr 13 – Fragment-Based Image Completion (Patrick Delfert)
  We present a new method for completing missing parts caused by the removal of foreground or background elements from an image. Our goal is to synthesize a complete, visually plausible and coherent image. The visible parts of the image serve as a training set to infer the unknown parts. Our method iteratively approximates the unknown regions and composites adaptive image fragments into the image. Values of an inverse matte are used to compute a confidence map and a level set that direct an incremental traversal within the unknown area from high to low confidence. In each step, guided by a fast smooth approximation, an image fragment is selected from the most similar and frequent examples. As the selected fragments are composited, their likelihood increases along with the mean confidence of the image, until reaching a complete image. We demonstrate our method by completion of photographs and paintings.

• Apr 27 – Water Drops on Surfaces (paper not available yet) (Ivan Lee)
  We present a physically-based method to enforce contact angles at the intersection of fluid free surfaces and solid objects, allowing us to simulate a variety of small-scale fluid phenomena including water drops on surfaces. The heart of this technique is our virtual surface method, which modifies the level set distance field representing the fluid surface in order to maintain an appropriate contact angle. The surface tension that is calculated on the contact line between the solid surface and liquid surface can then capture all interfacial tensions, including liquid-solid, liquid-air and solid-air tensions. We use a simple dynamic contact angle model to select contact angles according to the solid material property, water history, and the fluid front�s motion. Our algorithm robustly and accurately treats various drop shape deformations, and handles both flat and curved solid surfaces. Our results show that our algorithm is capable of realistically simulating several small-scale liquid phenomena such as beading and flattened drops, stretched and separating drops, suspended drops on curved surfaces, and capillary action.