Subdivision Surfaces

March 26, 2014, GPU Technology Conference, San Jose, CA—David Yu from Pixar talked about subdivision surfaces and their contributions to make the technologies available to all. The idea is to reduce the number of edges and vertices to a minimum at the entry stage, and refine and smooth the edges with other tools.

Specifying the geometry of surfaces starts with a polygonal mesh in Maya. Subdividing the edges provides a limiting surface for smoothness. The base mesh is considered level 0 for the algorithm and is the same as a cage view of that object, fairly coarse. By iteratively subdividing and smoothing, the object soon achieves the desired appearance without needing to be completely hand drawn. The techniques allow for the use of far fewer triangles to represent the surface.

Consider a coffee cup in a 3-D graphics program. A minimal view would be a wire-frame view with many sharp edges and barely recognizable structure. Applying the algorithms to the drawing causes the few faces at the boundary of the handle and the cup to be broken into hundreds of smaller triangles. These triangles are averaged for a common curvature to create a smooth fillet between the handle and the cup. A much quicker and simpler process than drawing all the faces by hand.

Comparing the efforts to use coarse grids and subdivision smoothing versus a full polygonal approximation shows why this technique is of great value. The polygonal approximation the artist to work at full detail on all objects, and generates hundreds of thousands of faces. The minimalist approach allows for arbitrary topologies to have minimal detail which only needs to be rendered in the final version.

For example in “Toy Story”, the characters only require about 10k faces. Woody and Buzz have a lot of detail that is not needed for many shots, but the continuity requires that the full detail be present. In “Cars” McQueen requires 70k faces, due to the details in the wheels and the larger facial animations. By subdividing the raw images through 4 iterations and integrating, the artists can generate characters quickly and repeatability.

Since everything is not smooth, they also created the capabilities for creases. These functional identifiers change a rounded edge to a sharp one. By defining areas for smoothing and for sharpening, rather than having to draw the details, reduced the necessary data and still permits great amounts of detail in the final render.

In “Wall-e” the effective use of the techniques allows the robots and other characters to be highly detailed without needing massive amounts of artist time and effort. The reductions in data volume also reduce render times and storage needs.

The technology is patented but Pixar has released all of the IP in open source. More on the technology is available at: http://graphics.pixar.com/opensubdiv/ and an open subdivision library is available at: https://github.com/PixarAnimationStudios/OpenSubdiv