Sunday Spent the day prepping for my portion of the Introduction to Direct3D 10 Course. My portion was originally billed as Film applications of DirectX 10 technology, but as the course notes developed, it was clear that what I really needed to talk about was the production pipeline in general. In order to accelerate the adoption of hardware solutions across the pipeline, I suggest that it is necessary for both the production pipeline, and the rendering pipeline to take on newer forms. I coined the phrases ad hoc, structured, and modular to describe the typical games pipeline, the typical film pipeline, and the proposed new form. I match the pipelines to the typical rendering pipelines I discuss in an earlier entry. I was pleased to see people liked my terminology, and I heard it a few times during the show. In discussions later with Bay Raitt, he corrected my impression that a modular pipeline and a shader based renderer are incompatible, through intelligent caching techniques. Monday Urban Design and Procedural Modeling was awesome, although I ducked out after the Need For Speed presentation. Pascal Mueller demoed the CityEngine tool set live for the first time in public; truly inspirational. "Surf's Up": The Making of an Animated Documentary was the second highlight of the morning. I particularly enjoyed the presentations on creating waves, breaking waves on the beach, and the discussion of techniques to sell the documentary look. Particularly interesting was their solution for cameras dipping below the water line whilst filming an object above the water. They encased the camera in a virtual air-tight housing, and put the clipping plane out in front of the lense. Depth of field blurred the intersection of the water surface with the housing at the clipping plane, and they comped in green, with suspended particle systems, and god rays under the water line. An overlay with some specular scratches and a few water drops at the clipping plane, and slight vignetting sells the illusion. At the end of the day "Happy Feet": Thawing the CG Pipeline was similarly instructive. I was slightly penguined-out at this point. Tuesday The system jointly developed by MIT, ILM, and Tippett, The Lightspeed Automatic Interactive Lighting Preview System was very good, although it wasn't new - been watching these guys build it for the last little while! Matrix Row-Column Sampling for the Many-Light Problem is very promising as we look into much more sophisticated GPU based lighting solutions. Plushie: An Interactive Design System for Plush Toys looked like an excellent evolution of Teddy style sketch based modeling, with a real world practical application that I'd love to get into the hands of my kids. Curl Noise for Procedural Fluid Flow was one of those papers that made we want to rush back to my hotel room and start coding. Wednesday Found myself in the Spiderman 3 sketch session, which had a really unexpected bonus at the end. Artistic Direction of Foliage was a presentation of the art directed procedural tree modeling system developed for Kung Fu Panda. Eikonal Rendering: Efficient Light Transport in Refractive Objects presented a solution to a problem that I have long been interested in - the problem of modeling the lighting problem as a wave based problem instead of a particle based (ray based) problem. Most of the investigation I've done to date has focussed on inverse transforms of wavefront propagation in frequency space; this approach having the drawback of requiring huge amounts of memory to model tiny packets of space. The method presented today instead presented the problem as wavefront propagation in time which brings down the amount of data required by an entire dimension. Lovely! Computing the Scattering Properties of Participating Media Using Lorenz-Mie Theory was extremely refreshing. If you've read some of my previous articles, you'll know that the thing I like about Siggraph is being challenged with approaches that I won't be able to use in real time for years. I like the challenge of finding future real time trends. The refreshing bit about this talk was when Jensenn revealed under questioning that the images of milk took four days to render. Now that's the Siggraph I love! Harmonic Coordinates for Character Animation. Okay I admit it, you win. Completely new, retrospectively obvious, paradigm shifting approach. Genius. Automatic Rigging and Animation of 3D Characters. This is sweet. It isn't harmonic coordinates, but it works really, really well. And to top it all off, they released the code, sample app, and test motion data for free on the Internet. Yowza. Thursday Managed to sneak in Image-Based Procedural Modeling of Facades, Image-Based Tree Modeling, and Approximate Image-Based Tree Modeling Using Particle Flows. Yes, it's true. I'm a sucker for procedural modeling. Of course, LucasArts and ILM: A Case Study in Film and Game Convergence was a winner. The Spore sketches were instructive. I was probably the last developer in creation who hadn't yet seen the end user tools in use. Very inspiring. There were a few other things I saw which escape me just now. Will update this post after grinding through my notes.