I originally wanted to link to all the pdf’s of the paper talks I went to but I don’t think I can because of publishing reasons. If you have a subscription to the ACM Digital Library, you should be able to access them.

Computer Animation Festival

Today was a really good day. Attending the paper talks were much more interesting and helpful than the classes. It’s also much easier to pay attention because the talks are only half an hour each as opposed to three hours. The highlight of today was most likely the Computer Animation Festival screening. Some of the works were a bit abstract but there were some real gems there. My favorites, in no particular order, were:

• Oktapodi
• Bolides
• Mauvais Rôle
• Carbon Footprint
• Our Wonderful Nature

Paper Talks

I spent the rest of the day, attending paper talks that I had took notice of during the Fast Forward paper session. The two ones in the morning were on Image Collection and Parallelism and the one in the afternoon was on Hair and Realistic Rendering. Here are the highlights from each:

Image Collection

The Image Collection talk probably had the coolest demos. One of the talks seemed to be an extension of Microsoft’s earlier work with Photosynth. This new work was more focused on HCI than graphics but it still had quite a bit of interesting, technical details. The link he gave out at the end of the talk is here. Check it out!

Another really neat paper was another extension of a previous work. It was seam carving for video resizing. This is a technique that allows for content aware resizing rather than just rescaling. They call it, “re-targeting.” Instead of just scaling everything, this algorithm looks for “boring” or stale parts of the image and removes that. This new paper had improvements on the previous technique as well as some necessary adjustments to work in the time domain.

Parallelism

There seemed to be a whole lot of excitement around Larrabee, a new many-core x86 architecture. The room looked almost as full as when Ed Catmull was speaking. I think the impressive author list may have had something to do with that. The goal of Larrabee was to provide

CPU Programmability with GPU Parallelism

It basically replaced most of the fixed logic hardware in graphics cards with x86 cores. This allows developers to better utilize hardware to meet their particular game/application’s need. It made the bold claim that “software was the new hardware” and that Larrabee could dynamically load balance all the stages of the pipeline for better utilization. It seemed very hopeful, but its current status was also rather vague.

There was a paper on generating Poisson disk samples in parallel which had a pretty neat idea behind it. I like these papers the most because speakers can very clearly provide the intution behind that idea and leave most of the details out of the talk. Sadly, the main idea behind this paper hasn’t been proved but his results proved promising in practice.

The very last paper addressed how to deal with image processing on data that could not fit into memory. It seemed a bit out of place with the rest of the parallelism talks as it talked more about pipelining than parallelizing. Still, it seems like a relevant talk. In the end, he showed some huge images of Seattle that he was working with. I think the data was something like 50gb which is just a ton of data. Also, I think this paper was the best presented talk I’d seen.

Hair and Realistic Rendering

These talks also featured some really nice eye candy. The first there were about hair, and the last one was about general rendering. I caught the tail end of the first talk of the session which was about generating hair geometry and image based rendering techniques. The results from this paper were outstanding, though it seems like a very complicated approach.

Next up was a paper I wanted to read and implement for my cs348b final project. They talked about efficient ways to simulate multiple scattering which in many ways was similar to our nebular project. However, they addressed this problem using spherical harmonics (no surprise there). I thought this presentation was really well done and the speaker explained the key points of the paper very clearly. Despite that, I was still a little confused afterwards and will probably read the paper later.

The last talk on hair was pretty good. Their goal was to render a “good enough” approximation to hair, in real time. I must say, it worked pretty darn well. In the end, they had a reasonably decent looking rendering at about 14fps. The previous paper had mentioned how with their technique, multiple scattering was no longer the bottleneck, but instead single scattering was dominating the runtime. This paper suggested that there were many ways to speed up the single scattering stage (as well as the multiple scattering stage). It’d be interesting to put the two techniques together (if they were compatible, that is).

I was really looking forward to this last talk but the speaker was not as good. The content was more directed towards sampling and reconstruction which I don’t understand completely. There were interesting ideas in the paper; however, most of the theoretical details were deferred to the paper, which I’ll probably have to read now. In fact, I should probably go and read most of the papers from today. Mostly, they were that interesting.

And so, that was my second day here at SIGGRAPH. I got to grab dinner with an old friend from COSMOS who I hadn’t seen in about 5 years. We caught up for a while and then I got dropped off back at the Computer Animation Festival to catch the tail end of the screening of Frederic Back’s “The Man Who Planted Trees.” John Lasseter just started to interview with Back when my brother called to pick me up. I would’ve liked to hear the rest of that interview with Back, but he was pretty difficult to hear anyways.