This sheer size of this convention never ceases to impress me – about 10,000 people spread over 50+ rooms and exhibit halls in the 900,000 square feet convention centre.
Room 408 AB holds about 800 people. This shot taken between sessions.
The latest and greatest academic papers on computer graphics along with talks, workshops and courses in anything related to the subject as well as an animation festival, trade show and job fair. Producers and directors presenting clips of the most impressing upcoming CG movies – pretty much any major movie is a CG movie these days – and details of the production.
As a result it has about 6-8 events going on at any one time, so it’s impossible to follow it all. Luckily, they record all the sessions, so I’ll get them all on DVD in a couple of weeks.
The conference has shrunk since I first attended in the ’97 where it had almost 50,000 attendees in the same location. In a post-Avatar world it’s obvious that most of the core problems of photo-realistic rendering have been solved. It was tale-telling that the Jury award in the animation festival went to a short that was first and foremost a great story, told with average animation.
It’s no longer about what we’re able to do technically, at least for movies, that’s pretty much anything except for some edge-cases. It’s more a question of how much content we can afford to create and to what quality level. That’s where the budgets of all these movies are going.
As anybody that has sat down in front of Maya or 3DS Max can attest to, creating 3D content has a steep learning curve and takes much longer that you thought when you first tried to do it. I don’t think any of those tools will go away, but it’s time to generate a new class of tools that focus on making it easier to tell a story with basic figures and then add detail as time and budget allows.
Think of the difference between Quark Xpress and Word – Quark Xpress is clearly more powerful and a better tool for a professional type-setter, but Word can do a pretty good job and is far more accessible.
That said there is still a lot of great new research going on:
Much of the work this year focused processing 2D images in interesting ways. Now that almost everyone has a video camera on them all the time, this field is getting renewed interest. From automated compositing of crowd-sourced video into multi-camera event footage to using user input about favorite pictures to automatically to teach your camera to take pictures in your preferred style. Here’s a list of my favorites:
- Image Deblurring Using Inertial Measurement – this can use inertial data from your blurry cell-phone snaps to de-blur them into tri-pod sharpness (or at least a lot closer).
- Unstructured Video-Based Rendering: Interactive Exploration of Casually Captured Videos – turning crowd-sourced event coverage into 3D-based cut scenes.
- Video Tapestries with Continuous Temporal Zoom – turn any movie into a zoom-able tapestry
- Personal Photo Enhancement Using Example Images – automatically adjust process your photos to look more like your favorite pictures.
On the more hard-core 3D side, one of the best papers was about using a new atomic object for simulating deformable objects – the Elaston. This is really brilliant stuff that greatly simplifies a lot of simulations of rubbery or otherwise deformable objects. There was also some great work (using Elastons) in realistically simulating knitwear and other types of cloth. For all the great work here we still have a ways to go before we get the full range of clothes and hair to render well – we’ve nailed spandex, crew-cuts and pony tails. Here’s a list of the best ones, in my opinion:
- Unified Simulation of Elastic Rods, Shells, and Solids – Introducing Elastons: an elegant atomic structure for handling elastic objects of almost any kind.
- Micropolygon Ray Tracing with Defocus and Motion Blur – optimizing rendering to take into account that most images (even real-time) now have more polygons that pixels, all our existing techniques assume that polygons have lots of pixels.
- Rigid-Body Fracture Sound With Pre-computed Soundbanks – yes, this one allows realistic-sounding shattering of objects in real-time.
- Interactive Hair Rendering Under Environment Lighting – this is really good, for polyester wigs.
- Subtle Gaze Direction – this explores using an camera to track eye-focus of viewer and imperceptibly direct their focus to specific parts of the picture by making small distortions outside of the focal area. It’s amazingly effective and ties in well with the development of devices that look back at you.
- Feature-Based Locomotion Controllers – this allows biped models to move realistically from A to B, either jumping, avoiding or focusing on obstacles, using a learning based reward model.
- Comprehensive Biomechanical Modeling and Simulation of the Upper Body – this is the best human simulation model I’ve seen. They’ve modeled (and made publically available) a human torso that simulates all the muscles, bones, cartilage etc. in near-real time.
- Efficient Yarn-Based Cloth With Adaptive Contact Linearizations.
The trade-show was showing off a number of new products. It’s clear that object generating machines continue to get cheaper and better:
A rose by any other name… this one is all plastic and came out of a desk-top sized machine.
Other things that caught my attention in the tradeshow was a company called EyeTech with an effective webcam-based eye-tracker that allowed pretty precise eye-tracking after a 10 second calibration.