| Event Home | Register | Speakers | Sponsors | Presentations |

Yoky Matsuoka

Yoky Matsuoka presented a high level survey of her work in neuroroboticsn -- the very forefront of human machine interfaces.  Her work focuses on haptic control devices that interface directly with the human nervous system or indirectly via brain wave analysis to enable human beings to interact with and control robotic devices and virtual environments at the speed of thought.

Here are three example projects presented:

Understanding neural control and biomechanics of human hand movement

Using neural signals to control external devices

Technique to extract neural adaptation state through haptic interaction

 

Steve Seitz

Steve Seitz presented several exciting new developments in 3D reconstruction and visualization from photos. Building on their recent "Photo Tourism" work, the 3D technology behind Microsoft's Photosynth, Seitz gave the following demos:

3D slide shows created from your vacation photos

Tag transfer technology that transfers image annotations to all other photos of the same scene

High resolution 3D meshes from casual photo collections

Building a 3D model and visualization of the interior of your house.

Progress on reconstructing entire cities ("Building Rome in a Day") from hundreds of thousands of photos on Flickr

 

Zoran Popovic

Zoran Popovic presented latest advances in realtime simulations for game environments, as well as a framework for authoring avatars and NPCs that move and behave intelligently.  He pointed out that we are at the cusp of a new age of character modeling where for the first time we can have believable and compelling virtual characters.  He also pointed out that the new framework makes the authoring sufficiently easy that it can be done by the game players themselves, thus transforming interactive animations in the games into a new storytelling media for everyone with a PC.

Below is a list of demonstrated projects:

A project on modeling high-fidelity fluids 5 orders of magnitude faster than the standard state-of-the-art offline methods

A model of crowds that can realistically simulate city-size crowds in realtime (licensed by EA)

An authomatic method for creating complex human locomotion controllers

A first model of fully simulated realtime human motion that is indistinguishable from motion capture

A way to evolve creatures to bets achieve locomotion tasks.  Sort of like Spore's Creature Creator, but full dynamics and allowing the bones and morphology of the creature to change to be a better runner for example.

Not published yet (under review): a character AI framework that allows characters to intelligently play competitive games such as fighting, by accounting for the opponents abilities and figuring out the best strategy.

Not published yet (under review): a way to create very hard AI navigation controllers automatically in such a way that it can run optimally on many architectures from XBOX 360 to a mobile phone, thus drastically reducing game development costs

 

Brian Curless

Brian Curless discussed work in the areas of human body and face shape modeling and computational photography:

Spacetime Faces: High-Resolution Capture for Modeling and Animation
We have developed a scanner that captures high resolution models of human faces at video rates, enabling 3D facial expression modeling and animation.

The space of human body shapes: reconstruction and parameterization from range scans
Starting from a database of detailed human body scans across a population, we develop statistical models of shape variation, including predictions of how an individual's body will appear as weight changes.

Parallax Photography: Creating 3D Cinematic Effects from Stills
From a few photographs around desired viewpoints, we demonstrate a method for creating video slide shows that subtly yet effectively reveal the 3D nature of the original scenes.

Using Photographs to Enhance Videos of a Static Scene
Enhancing and Experiencing Spacetime Resolution with Videos and Stills
In these two projects, we combine high resolution photographs with low resolution videos to enable a range of applications that include high resolution video reconstruction, editing videos in some cases as easily as one would edit a photograph, and shifting time to recover that moment missed when shooting photographs.