This year, sept 3-7, I had the opportunity to attend the Eurographics Conference in Prague, which has gathered more than 500 Computer Graphics researchers and enthusiasts. This certainly makes this conference the biggest of its kind in Europe. The week has been stuffed with a lot of scientific presentations (tutorials, papers), non-game and game industry related presentations, and a bunch of social events (welcome reception, poster+beer party, diner conference). So that every day was really fully booked. Here are some comments on some of my favourite presentations:
Tutorial: Populating Virtual Environments with Crowds
This is the tutorial we presented, we got some nice feedback from some students working in related fields.
Tutorial: Computational Photography
Interesting tutorial on a very trendy domain. Computational photography researchers want to change the conceived idea of pictures as static array of pixels. More and more papers are presented each year at the major conferences. From the speakers, it groups a lot of sub-domain like image enhancement, HDRI, cameras, big images, video processing, ...
Tutorial: Inverse Kinematics and Kinetics for Virtual Humanoids
All IK methods were compared and studied, jacobian, analytic, pseudo-inverse, CCD, hybrid. At the end of the tutorial, we had a very clear idea what method to use for a specific applications. For example, for crowds, a hybrid CCD + analytic method shown in a real-time demo with 500 characters was really convincing.
Paper: Omni-directional Relief Impostors
Next-gen impostoring. Nice demos. With this technique there is no more popping artefacts when rotating around the impostor because the mesh is very nicely reconstructed. Comparing the method with impostors used in crowds, it is much more costly to render. I would have liked some comparison with usual mesh rendering to really know when it's cheaper to render the impostor (complexity on the number of pixel to render).
Paper: Efficient Rendering of Interreflections for Dynamic Scenes
Nice interactive global illumination results for scene of a few objects. Usual huge preprocessing time for these kind of effects.
Paper: Adaptive Space Deformations Based on Rigid Cells
Impressive method to deform arbitrary shapes while preserving details. The model is first adaptively voxelized.
Paper: Real-Time Simulation of Thin Shells
Cool real-time thin shell physics (leaves, cowboy hats, dinosaurs).
Paper: Crowds by Example
Nice navigation behavior with collision avoidance results for crowds of less than 100 characters based on manually extracted video data. I like the couples forming temporarily and also characters naturally stopping to watch shop windows. But the method is not real-time (~1 hour for a 2 min. animation) and also it's not possible to control the simulation without adding other video shots.
Paper: Pedestrian Steering for Crowd Simulation: A Predictive Approach
Low-level real-time pedestrian collision avoidance steering based on motion captured data (impressive crowd mocap videos). Scalable for up to 1000 characters. Though not very convincing crowd rendering, the steering method is efficient and very convincing.
Paper: Soft Articulated Characters with Fast Contact Handling
Cool demo of a duck jumping on a running deer. Nice results but still a bit slow to integrate it in a game where a lot of other computations are to be done.
Paper: Context-Aware Skeletal Shape Deformation
Next-gen skinning to reproduce bulging effects not present with the traditional method. Not yet suitable for crowds, maybe conceivable with a GPU implementation.
Industry Presentation: Collada
Collada 1.5 will include among others IK and cloth simulation.
Saturday, September 8, 2007
Eurographics 2007
Posted by
Jonathan Maïm
at
11:33 AM
Labels: computer graphics
Subscribe to:
Post Comments (Atom)

0 comments:
Post a Comment