Final Activity Report Summary - ScalableGlobIllum (Scalable, Flexible Global Illumination and Natural Lighting)
The creation of realistic computer graphics images involves complex computations to simulate the interaction of light and surfaces. Global illumination algorithms compute physically-based and accurate solutions to this problem. Due to the inherent complexity of this task, the computation time is tremendous and, as a consequence, different approximate algorithms are applied in interactive applications. These algorithms are typically based on different simplifying assumptions, and thus neither convergence with increasing computation time nor continuous transitions to more accurate methods can be achieved.
In this research project a scalable and flexible global illumination framework was developed. The key contribution was a reformulation of the process of light transport. Instead of computing mutual visibility of surfaces, which was the most time consuming task in all global illumination algorithms, a concept of negative light was developed which treated visibility implicitly (refer to publication 1).
By computing global illumination solutions with this implicit visibility we could perform the computations on fast parallel architectures, such as graphics hardware, and adapt the accuracy of the lighting solution to the available computation time. The accuracy of the global illumination ranged from a plausible approximation, achieved at interactive speeds, to a solution comparable to the one obtained from traditional, physically-based, off-line algorithms. With our new solution, a single algorithm computed a solution which was as good as the computation time allowed, without switching between distinct, unrelated algorithms which were not designed to communicate with each other nor guaranteed continuity in quality.
Our new framework was also flexible in terms of scenes representation. It treated traditional polygonal description of surfaces and volumetric representations through clustering. In a similar context we further developed a new data-structure for storing surface data, e.g. lighting information and textures. The novel solution, called the TileTree data structure combined the flexibility of volumetric data structures with the efficiency of image-based representations (refer to publication 2).
We also explored a research area which had recently attracted much attention in the computer graphics community, i.e. perceptual rendering. This concept was inspired by the observation that the human visual system, and thus the human perception, was affected or influenced by local features in images, such as spatial and contrast masking. This could be considered during the creation of computer graphics images. Image regions could be identified in cases a less accurate rendering was tolerable or not noticeable at all. Our research in this area introduced perceptual rendering to interactive computer graphics for the first time (refer to publication 3).
In order to create immersive virtual worlds we also investigated the high-quality modelling of sound propagation and scattering in complex environments. We developed a method which efficiently computed a first-order scattering approximation of sound propagation. The algorithm was designed to exploit the computational power of graphics hardware, which was additionally used for the creation of images of the virtual world and for sampling the scenes surfaces, thus sharing resources and creating synergy effects.
The most prominent publications of the project were the following:
1. Implicit visibility and antiradiance for interactive global illumination, by Carsten Dachsbacher, Marc Stamminger, George Drettakis and Fredo Durand in the proceedings of the Association of Computing Machinery (ACM) transactions on graphics ACM SIGGRAPH 2007
2. TileTrees, by Sylvain Lefebvre and Carsten Dachsbacher in the proceedings of the ACM SIGGRAPH Symposium on interactive three-dimensional graphics and games in 2007
3. A novel perceptual rendering pipeline using contrast and spatial masking, by George Drettakis, Nicolas Bonneel, Carsten Dachsbacher, Sylvain Lefebvre, Michael Schwarz and Isabelle Viaud-Delmon in the proceedings of the Eurographics Symposium on Rende.
In this research project a scalable and flexible global illumination framework was developed. The key contribution was a reformulation of the process of light transport. Instead of computing mutual visibility of surfaces, which was the most time consuming task in all global illumination algorithms, a concept of negative light was developed which treated visibility implicitly (refer to publication 1).
By computing global illumination solutions with this implicit visibility we could perform the computations on fast parallel architectures, such as graphics hardware, and adapt the accuracy of the lighting solution to the available computation time. The accuracy of the global illumination ranged from a plausible approximation, achieved at interactive speeds, to a solution comparable to the one obtained from traditional, physically-based, off-line algorithms. With our new solution, a single algorithm computed a solution which was as good as the computation time allowed, without switching between distinct, unrelated algorithms which were not designed to communicate with each other nor guaranteed continuity in quality.
Our new framework was also flexible in terms of scenes representation. It treated traditional polygonal description of surfaces and volumetric representations through clustering. In a similar context we further developed a new data-structure for storing surface data, e.g. lighting information and textures. The novel solution, called the TileTree data structure combined the flexibility of volumetric data structures with the efficiency of image-based representations (refer to publication 2).
We also explored a research area which had recently attracted much attention in the computer graphics community, i.e. perceptual rendering. This concept was inspired by the observation that the human visual system, and thus the human perception, was affected or influenced by local features in images, such as spatial and contrast masking. This could be considered during the creation of computer graphics images. Image regions could be identified in cases a less accurate rendering was tolerable or not noticeable at all. Our research in this area introduced perceptual rendering to interactive computer graphics for the first time (refer to publication 3).
In order to create immersive virtual worlds we also investigated the high-quality modelling of sound propagation and scattering in complex environments. We developed a method which efficiently computed a first-order scattering approximation of sound propagation. The algorithm was designed to exploit the computational power of graphics hardware, which was additionally used for the creation of images of the virtual world and for sampling the scenes surfaces, thus sharing resources and creating synergy effects.
The most prominent publications of the project were the following:
1. Implicit visibility and antiradiance for interactive global illumination, by Carsten Dachsbacher, Marc Stamminger, George Drettakis and Fredo Durand in the proceedings of the Association of Computing Machinery (ACM) transactions on graphics ACM SIGGRAPH 2007
2. TileTrees, by Sylvain Lefebvre and Carsten Dachsbacher in the proceedings of the ACM SIGGRAPH Symposium on interactive three-dimensional graphics and games in 2007
3. A novel perceptual rendering pipeline using contrast and spatial masking, by George Drettakis, Nicolas Bonneel, Carsten Dachsbacher, Sylvain Lefebvre, Michael Schwarz and Isabelle Viaud-Delmon in the proceedings of the Eurographics Symposium on Rende.