Compressed Data Structures for Real-Time Rendering

• Applicant: Professor Dr.-Ing. Dieter W. Fellner, since 5/2022
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing ; Data Management, Data-Intensive Systems, Computer Science Methods in Business Informatics
• Term: from 2018 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 407714161

Traditionally, real-time rendering uses rasterization for image generation. These algorithms split each individual primitive, usually each triangle, into a separate list of fragments. The advantage of this approach is that the complexity increases only linearly with the number of triangles or fragments. However, interactions between triangles, such as global illumination effects, cannot be handled and have to be approximated. Physically based rendering algorithms and hardware on the other hand are far from being able to generate images in real-time.The reason is the sheer amount of computations required to solve the area integrals of the rendering equation numerically. Even though there are existing approaches to either speed up the convergence of the numerical integration or to approximate the solution, including approaches that take neighboring pixel or samples into account, they usually do not that the human visual system into account. Therefore, only scenes with up to a million triangles, a handful of point light sources and less than 10 light paths per pixel can be handled in real-time. Even when filtering the generated image, it is far from being perceptually accurate. The goal of this research project is to enable perceptually accurate path tracing algorithms in real-time image generation.For this, we need a perceptual quality metric that can be evaluated in real-time and takes the image generation process into account. It forms the basis for a post-sample reconstruction algorithm that will be able to generate images without noticeable artifacts in real-time. Additionally, we need an output sensitive path tracing algorithm whose computational complexity only depends on the visual complexity of the output image in order to handle scenes with a multitude of area light sources and millions of triangles. We also require dynamic, compressed acceleration structures that can be generated in real-time in order to not only generate still images but a whole sequence of frames.On the hardware side, we need to develop a programmable path tracing hardware and work scheduling approaches. For maximizing the available memory bandwidth, we will also research hardware based compression algorithms for triangle meshes and textures that allow for random access both for reading and writing.Together, we will aim to both reduce the complexity of the computations that are required and to increase the amount of computational resources that are freely available by several orders of magnitude. This will enable path tracing algorithms to generate images in real-time even if scene complexity is increased by an order of magnitude compared to today.

• DFG Programme: Research Grants
• Ehemalige Antragsteller: Professor Dr.-Ing. Michael Goesele, until 2/2019; Dr. Stefan Guthe, until 5/2022