Dynamic Benchmark for Graphics Rendering
2021, Alanko, J., Master of Science thesis, Tampere University.
Abstract: Most graphics rendering algorithms used in both animated feature films and real time games can enjoy the performance and quality boost that comes with temporally reusing previous computation. However, there is a lack of proper rendering benchmarks that would allow people to have detailed and objective comparisons between different temporal methods. Currently, very slowly moving cameras, improper scenes, and animations are used, which results in an unequaled playground for comparisons, having an obvious bias towards the proposed novel methods.
In this thesis, we describe a framework that can be used to capture 3D animations out of interactive scenarios and compile them to a dataset that is compatible as a dynamic benchmark. The capturing framework is used in the creation of two datasets: EternalValleyVR and EternalValleyFPS. We verify the quality and the dynamic challenge these datasets put on the algorithms. By surveying the input features used in the state of the art temporal reuse algorithms, we form metrics of change in features that happen throughout the animation. The proposed dynamic benchmarks are shown to surpass the previously released animations in temporal complexity.