Meshless Animation Framework

Abstract

This report details the implementation of a meshless animation framework for blending surfaces. The framework is meshless in the sense that only the control points are handled on the CPU, and the surface evaluation is delegated to the GPU using the tessellation shader steps. The framework handles regular grids and some forms of irregular grids. Different ways of handling the evaluation of the local surfaces are investigated. Directly evaluating them on the GPU or pre-evaluating them and only sampling the data on the GPU. Four different methods for pre-evaluation are presented, and the surface accuracy of each one is tested. The framework contains two methods for adaptively setting the level of detail on the GPU depending on position of the camera, using a view-based metric and a pixel-accurate rendering method. For both methods the pixel-accuracy and triangle size is tested and compared with static tessellation. Benchmarking results from the framework are presented. With and without animation, with different local surface types, and different resolution on the pre-evaluated data.