Technical Report: tr-ri-13-332 Pull-based Dataflow for Multi-Volume Visualization
Stephan Arens, Matthias Bolte, Gitta O. Domik · 2013
(a) (b) (c) (d) (e) Figure 1: A visualization of multimodal volume data exemplifying the problem with sampling patterns. (a) Two volumes, one containing a cube and the other containing a sphere, shown in a multi-volume rendering. (b) - (e) Logical combination of the two volumes, illuminated using the cube's gradient (b), the sphere's negative gradient (c) and the true negative opacity gradient of the logical combination (d). (e) Three special sampling patterns (deformation, gradient, ray-casted shadow) connected correctly in series. Abstract In multi-volume visualizations it is not only important to display several fields and modalities, but also to combine their provided information on a per sample basis in order to be more expressive. This significantly complicates the computation of the negative opacity gradient for illumination or, generally speaking, the computation of all sampling patterns. In this technical report we propose the concept of pull-based dataflow for volume rendering to solve this processing problem, avoiding case handling that is necessary elsewhere. This technique sim- plifies the specification of the dataflow considerably and is well suited for GPU-based volume rendering without performance loss. In fact it improves the rendering performance in many cases massively as it is able to skip unnecessary computations in contrast to the normally used push-based dataflow. To show its applicability and usefulness in the important field of rapid prototyping, we replace the evaluation model of a graph-based frame- work for GPU-based multi-volume rendering by the proposed technique and compare exemplary visualizations in terms of usability and performance. We explain how to implement this concept in an easy way using the shaders' preprocessing stage and call-by-value-return evaluation strategy.