Abstract
Virtual reality offers unique affordances that can benefit the scientific discovery process. However, virtual reality applications must maintain very high frame rates to provide immersion and prevent adverse events such as visual fatigue and motion sickness. Maintaining high frame rates can be challenging when visualizing scientific data that is large in scale. One successful technique for enabling interactive exploration of large-scale datasets is to create a large image collection from a structured sampling of camera positions, time steps, and visualization operators. This paper highlights our work to adapt this technique for virtual reality, and uses two authentic scientific datasets - a) a large-scale simulation of cancer cell transport and capture in a microfluidic device and b) a large-scale molecular dynamics simulation of graphene for creating extremely low friction interactions. We create a collection of omnidirectional stereoscopic images (three-dimensional surround-view panoramas), each of which captures all possible view angles from a given location. Therefore, virtual reality devices can always render local movements at full frame rates without loading a new image from the collection.