March 19, 2013, GPU Technology Conference, San Jose, CA— Shailendra Mathur from
Avid described a new toolset that allows editors to smoothly transition their monoscopic storytelling skills to the stereoscopic format. The tools enable users to take advantage of existing assets and workflows.
One goal was to preserve editing and data management functions, operational speed, and image quality from existing tool flows. The trends of increasing resolution, higher frame rates, and higher bit depths of the video, create problems for many directors. In addition, the new workflows require an increasing number of views, while both monoscopic and stereoscopic images are required as the deliverable.
The increased elements within the scene require a data model that can manage multiple views and resolutions while supporting common editing and data management functions and workflows. The advent of plenoptic or light field cameras further increases demands for the editing suite.
Light field cameras do not require a parametric model to manage all the different possible light paths from the scene. This multiview requires sampling across time, space, and color to capture the final image. In the same way, a multiple camera shoot also requires color, spatial, and temporal synchronization to capture the multiple images within a scene. As a result, they created a new editing system that can simultaneously handle the myriad channels, resolutions, and streams from various inputs.
To achieve these capabilities, a proxy editing and multi-mastering flow starts by transcoding the images to side-by-side 3-D for further processing. The output can be full frame 3-D and also mono with synchronized audio. The data model groups multiple views, synchronizes those views, arranges resolutions within those views, and resamples the scene based on output context.
These multiple views require much more processing for codecs, visual effects, and graphics. As a result, the tools were optimized for a heterogeneous compute environment comprised of multicore CPUs and GPUs. Unfortunately, heterogeneous compute usually requires multiple pieces of code for each compute domain, and dynamically balancing across compute domains. By splitting shell optimize processes and compute optimize processes and providing a single API contract, the domain-abstracted player eliminates many of these contentious problems.