Real-time collaboration has become ubiquitous. Google Docs, Dropbox Paper, Microsoft Live Share are example of these systems used by millions of people worldwide. In this dire times, we have relied on collaborative tools to use us working remotely and safely. At this time though, many 3D design applications do not have collaborative features.
All Computer Graphics applications need to represent complex 3D scenes that model the cameras, shapes, materials, lights and layouts of real-world environments. The two most popular data structures used to encode this data are scene graphs and node networks. The former encodes scene data as direct acyclic graphs, where node properties can be inherited in the hierarchy. The latter are forms of data flow graphs that describe scenes procedurally. These scene representations have served the graphics community well over the years. But, as we will discuss in the project description, collaborative editors cannot be built directly on scene graphs or node networks.
Inspired by Git, we propose to investigate SceneHash, a content-indexable data structure for 3D scenes. In scene hash, all scene elements, from large objects to shape vertices, are referred to by their content hashes. Scene data structures can be built using these hashes in leu of explicit references. Since in this representation, a local edit changes the identity of the containing object, tracking changes in large data structures becomes easy. This in turns means that scenes can be concurrently edited at their finest granularity, while ensuring that the scene data structures remain well-formed after each edit.
We will use this data structure to (1) build a prototype 3D collaborative editor where scenes can be seamlessly edited, (2) speed up typical graphics editors that use concurrent processes to provide an editing-while-rendering workflow (3) seamlessly provide a scalable and robust caching system for large data flow graphs used in procedural modeling and shading.
We believe this project will make a significant impact for three reasons.
First, ours will be the first demonstration of seamless collaborative editing in 3D environments where scenes scale to large complexity and where edits are arbitrarily large or small. We know of no other system design, both in academia and commercially, that can support these edits.
Second, while we focus on collaborative editing, we believe that our scene data structure is useful in other computer graphics applications, such as animation and modeling-while-rendering systems. The shortcoming of today's scene graphs have been already observed when bringing interactivity at scale [1,2,3,4,5]. Many new data structure are being explored to address the needs of modern graphics applications. In our opinion, SceneHash can address some of these shortcoming since it can represent variations of large scale environments very efficiently.
Finally, to the best of our knowledge, our work is the first to explore the scalability of read-only context-indexable data structures. While we will focus only on graphics data, it is possible that our work will have impact on other editing domain that today are not made collaborative due to data size.
Besides improving upon the academic literature, we strongly believe that the real impact of our work and to shoe that Seamless collaborative 3d design is possible at scale. We hope this will foster solutions in this space that will eventually allow designers, artists and architects to work remotely and safely, just like today is possible for office workers and computer engineers.
[1] Georgiev et al. Arnold: A Brute-Force Production Path Tracer. ACM Trans. Graphics, 2018.
[2] Christensen et al. RenderMan: An Advanced Path Tracing Architecture for Movie Rendering. ACM Trans. Graphics, 2018.
[3] Burley et al. The Design and Evolution of Disney¿s Hyperion Renderer. ACM Trans. Graphics, 2018.
[4] Unity Inc. Unity Data Oriented Technology Stack. 2019.
[5] Pixar Inc. Pixar Unified Scene Description Documentation. 2019.