Pipeline Developer Forum

Forum - August 2019 | Thursday, Aug 8, 2019

As always, a big thank you to everyone who came out this month, especially so soon after the long, tiring week that is Siggraph!

Siggraph 2019 Post-Mortem

This month, we sat down to discuss what we saw, and didn’t see at this year’s Siggraph conference. This is by no means a complete coverage of the conference, and only represents the perspectives of who made it to the forum. Also, as usual, the discussions are left to weave and track however they want to, so there may have been topics that we had hoped to bring up but didn’t, in addition to topics that we never meant to cover, but did anyway!

The Pipeline Conference

This year was the inaugural event for The Pipeline Conference, which spawned out of the ever-popular “Global VFX Pipelines BoF"Birds of a Feather - Siggraph Session which has been running for many years. This took place at the very beginning of Siggraph week and is particularly relevant to our group so seemed like a good place to start the forum.

Overall, those that attended definitely enjoyed the day. There were some very interesting and cool talks that went on. Notably, the Nvidia Omniverse presentation was an interesting take on real-time workflows and the power of USD as a standard format for communicating scene data. Everyone seems a little uncertain exactly what is in it for Nvidia, but certainly relevance in the industry and the ability to disseminate technology which can leverage their hardware can be seen as beneficial in some way.

The keynote was another point of discussion, and was well praised. Hank Driskill (CTO, Blue Sky Studios) spoke about the state of our craft from a cultural perspective, questioning the benefits of closed-source pipelines. Although I won’t do it justice here, it was a great reflection on our purpose as a community, and responsibilities as members of such a wonderful group.

We didn’t speak too in-depth about any other particular topics at The Pipeline Conference, but did reflect on the day as a whole. In conclusion, everyone who attended enjoyed the day, though it seems like the hope was to see it be more of a giant extension on the BoFBirds of a Feather - Siggraph Session Siggraph session rather than a day of presentations. There was little material of real value in our day-to-day, and as amazing as it is to dream about the future and see people exploring the bleeding edge there was a hope that we would talk about the aspects of current-generation pipelines. Cool, but not applicable being the general theme.

DigiPro

At this point, we are still not getting in to Siggraph proper! DigiPro is another one-day conference that takes place just before Siggraph and has been running very successfully for many years. These sessions are more focused on “digital production” but are sometimes still relevant to production pipelines.

This year there was a lot of eye candy, some USD-based technology for doing sequence-based lighting, efforts by The Foundry to make machine learning more accessible inside DCCDigital Content Creation apps, a new (re-)lighting tool focused on making the whole process fun and enjoyable, as well as a few others.

We also discussed the session on Dreamwork’s aptly-named “Insanity” tool, which essentially aims to provide something like real-time, collaborative lighting by leveraging an enormous compute cloud scaled up for such a task. Pushing out full-quality renders this way takes a lot of resources, and is expensive but the idea is to spend that money over a few hours early in production to save countless iterations later on in the classical render farm system.

It seemed to be a general theme this year that studios are exploring this kind of front-loaded pipeline technology. The idea being that an investment in the processes, workflows and resources available early on in the pipeline allows creative iteration to happen as soon as possible. This, for the most part, has proven to reduce cost and time overall by increasing the efficiency and smoothness of all departments later on.

Pipeline UX

Thinking about the spLit tool, developed for lighters with the single criteria of making them smile when they used it, we found ourselves talking a little more generally about UI/UX in the pipeline.

The idea of dedicating resources to or hiring UX people comes up a lot, and sounds like a great idea, but it always seems like the time/money/energy is just not there. Perhaps there is a world where we all follow some kind of UI standards (Material Design?) which would provide consistency across studios both in development skills and feel… but the look of a UI is really only a small fraction of what matters to the experience as a whole.

One thing we definitely agreed on is that some of the biggest savings in time, energy and overall satisfaction is often in the UX space. One idea that came from Siggraph this year was to consider UI/UX training for developers, which can set the tool builders up for a better chance at making something great while they hack together GUIs.

Siggraph

At this point it’s worth mentioning that we are bringing up information and ideas from Siggraph week itself, and not just the first two conferences so this is as good a place as any for the header.

The aforementioned front-loaded pipeline was something that extended though Siggraph this year, from real-time technologies in use on set to more Omniverse and other efforts aimed at increasing collaborative creativity and getting much of the iteration time out of the way early.

Speaking again about Omniverse, we discussed how they mentioned this idea of developing the HTTP of USD/scene data. This is an interesting concept: that our scene descriptions are much more akin to a communication protocol rather than a file format or singular data object. Apparently they also put a lot of work into optimizing some key areas of the USD libraries, which would be amazing to see passed back to the open source community (which they said they would do!).

This point brought out the question:

What do we really think about this dream of a completely open, standardized pipeline?

In some ways, we are inching our way there. There are some fundamental pieces that have gone open-sourced, and the ASWFAcademy Software Foundation is now helping to formalize and provide resources to some of those communities.

The Pipeline Conference keynote touched on this idea that our pipelines are not our competitive edge, but rather it’s our storytelling abilities. This may be true, but given how differently studios work should we ever expect to be able to share an entire tech stack?

And what is the business model around a completely open-sourced pipeline? Who pays for the development of such a pipeline, if it’s not a commercial offering? Probably for the ability to help drive the future of these libraries, and even just present so as to bring in better talent.

Perhaps, however it just becomes that the baseline shifts. Someone open sources a key piece of tech and so we all just shift focus to other areas. USD seems like one of these technologies - in a similar way to how Kubernetes as sort of taken over a large chunk of the service management and deployment space.

Cloud Native BoFBirds of a Feather - discussion session

We noticed a lot of cloud vendor presence at Siggraph this year, and they are really pushing hard for the cloud as a future for this industry. There are definitely people leveraging cloud technologies in their pipelines, but the way it’s often framed is that you absolutely need to pay for it all and should plan to move all your infrastructure to the cloud over the next 5-10 years (or whatever).

Something that we should all remember is that there are a lot of ways to leverage cloud technologies with our existing hardware and data centers, especially when the investment has already been made. This is also true for infrastructure talent; we must be conscious of the knowledge and skills that we lose by moving into the public cloud - to be sure that it’s what we want.

Another interesting presentation that was given in this session was by a chap from the NBC Universal feature/live action side of things. Apparently there is a large effort going on with them, MovieLabs, and the Entertainment Technology Center.

Their goal is to form some kind of standard and standards body around asset metadata, with the intention of making it easier for client and vendor to share the appropriate information. Exactly what this looks like is unclear, and obviously there is skepticism, but he was very confident that they were going to achieve this so we are keeping an eye out!

Metrics / Data / Machine Learning

There were a few related Siggraph sessions in these areas, but we talked a little more generally about the ideas. Some of what we talked about can’t really be posted publicly, but here’s some notable points:

OpenCue BoFBirds of a Feather - discussion session

OpenCue was recently made open-source and adopted into the ASWF. There’s a general uncertainty to this one as it is not widely adopted, if at all. It also doesn’t reach feature parity with what other studios are used to, but we are definitely curious to see how it evolves.

Python in Production BoFBirds of a Feather - discussion session

Not as chatty as previous years. Some talk about API layout and how much code is left to be “public” (via Python naming conventions). Also some talk about challenges with long Python search paths (PYTHONPATH) causing NFS load. The blister package was mentioned as being able to help with this using some caching mechanism. A general poll showed that CPython is nearly universal as the distribution being used.

VFX Reference Platform BoFBirds of a Feather - discussion session

In case you haven’t heard the platform is moving to Python3.7 in 2020. Of course, this is a big deal for all of us with immeasurable amounts of Python2.7 code so it was largely the topic of conversation. There was a case study moving a relatively large open-source application, as well as some other tips. One thing to note is that even though all of the migration tools are helpful, at the end of the day you still need to learn the differences and know what you are doing during conversion.

State of the {Rigging, Hair, Animation} BoFsBirds of a Feather - discussion session

Unfortunately none of our members were at any of these sessions but they sounded like good discussions. Generally the community coming together, seeing how we are feeling, and looking at upcoming technologies that might disrupt.

Ephemeral Rigs

This is an interesting technology that was presented this year, which generates rigging controllers on the fly based on joint selection and key combinations (etc ik, fk, slide, etc). This is intended largely for static posing or maybe some pose-to-pose animation.

We loved this idea, as it’s always amazing to see genuinely novel approaches to age-old problems/workflows.

Conclusion

If we had to summarize everything, there seemed to be a large amount of flashy, eye-candy stuff at Siggraph this year. Many of us felt like there was very little discussion or presentation which delivered immediate, tangible value to our day-to-day problems and workflows. Of course, it was still a great conference and a great week in LA.

Thanks for reading!


Notable Imagery

Building pipelines is kind of like hanging out by a waterfall. We want to put a little wire across the edge, and let people get as close as they dare go, with just enough to catch them from going all the way over.

This industry is like taking tech in one hand, art in the other, slamming them together over your head and seeing how fast you can run without the whole thing falling apart.

Production Pipeline Fundamentals for Film and Games came up as a book worth checking out.

Pipeline Patterns is an ongoing effort by Bill Polson to formalize and discuss the pipeline development ecosystem.

Solaris was announced by SideFX at siggraph this year, and we are curious to see where it goes.