Poseidon is now well and truly out there, with its stunning visual effects and strong technological advances on display for all to see. We gave you the opportunity to ask questions to the guys behind the scenes and quite frankly you jumped at the chance. Indeed you asked so many questions, that we thought it best to split them between two of the industries finest.
We would like to present Philippe Rebours, CG Supervisor at ILM, and Rolf Herken, CEO and CTO of mental images. They have kindly agreed to answer your questions on Poseidon and mental ray.
Philippe Rebours
Born and raised in France, Rebours received the equivalent of a Bachelor of Science degree in Engineering, with a focus in Computer Science from L'ecole Centrale de Lyon in France.
Upon graduation, he worked at BUF Compagnie in Paris, known for their work on Batman and Robin, and City of Lost Children, where he designed software for commercials, music videos and feature production. He joined Industrial Light & Magic in 1998. He soon transitioned over to feature production where he worked on several projects as a technical director before being promoted to a sequence supervisor on Star Wars: Episode II “Attack of the Clones.” His latest work has been as CG Supervisor on Poseidon.
Industrial Light & Magic
http://www.ilm.com/
Rolf Herken
Rolf Herken founded mental images over 20 years ago and is the driving force behind its innovative technologies and products as well as its successful establishment as one of the global leaders in 3D visualization. He is the CEO and CTO of mental images GmbH in Berlin and president and chairman of metal images Inc. He has been serving as the director of R&D with primary responsibility for Research, Development, Technology Strategy, and the company's OEM business.
Trained as a theoretical physicist with a strong interest in Theoretical Computer Science, he established the company in 1986 while working on a Ph.D. thesis on computability aspects of quantum gravity. Rolf also has a background in visual arts.
Mental Images
http://www.mentalimages.com/
Firstly we will speak with Philippe to answer your questions regarding his work on Poseidon.
Philippe, can you please tell us a bit about yourself and your position within ILM?
PR: I’m what you would call a CG supervisor and on Poseidon I basically did the set up of the boat, the lighting, materials and a few sequences, mainly the first sequence.
Can you tell us about your previous projects?
PR: I was the CG Supervisor on Lemony Snickett last year in which we made a CG baby. I worked on the lava scenes of Star Wars Episode III, prior to that Terminator 3, Peter Pan, Minority Report, Star Wars Episode I and II and A.I. I have been working here at ILM for eight and a half years now starting as a CG Technical Director then Sequence Supervisor and now CG Supervisor.
What was the hardest scene you worked on in Poseidon and why?
PR: The hardest scene was definitely the first shot and why…because it was a 4000 + frame shot, 3 minutes long. It starts underwater and then you go right around the boat following Dillon, (one of the main characters,) as he jogs around the boat. This type of long shot requires so much more management due to the time it takes to render. In a long shot like that you need to be able to see the people, the chairs, the deck and many more objects in great detail. We worked for 8 months on that scene.
What software did you use?
PR: As you know we used mental ray, we used Maya to model, we did some painting in photo shop, Physbam for water simulation, Renderman for particles and ocean shots. We used our own software for a great portion of it called Zeno, but for rendering the boat we used mental ray and only mental ray.
Was there anything that you wanted to do, but because of the sheer size and scale of it just couldn’t?
PR: When you are a perfectionist I think you always want to do a tiny bit more. I would say though that we are pleased with the film and we managed almost everything we wanted to do. You still just wish for one more week though, especially on that long shot.
How was the texture of the ship created?
PR: We had a lot of references. They went out and took a lot of photos of a cruise ship. Otherwise a lot of it was painted with a lot of detail. You see it’s like architecture. There are all different pieces that needed to be assembled to create the boat. Each one is treated like a seperate creature that needs to be modeled and painted. We usually use three, four or more types of texture per object.
What other research did you have to do?
PR: Photos, we developed a Global illumination system that we call Cine Radiosity, and that gave us the possibility to create global illumination with a good rendering time that we could work with. If you take that first shot for example, all we had was the sun, a little ambience and the reflection in the water, so basically the whole piece required global illumination.
How long did it take to render the shots?
PR: About an hour for the pre pass and maybe four to five hours per frame. That first shot was 4000 frames…so it took the equivalent of five to six days non stop rendering on all of our CPU’s assigned to the shot. Night time shots could take longer...Up to 7 or 8 hours a frame because we were using more than 1000 area lights.
So how many processors were needed to get the job done?
PR: We used 300 processors just for that long shot.
What exactly does Ray tracing Global illumination involve?
PR: It searches around to find light. It questions every single geometry. In mental ray there is a final gather which does that, but it’s very important to be able to tweak it, we created our own shaders to do this.
Does this type of work require custom and optimized shaders or are default shaders still usable?
PR: You can start off with default shaders, but for this piece we used custom shaders. We needed to be very specific. You tend to get a more pleasing results with custom shaders, but you can definitely make a start with default shaders.
Which scenes are you most pleased with?
PR: Obviously the first shot, I am very pleased with, but also there were some of the night shots which just looked beautiful. Oh and there is another when the boat is coming towards you head on…that did look really real to me...in fact there are so many of these scenes that you just really would not think they are not real.
Is there anything else you would like to add Philippe?
PR: It really was an interesting experience. It was hard and fun to work on. We certainly learnt a lot about rendering techniques. For the future it taught us what we can do and where we can go. I am pleased with the results.
Philippe, thank you so much. It has been great talking with you.
Now we move into a Q&A Session with Rolf and try to answer the questions our users posted!
Rolf, can you please tell us about yourself and your position within mental images.
RH: I founded mental images 20 years ago. I am the CEO and CTO of mental images GmbH in Berlin, and President and Chairman of mental images, Inc, our wholly owned US subsidiary. My background is in Theoretical Physics and Computer Science as well as in visual arts.
With applications such as Gelato allowing some renderers to take advantage of GPU rendering, is this something that mental images is looking into for the future?
RH: We showed the use of GPUs in mental ray for accelerating rendering by performing shading related calculations for the first time at Siggraph 2004. In particular, we are providing extensive support for hardware shaders written in CG through the mental ray shader API since version 3.3 of mental ray.
With the most recent, enormous improvements in compiler performance that were accomplished by NVIDIA, this feature will become a viable option for anyone seriously interested in significantly cutting rendering times, for example in feature animation production.
(We expect this to really take off as soon as we deploy our new rendering target-platform independent representation of shading by means the target universal shading language MetaSL from the forthcoming version 3.5 of mental ray on. See below.)
As to the use of future GPUs to accelerate our core raytracing rendering algorithms we have embarked on a fundamental project in close cooperation with a leading graphics hardware company.
As of today, there is no rendering software that is using GPUs or even custom designed raytracing processors that can beat mental ray's raytracing or global illumination performance and quality. Even custom designed raytracing processors cannot match it. mental ray exploits thread-parallelism and host-parallelism to accelerate rendering with very high scaling efficiency, literally reaching 100% scaling efficiency on multi-core processors.
mental images has a very strong working relationship with ILM, (such as ILM having access to mental ray's source code). Does any of the development ILM makes with mental ray internally make it into the general release that the public can get access to?
RH: Our relationship with ILM is very strong indeed. We provide custom support and development to ILM on a 24 hour turn-around basis. We also have extremely qualified staff on-site at ILM if it is required by them. However, ILM does not have access to mental ray's source code. In fact, ILM gets what they need with the additional benefit of not having to worry about the source code. All mental ray core development is done exclusively by mental images and all features and improvements in mental ray that result from the working relationship with ILM will be released to the public as part of one of the next general releases of mental ray.
Renderman has quite a strong share of the feature film market. Now that machines are getting faster and 64 bit machines are almost standard, and GI is becoming more attainable, do you see mental ray taking over Renderman seats in the future?
RH: Yes, of course. mental ray will be the rendering software in the majority of all professional feature film rendering pipelines. We are working closely with our OEM partners on making this happen.
Rendering algorithms such as radiosity and sub surface scattering are now something available to everyone. What new rendering methods and/or challenges do you see mental images working on in the future for mental ray?
RH: Well let’s see...in the immediate future mental ray 3.5 will be released.
The development of version 3.5 is focused on improvements of performance and usability, while ensuring a high level of compatibility with earlier versions. The following areas are at the focus of the improvements:
- global illumination and final gathering ease of use and performance
- ease of shader creation and shading performance
- rasterizer performance
- ray tracing robustness
- ease of parameter tuning for options and scene data
- backward compatibility
- physically based rendering
For the future we are working towards the ultimate goal of being able to render incredibly complex scenes (with billions of triangles) using raytracing based global illumination algorithms with complex shading including BRDFs, subsurface scattering, and frequency dependent phenomena
*in real-time*.
Our next generation software rendering technology that is built into RealityServer, achieves rendering times of less than 40 milliseconds per frame (i.e., realtime) with full raytracing (and very soon full global illumination) on small 64-bit commodity multiprocessor machines or blades on scenes with tens of millions of polygons.
Parts of this entirely new, revolutionary rendering technology such as our dynamic BSP acceleration methods are being integrated into mental ray, beginning with version 3.5. In mental ray 3.6 we will have support for assemblies, allowing rendering of arbitrarily large scenes, e.g., crowd scenes with a million creatures, even with our raytracing and global illumination algorithms.
Finally, we work on the ultimate solution to the shader creation problem. It is not only the "Holy Grail" of the games industry but also for the VFX/Feature film production - given that the two will converge eventually. In fact, we venture to say that only with our forthcoming mental mill technology and the underlying generalized shading language MetaSL, it will be possible to truly leverage the power of the evolving graphics processor and rendering software technology without being forced to constantly rewrite shader assets for each new hardware rendering platform release, including the variety of games consoles
MetaSL is the universal shading language that we have designed as a superset of all shading languages. It is independent of any target rendering platform, and hence permits the creation of shader assets independent of todays as well as of future hardware and software rendering target-platforms, including GPUs and custom processors as well as software renderers such as mental ray 3.5 with its C++ API.
mental mill is currently in an early Beta phase and will be released this year.
RealityServer 2.0 is in Beta 2 and will be released in the third quarter of this year.
Also, this year we will release mental queue. It will be the most powerful, fully configurable general process management system for production pipelines commercially available. In particular, it will allow you to control any mental ray standalone or batch rendering pipeline with any number of nodes and individual users.
Many renderers specialize in certain tasks, (such as the new Maxwell render focusing on the visualization side of the industry, and Renderman on features,) mental ray is in quite a unique position of dipping it's toes into all of these markets... Is this something you see mental ray continuing, or will it eventually focus on one area?
RH: Yes, we will keep mental ray in its unique position. After all, these various "sides" of the industry are not unrelated at all. If a car needs to be rendered in a VFX movie, why not use the original CAD data and the correctly measured BRDF for the car paint for it? Or, replace the car by a human face with its slightly translucent skin. mental ray can handle all these situations and allows the achievement of true virtual cinematography. A product designer is just as interested in a realistic look of the product as the cinematographer.
I don't think mental ray "is dipping its toes into all of these markets". mental ray rather has a very large, if not the largest footprint in each of these markets. mental ray is the integrated high-end renderer of the world's leading CAD software products as well of the world's leading DCC software products.
Poseidon Ship Only Render Pass
Well thanks for clearing all of that up gents. I had a great time speaking with you and all at VFXTalk thank you for your responses to their questions.
Written by Kelly Campbell
VFXTalk.com
Poseidon Waves Render Pass
Poseidon Waves Fluid Dynamics
Poseidon Final Render Pass
VFXTalks Meets the VFX Masters Behind Poseidon –
