 Hallo und herzlich willkommen zurück auf dem FAM-Kanal. Die nächste Speakerin ist Franca. Sie hat sich mit Computer-Grafik befasst und zwar genauer mit... Hallo und willkommen zurück auf der FAM-Channel. Die nächste Speakerin ist Franca. Sie hat sich mit Computer-Grafik befasst und zwar genauer mit unterschiedlichen Renderungen. Es gibt keine Q&A auf dieser Channel, sondern eine Breakout-Session im Städt. Wir zeigen euch genau woher sie ist. Sie ist auch im Chat. Vielleicht kann sie euch direkt beantworten. Jetzt viel Spaß mit dem Talk. Willkommen zum Battle of Render Engines. Heute soll es um die Frage gehen, ob das gleich mal... Hallo. Willkommen zum Battle of Render Engines. Es geht um die Frage, ob das gleiche Material, Plastik oder Metall in unterschiedlichen Renderungen wird. Mein Name ist Franca Bittner und ich studiere Medientechnologie an der TU. Ich bin jetzt ein Master. Vorher habe ich in der Audiovisual-Media-Studie im Hochschule-Medien-Stuttgart eine Bachelorstudie gemacht. Die Resultate haben in den skilligen Procedungen der Computer-Science, unter dem Titel, Comparison of Material Models in Modern Physically Based Rendering Pipelines, und die erste der vier Renderer, die ich compared habe, ist Arnold von der 3D-Grafik-Software Maya. Arnold ist ein Path-Tracer, das gleiche als die Cycles der Render-Software. Und beide Renderer sind offline Renderer, das heißt, es dauert eine Weile, bevor man die Resultate von Renderern bekommt. Und sie arbeiten nicht in real-time, wie EV, die auch von Blender, und die Unreal Engine 4 sind both real-time Renderer, also arbeiten sie in real-time. And the Unreal Engine 4 is also the only gaming engine that I have also analyzed. The categories in which these compete against each other is Roughness, Metallic or Metalness, Spectre, as well as Clearcoat. You can see here the effect of all those a little bit, and I will explain that in a little more detail later what all this means. Why did I do all this? It's about physically based rendering. And this means that the real world is emulated in the virtual world with the help of physical laws. And the background is why this is done, is that in the end it simplifies the workflow of the 3D artists. In earlier times this wasn't done, and every phenomenon that was observed was modeled for the concrete case, where you wanted to use specular highlights, for example light and shadows have been baked into the textures. But if you move an object in a scene or a light, then of course the new scene had to be re-rendered, and the texture had to be re-baked. So, to decouple those from each other, the place of the objects and the creation of the materials, there was a new abstraction layer, and this is now based on physical laws, and it was re-implemented or at least aligned with physical laws. And in the end this also caused more realistic results, especially if the light situation changed unexpectedly, and it's not so easy to recreate this manually. But why do we need a comparison between the renderers? There are many different material models, which all have different features, so to speak. So, with some material models, some things are possible, which are impossible with other models, and the question is, or what you really want is that intuitively you can use the models, which you can easily work with as an artist. So I analyzed if and how the material models differ between the renderers and which model is also more intuitive to use. So, to make this a fair competition, we have to use the same object for rendering everywhere, to show this. And I used a simple convex sphere, because it's very easy to implement. And I have, there are two lighting situations, which they compete with each other. One is the very unnatural lighting. It's a simple unidirectional light source that doesn't really exist in reality, but it is very easy to implement. And that's why it is included in the results, because we expect that in this situation all the renderers should behave very similarly. And the other case is the very unnatural, lighting, which we use with several environment maps. And it's image-based lighting with environment maps. And I have evaluated this qualitatively in a table, in a material table. And down here you can see the five different lighting scenarios. One is none, this very unnatural lighting, with only one directional light source, which doesn't exist in nature. Then there's the different natural scenarios. What is in the background here, and what's meant by moonless golf. And so those are the three, four different arenas, so to speak. And in the end those are the pictures that are used for the image-based lighting. The background, that is lit, and casts the light back onto the sphere in simple terms. And now since all the basics have been explained, let's go and fight. And here are the results for the result roughness. And at first maybe you see that not much is very different in this unnatural lighting situation, but also in other scenarios. The results to the roughness are very similar. What does roughness mean, really? It's about an object that's very rough or has a very smooth surface. And so as a result reflects the light really focused or just distributed. And here we can see how I did this in this material table. For each renderer I created a rendering framework, tailored to the specific renderer, since there was no unified interface. This will slowly arrive with Python now, but it wasn't at the time. In steps, with a step size of 0.1, I changed the parameter. That's for all parameters I used the same step. And as you can see here, I have taken one picture and then put it into this table. So from roughness equals zero, which means a very, very smooth. So smooth that you can't even see the highlight until the roughness 1.0, which is the perfectly Lombardian rough material. It's not really that important. And here we can see in the material table with different pictures what has changed from step to step. So if roughness has gone from 0.1 to 0.1, what has changed. And when the color is red, it has become brighter. Zero is white as nothing has changed. And if an area is blue, that means it has grown darker compared to the picture before and on the right. I have shown before and after between roughness 0 and roughness 1. And as you can see, this is especially between the renderous cycles and Arnold compared to Evian Unreal. There's a difference because at the edge of the sphere, especially with cycles at the inner edge, where there's the Terminator, there seems to be a special feature of path tracing algorithms, but it's hard to say, but maybe it is caused by capturing the effects much better. And in the end, if you look at these two in detail, you see that cycles in the right-hand side, shown in yellow, at the edge of the sphere, there's also an increase in brightness. What we see here is a single line across the picture of the green channel. The green channel is representative for what we humans see as brightness because the biggest part of brightness is from the green channel and the little rise in the end at the edge of the sphere, where it is even brighter than in the middle of the sphere, that is called retro-reflectivity because only cycles is modeling that correctly, not even Arnold or any of the other renderers. None of them model that correctly. Also below here, what's in gray here is the sphere itself, it's a little stripe, a little strip from the picture itself to show that much better. Let's go to the next one, to the next parameter, back. Hier are bigger differences, especially if you look at the Unreal Engine compared to the other three. And here is the overview. And Unreal really stands out. Yeah. But especially if you look especially what happens from one step to the other, it looks quite similar for all renderers, says that the parameter is really just a linear interpolation and I really looked at the implementation in the source code of all the renderers and nearly all of them metallic is really an influence between the state of non-metallic to fully metallic and you can really see that here except for Unreal ist also one line across the image from on the left side non-metallic, on the right side fully metallic and I wanted to be sure that I haven't made a mistake in modeling the light situation which would have made Unreal so darker. It is somewhat darker but it's really similar to the others but nevertheless the difference in the metallic property is quite big and I have no idea why why that happens, I couldn't really find it out but in my experiments this was the case. Yeah. The next parameter is Specular and the influence of Specular is a little bit more subtle and for Arnold it is hardly visible at all by Evian Cycles the edge something changes quite noticeably but you can see that this highlight of this, the reflection of this window is fading in and out and it becomes visible and then goes away again and in the end that's what Specular does it's about in the lower three Evian and Unreal the Specular term influences the reflection index of reflection and in Arnold it is really a multiplier a simple value that is multiplied with the actual brightness value the luminosity value and this parameter has a functionally different way in the programs of course if you look at the programs and work with the programs you don't really know but you think it's the same name so it's probably the same thing that is being influenced but it's not the case here and you can see it here in the difference pictures with Arnold it gets darker and darker towards the edge whereas with Cycles and Evie the more you increase the Specular parameter from 0 to 1 and that is perhaps also the case because the Fresnel effect has an influence here I can't talk about this in the scope of this talk much more it's about the reflectivity how something is mirrored is reflected at the edge of an object and in the end it's really correct if the light from a physical point of view as far as I can see as far as I can tell although I'm not really an expert if the light is being reflected stronger at a different point of view and the last category is clear code it's also an effect the effect is quite subtle in all four renderers and you can see it the strongest in Arnold so maybe you can hopefully you can see it that the edge becomes a bit darker the edge of the sphere maybe if you couldn't see clearly in the difference pictures here and there's really the effect what clear code does perhaps I have to explain that a little bit there's a difference between clear code and roughness which is rather subtle so you really have to look very closely at least in the results of my evaluations here maybe in other scenes it can be seen more easily or in other object shapes but here are some examples on the very left there's roughness 0.2 which is the default value and clear code is set to 0.0 and if you increase clear code from 0 to 1 you can see that this light orange highlight is still in the background a little more blurry but if instead you reduce the roughness to 0.1 and remain leave clear code at 0 and the lightly blurred background is missing then and that means that clear code means that you have sort of a second layer onto the object which interacts with the light in a different way and that is supposed to be the effect and the name itself is also slightly somewhat intuitive so so I thought it was an intuitive parameter even if the effect is rather subtle and that's more or less what I wanted to say as a conclusion maybe not all renderers display materials the same which are supposed to be the same well supposedly the same, I don't know if it's really the same material that I created in all renders but I tried as well as I could to set the same values for all of them and especially where the values um I used the same the same material parameter names I used, I used the same values and that means that there is a problem for 3D artists where if the exchange of the materials between the tools is getting more difficult by this you can't really just use the green shader which really means in this context means material so you can use the one from Arnold copy it over to the other engine without information getting lost or being interpreted differently and the biggest visual differences are created usually by rendering technology itself so the path traces usually have a more realistic result than the real time renderers which use a simplified model and maybe in the future there will be a standardized material model and I've seen works about the web based display of materials and I think that would be interesting and other things to be aware of is that the influence of algorithms for light transport for example in the environment maps I haven't really evaluated that in a systematic way and the selection of environment maps and the shape of the object have an influence as well as the Albedo which is the base color of an object and the texture and in addition it would be interesting to evaluate if and how because was you do is you don't change only one parameter but you change a lot of parameters normally and of course it would also be interesting to look at other rendering tools for example Unity or the Godot engine and look at those and of course also look at different material models for example anisotropic or materials or there is subsurface scattering for example as human skin and so if you have kept listening until now thank you for your attention you can find the source code and the publication on the website for this talk and for the paper thank you very much yes thank you Franka for this very interesting talk as I said there will not be a question and answer session here but there will be a breakout session in the RC3 world and it will be in the FEM assembly and it should be here on the screen so if you don't have a ticket you can look at the RC link directly so thank you very much for your attention also from the virtual translation booth and you just heard the talk battle of the rendering engines and the translation was by pink dispatcher and if you have any feedback for us please use the hashtag c3lingo bye bye