Ghonma, I'm confused about this conversation.

Your answers now seem to be in response to the idea that Poser can do these things now, whereas I was talking about what software in general could do if it were written a different way.

Specifically, what caused me to write about how to do soft speculars was when you said this:

Quote: It's not possible in any software.

You didn't qualify that by "any software that exists today". I can only respond to what you say, not what I think you meant to say.

Quote - All of which is irrelevant, since 'asking' the renderer for baked maps at rendertime, is yet another feature that poser doesn't have. You may as well talk about using lightmaps for SSS, or brickmaps for AO or whatever for all the good it does. It's not doable in poser without a major overhaul of the rendering system.

It does not require a major overhaul of the rendering system. Have you written a renderer? I have. It is not a major overhaul to add convolution.

Before Poser 7, mip-mapping did not exist in Poser. It was added, but the renderer did not change in any way.  Similarly, I could add a specular convolution module to Poser and consult the resulting map as a lookup, same as is done with the Sphere_Map node.  Ice-boy is doing this by hand, but it works. He makes a specular convolution corresponding to a particular sharpness of specular. He loads the resulting image into a shader, and connects a Sphere_Map node to it. This works already, in today's Poser. The only problem with it is that he has to make and load the convolved image manually, and the Sphere_Map node does not pay attention to any other objects that might be occluding it's view of the specular convolution map. It would be a simple matter to write a new version of the Blinn node that does include shadowing, but instead of simply delivering a fixed amount of specular with a fixed color, would look up what to deliver from the specular map.

I'm sorry if this is not clear to you, but it is clear to me and other people who know how to write 3D software, and I wish you'd stop claiming that stuff is impossible and not doable. It's very distracting. Please don't keep repeating it.

Quote - It's not a bad idea, but IIRC the larger bottleneck in specular convolution is the fact that specular convolution requires calculation of a virtual phong specular for every single pixel in an HDRI

Sigh. First of all, I just got done explaining in my previous posting that unlike diffuse convolution, specular convolution does not have to use data from an entire hemisphere of the map from each point. Since the specular "cone of influence" is much smaller than 180 degrees, the cost of each point is orders of magnitude less work.  Suppose you're using a 2 megapixel environment map. For the diffuse convolution, fully half of that, 1 megapixel, must be consulted for every single point in the map. But consider a specular effect that has only a 60 degree cone (the coefficients outside the cone are 0 and can be ignored.) Instead of 1 million pixels, this only requires adding up the values of 111 thousand pixels. Further, we don't actually need the full resolution and we can mip-map this down to 10,000 pixels. That is 100 times faster than the diffuse convolution. If we're talking about a tight specular with a 10 degree cone, then it would be 10,0000 times faster. There is no bottleneck - you made that up.

Furthermore, Poser is already doing this optimization. Did it never occur to wonder why HDRSHOP needs hours and hours to compute the diffuse convolution of a diffuse IBL map, but Poser does it in a fraction of a second just before rendering? The reason is because when the cone of influence is large, you do not need the detail provided by the map. That's why in HDRSHOP they tell you to drop your image size to something like 200 by 100 pixels. You don't need resolution for a 180 degree convolution cone.

As the size of the cone drops, you need more resolution, but the zone of influence correspondingly drops in exactly the same proportion. So you do more iterations on the one hand, but less iterations on the other. The upshot of this is that the time it takes to produce a convolution, of any size, is pretty much a constant.

---

Renderosity forum reply notifications are wonky. If I read a follow-up in a thread, but I don't myself reply, then notifications no longer happen AT ALL on that thread. So if I seem to be ignoring a question, that's why. (Updated September 23, 2019)