OTOY Forums

Damn poltergeists!

Well, here is a SECOND curious "problem" with Shadow Catcher. If your emitters are all off, so that the first bug already discussed does not happen, and you render with Alpha Channel ON and Environment ON, you get this:

Notice that the shadows have picked up the texture of the underlying HDR image. That is OK if you are compositing over the same image. But if you are not, it spells trouble.

With Shadow Catcher working this way, you cannot render shadows to Alpha while at the same time checking the render as a whole periodically to see how it is looking overall.

And a THIRD issue. This one may be a feature of Shadow Catcher and not necessarily a bug. If it is a feature, this causes a whole can-of-worms of other problems:

On the left: Alpha Channel OFF, Environment OFF, Shadow Catcher Surface Disconnected.
Reflections are what they are when Shadow Catcher not active.

On the right: Alpha Channel OFF, Environment OFF, Shadow Catcher Surface Connected.
Look closely a the reflections in the car and the ground reflections are changed as if Shadow Catcher has projected the bottom details in the image up to the catcher surface:
It does not matter if Alpha Channel is on or off...Same problem in the reflections. This makes it impossible to do a beauty pass of the mesh w/o shadows and get it to mate properly to the pass w/ shadows.

I see that as a feature, and works for the way I use matte objects. This would only not work if you wanted to composite the render on to a different background I guess.

Are you referring to the 2nd or third issue post?

The 3rd problem just seems like projection inconsistencies.

treddie wrote:And a THIRD issue. This one may be a feature of Shadow Catcher and not necessarily a bug. If it is a feature, this causes a whole can-of-worms of other problems:

On the left: Alpha Channel OFF, Environment OFF, Shadow Catcher Surface Disconnected.
Reflections are what they are when Shadow Catcher not active.

On the right: Alpha Channel OFF, Environment OFF, Shadow Catcher Surface Connected.
Look closely a the reflections in the car and the ground reflections are changed as if Shadow Catcher has projected the bottom details in the image up to the catcher surface:

It does not matter if Alpha Channel is on or off...Same problem in the reflections. This makes it impossible to do a beauty pass of the mesh w/o shadows and get it to mate properly to the pass w/ shadows.

I could be understanding this wrong, but it seems like that should be the expected behavior. When shadows are present on a surface, even if it's matted, they are visible in reflections. Otherwise, how would you comp in ground reflections back onto the car?

It's a feature. It's implemented since v1.51:

http://render.otoy.com/forum/viewtopic. ... 62#p177710
http://render.otoy.com/forum/viewtopic.php?f=9&t=38911

Got it.

I figured it had to be something like that, I guess, since the HDR image is projected out there on a big sphere. It is not possible to pull out reliable data as to what is ground and what is not. WE see it with our eyes because we are used to looking at images and our brains fill in the required information. If I am correct in understanding this, if the HDR image was shot on a level flat piece of ground, it is theoretically possible for a program to understand where that ground plane is in the image. Unfortunately, there is no way for it to know where the boundaries of that plane are and just how everything beyond it was oriented in 3D space. It would be even worse if, say, the HDRI image was shot directly above a bunch of stair steps. So taken to the extreme, there is no way, without additional image data, for a program to know where ANYTHING really was in 3D space when the image was shot. So if a mesh is sitting on a ground plane which will be used to generate a shadow catcher, only a kludge can make it look like the ground plane of the shadow catcher is coincident with the ground plane in the HDR image. It's all hocus-pocus. It might be possible to get better mapping, maybe even remarkably better mapping, if you could draw vector paths over the background image for a particular shot, and tell Octane, "this area is flat ground, this area is a vertical wall, and here is a basic perspective grid to help map out the HDRI back into true 3D space."

I imagine this could be resolved with a stereo pair of HDRIs, but then that would be a bastard of a pattern recognition problem. Not to mention requiring super hires images that would choke a current day computer or GPU.

I don't know any software that does the camera mapping automatically. The most automatic you can get would be a 3D Scan I guess. Otherwise it's common practice to reconstruct the environment partial.

I'm confused. What specifically are you saying is wrong/inconsistent withe the shit in the right? Which I think your saying is the one wit the shadow catcher ground plane?