Deep Exr 360 FOV- Problems

[mogw4i]

Hey guys!


I´ve got some troubles with the deep images and negative spaces out of a panorama view and I´m trying to figure out where the problem occures. I hope you can help!

The rendering comes out of maya 2014 & Vray v2.25.01. It´s a deep exr-file. The camera in maya is a spherical camera which gives me the panoramaview with 360°degrees FOV from inside the 3D model. ( I´ve added some testfiles for better understanding )

I choose the deep images to get my multipass-file plus all the information the model brings in deep for holdout usage later.

As you can see the rgba-channel show the information uncut but the deep and the depth pass seem like they´re values are inverted at the far left and far right part of the image. I´ve graded the depth a bit for better visibility that there is indeed information.

I´ve tried to multiply all the negative values by -1 to receive a panorama with positive values but there´s something wrong. At the connection between negative and positive values a very hard edge appears.

This connection edge stays at its 180 degrees FOV wherever the camera points to. So far it seems the panorama view out of the camera cuts the image into a positive and a negative space exactly before and after this 180 degrees FOV.

These values don´t seem to go on with 181 instead of that more like 179 180 0 -1 -2 etc. Is there´s a possibility the values switch after 180 degree?
Or is it me, the user?



I appreciate every kind of suggestion or idea!!!


Cheers,
Olli

[vlado]

We found out that Nuke wants as deep z-depth the distance to the camera plane, so that's why the values invert for points behind the camera.

Best regards,
Vlado

[mogw4i]

Thanks Vlado,

we inverted the negative values but still have issues at the shift of negative to positive values (around 180 degree horizontal and vertical).
The values of the depth around this area is over 1 and on the bottom of the image around 2000
The left image is a rendering in nuke with a sperical cam and the depth converted out of the deep image (right side) is supposed to look like the left one.



Besides that we are not pretty sure what exactly is stored in the deep.front and deep.back, when we compare the two channels, the deep.back looks different.
Can you explain these two channels and how vray generates them?



Thanks a lot,
Olli

[vasil.minkov]

Hello,

The deep channel contains distance from the camera plane. Deep.front & deep.back define the validity of the deep sample. Using a range of two values instead of a single deep value is useful for volumetric composition. It seems that currently deep.back is not generated correctly for negative deep values and - it is clamped to zero.
V-Ray generates additional layer, currently named "other.ZDepth", that stores the distance from camera. Could you take a look at "other.ZDepth" and see if it is useful for you?

[ultrasonic]

We found out that Nuke wants as deep z-depth the distance to the camera plane

Can u post where you found that info? For me that doesn't make sense. I tried rendering a spherical camera as deep camera map in houdini 13(alpha) and mantras behaviour in that regard is totally different. It seems to store deep info also based on the distance to the actual camera-position and not to the camera-plane...

Would that mean, that deep images from mantra and vray are not compatible?

thx
Oliver

[vasil.minkov]

Using distance form the camera plane makes sense for the default plain camera without a distortion. Distance to camera will not work for a plain camera, since it would distort the image.

Currently VRay always exports depth values as a distance from the camera plane, so other camera types - spherical, fish eye, etc. may not produce useful depth values. Default camera with a distortion also may not work as expected.

There is a workaround for the particular case of a spherical camera. Use DeepExpression node to replace "deep.front" and "deep.back" with "other.ZDepth". The resulting DeepExpression output should be useful and compatible with other deep images.