aliasing issue with Maya-vray zdepth pass in NUKE ZDefocus

[schnitt]

Hi there,

Got some issues with DOF using a zdpeth pass with NUKE's Zdefocus.
Got a simple scene: two area lights, a few models and a curved background plane, outputting with Vray to a 32-bit EXR scanline zip.

For the zdepth render element I tried the following settings:
Depth Black: 0
Depth White: 220
depth_clamp / filtering disabled

also tried with swiched black and white values, so:
Depth Black: 220
Depth White: 0
depth_clamp / filtering disabled

I tried different Zdefocus math settings, but with all settings I get an aliased DOF. (you can check the screenshot I upload.)
Also tried comping a double resolution version of the zdepth pass (using the Reformat node and trying its different filters) without succes. Results in the same aliasing.

I didn not try to seperate background and my models and then apply blur, because: if I render a zdepth-pass with Arnold it works fine. No aliasing at all. Because with Arnold I have to divide "1/zdepth" I figured that perhaps I'm doing something wrong with the Vray pass. Some sort of math/value issue.

Anyone got a clue here?

screenshots:
http://cdn.imghack.se/images/c46e975...a11415e45a.jpg
http://cdn.imghack.se/images/9407647...b6a310de28.jpg
http://cdn.imghack.se/images/28228db...4a3f4842e2.jpg
http://cdn.imghack.se/images/4923639...4cf270e27c.jpg

[Morbid Angel]

How does zdepth from arnold look?

[schnitt]

I see I forgot to name the links to the uploade files:

vray zdepth pass: http://cdn.imghack.se/images/c46e975...a11415e45a.jpg
arnold zdepth pass: http://cdn.imghack.se/images/9407647...b6a310de28.jpg
dof with vray: http://cdn.imghack.se/images/28228db...4a3f4842e2.jpg
dof with arnold: http://cdn.imghack.se/images/4923639...4cf270e27c.jpg

@Morbid Angel: You have any suggestions on how to get rid off the aliasing?

[Morbid Angel]

Its hard to tell without looking at your scene. Both of the depths have the same aliasing as I can tell, the arnold's depth is darker though. Have you tried matching vrays depth to arnolds by depth value? Also how are you writing the pass out is it exr?

[RockinAkin]

You cant have (and don't want) an Anti-aliased depth pass - as that would distort the depth values of the edges of your objects.

What are you doing in comp to get better results with Arnold's (also aliased) depth pass?

[schnitt]

@morbid angel:
Output is EXR. As you suggested I corrected the values from the vray pass to match the arnold pass. And then there's a some difference in contrast; arnold has a lot more values. So I tried some different depth settings in maya. Then I noticed something really weird: the arnold zdepth is actually a little bigger. And by bigger I mean that the edge of the model against the background extends 1 pixel more then the vray. You can check the screenshot if you don't get what I mean. (Also it's probably helpful to add that the Arnold zdepth works well with both the vray beauty as well as the arnold beauty-pass)
http://cdn.imghack.se/images/c8feaa5...6d9b8e0d11.png

@rockinakin:
I'm really not doing anything special. The arnold zdepth is a little different; first get the inverse of the zdepth pass, so 1/zdepth. Then shuffle it to the depth.Z channel, and then connect the ZDefocus with "math" set to "far=0".

@all:
But if we forget the result from Arnold for a minute. How would you guys comp a Vray z-pass over your beauty? I think it should be as straightforward as described below right?

* output zdepth render element (I found that black=far, white=close works best with Nuke)
* make sure it's not filtered or anti-aliased
* in Nuke shuffle it to depth.z
* connect ZDefocus (math far=0)

With vray it should be possible to get DOF with just a beauty and a zdepth pass right? Who has experience with this? What am I missing here? And does the difference in size ring a bell?

[Morbid Angel]

What I would suggest is to try rendering a multichannel exr and using the depth as it comes there, also try render to 32 bit. But generally speaking its not uncommon to erode out the depth path a little to solve some of those edge issues in depth. That is if you have the objects and bg separated, perhaps arnold does that behind the scenes? who knows...

[vlado]

Oh that... keep filtering off for the z-depth; go to the "Render elements" tab of the V-Ray settings, scroll down to the bottom and where is says "Unfiltered fragment method" and change it from "Best coverage" to "Nearest to camera"; see if that makes it any better.

Best regards,
Vlado

[schnitt]

Yess!!! Excellent!

Thank you very much mister Vlado

[hsz]

I thought that (Math Far = 0) is used for Renderman ... for vray or Arnold is (Math = Depth)? is this incorrect? can someone explain this please? Thanks.

[schnitt]

For Arnold, this image shows the zdepth with math:far=0
https://support.solidangle.com/downl...7184000&api=v2
However another older tutorial on their website says: "Z should be set to 'depth.z' and 'math' should be set to 'depth' (the Z value in the depth channel is the distance between the camera and the pixels in the scene)." If it works.....

[ARTECONI-CGI]

Hi All,

I get the same problem but in a max environment.
Those options that Vlado has mentioned are not available in vray for max, can anyone suggest a solution?

Kind Regards,

Giacomo.

[vlado]

The filtering option for the Z-Depth render element is available in 3ds Max. The option for the unfiltered fragments is available in the VRayOptionRE render element. There is more information about it here:
http://docs.chaosgroup.com/display/VRAY3/VRayOptionRE

Best regards,
Vlado

[ARTECONI-CGI]

Hi Vlado,

Thanks for this.
I was aware of the filtering option available on the zdepth pass but I didn't know about the VrayOptionRE.
I wil try it immediately.

Now that I see those options about Deep data compositing I reminded of a problem we have at work comping deep data and refractions.
I refer to this thread here: http://forums.chaosgroup.com/showthr...660#post625660

I wonder if there could be any relationship between those options available in the VrayOptionRE and the problem we encounter when comping deep data with refractions.
Maybe setting the Zdepth merge threshold to "by Zdepth" could fix it?

Kind Regards,

Giacomo.

[vlado]

Yeah, I still don't know how to deal with the refraction problem :/

Best regards,
Vlado