(Open) [HOU-1499] (Open) [HOU-1495] (Fixed) [HOU-1424] Cryptomatte | Artifacts on edges / Sawtooth

[maxtrks28]

is this supposed to work with Depth of Field? I am using depth of field and i am getting the jagged edges in the cryptomate file.

VFH build: vray_adv_43001_houdini17.5.460_8fdcce1_6154
Houdini 17.5.460
Windows 10

[maxtrks28]

here is the pic with matte mask:

[maxtrks28]

in case you want to see the exr, here is the link: https://www.dropbox.com/s/f12df4nuge....0001.exr?dl=0

[Gosho.Genchev]

Hey, maxtrks28 ,

Please correct me if I'm in the wrong here. So, I've consulted a couple of colleagues with more experience in compositing and the general consensus is that the result appears correct. You may want to increase the Subdivs further to clean up the aliasing in the ice cubes in the background.
I played with the EXR you gave us using the cryptomatte as a mask for a Color Correction node. I've attached a short video here: https://drive.google.com/file/d/17D_...ew?usp=sharing

Regarding your first post - I assume this is a screenshot from the V-Ray Frame Buffer ? If so, we'll try to reproduce this here and fix the preview in the VFB although if it's not blocking your work, it may take a while.

Cheers!

[maxtrks28]

Thank you for the quick reply, yes, that first image was from vfb, but the result is basically same when i bring it into After Effects. What i was doing it masking out the Light at the back, not Color Correcting. Is it possible for you to try that? it is the background. It may be acceptable for color correcting, but it may have issues masking it out.

To make sure it is not me, I also tested the same scene with Redshift.

I rendered in Vray at Min/Max subdiv 1/24, which took 2:49 to render.
Rendered in Redshift fairly similar settings (I think), took 1:13.


If you can, please check out the exr files from both VRay and Redshift, specifically, masking out the background. Here are the links:

V-Ray exr: https://www.dropbox.com/s/5qhehi9i35....0001.exr?dl=0

Redshift RBG exr: https://www.dropbox.com/s/luq2osu5su....0001.exr?dl=0

Redshift Cryptomatte exr: https://www.dropbox.com/s/hox7jdh9at...matte.exr?dl=0

Also, below are screen grabs from the masked out results in After Effects. I feel that the difference is pretty drastic. Please let me know what you think.

Once again, thank you for your time!

[maxtrks28]

This is Vray:

[maxtrks28]

this is Redshift (the white dashes b/c it is a Demo version)

[Gosho.Genchev]

Hey, Maxim,

just checked the files. I still think this is a matter of not enough AA samples - could you please draw a small region around the area you've drawn on the picture above, and try rendering with higher Max Subdivs ( e.g. max 48 ) - that way we'll know for sure if that is indeed the cause for the problem ?

As to the render speed comparison - we do compare V-Ray to other rendering software. I can't tell for certain what is causing the difference in this case, though. I'll try to rebuild something similar to check. However, looking at the image:
1. Multiple layers of refraction (as in the V-Ray render of the bottle) can slow things down significantly.
2. The intensity of the lighting will also affect speed - in general, the dimmer the lighting, the noisier the render.
3. I'm by no means an expert in Redshift so take this with a grain of salt - the trace depth for the refraction of the ice cubes in the RS image appears to be capped.

Cheers!

[maxtrks28]

Thank you once again for looking into it. I did as you suggested and raised the subdivs. I rendered at Min/Max subdiv 24/96, which made the transition smooth.
I adjusting materials by disconnecting all the textures so that it would be more of a leveled field with the RS comparison, I then re-rendered the whole image just to see how long it would take--it took 8 min 45 sec. (Below is the screenshot of the result)

It seems that at higher subdivs the transition is as smooth as RS (at relatively much lower settings). But as the subdivs are lowered to get faster renders, the transition degrades drastically as seen in the earlier post.

I also rendered in RS at much lower sampling levels (render times a bit over 30 sec), and the transition degraded a bit, but was still acceptable levels.

I am not saying that one is better than the other, i just wanted to compare the results to ensure that i was not doing something wrong.

My question is, would it be possible to have better quality transitions in cryptomatte files at lower subdiv levels?

Here is the link in case you want to look at it: https://www.dropbox.com/s/sifbom5rr1....0001.exr?dl=0

Thank you

[maxtrks28]

the is the screenshot of subdiv 24/96

[Gosho.Genchev]

My question is, would it be possible to have better quality transitions in cryptomatte files at lower subdiv levels?

That's definitely something we'll look into, and I'll ping the dev team with the information you've provided us with. Thank you for giving a shot to my suggestion, and thanks for the EXR as well!

Please ping us if you hit any issues.

All the best!

[maxtrks28]

thank you for your time!!!

[Gosho.Genchev]

Hey, maxtrks28 ,

last evening I spent a couple of hours trying to reproduce the problem here - even at 1/8 min/max subdivs both for CPU and CUDA, the aliasing is nowhere near as jagged as in the screenshots above. Would you mind sharing the hip file? Feel free to remove anything but the ground plane and the single ice cube in the back. If sharing the file is not an option, hopefully a couple of screenshots showing the physical camera settings and the ground plane's material would shed some light.

Cheers!

[maxtrks28]

thank you for looking into it, i sent you the hip file via email. Please let me know if you did not get it. Thank you

[Mr.Hoo]

I am following and looking forward to news