30k render (strip renders) do not match near shadow catcher plane

[JeffPatton]

I'm running some tests and one of which is a 30k render because I have those type of projects come along from time to time.  With Corona (tested RC2 and RC3) I'm not able to directly render @ 30,000 pixels wide (it crashes out of memory). 

I used the network/backburner/strip method to render the image out in strips.  That kind of worked.  It at least rendered the image.  However, I'm getting one sequence that remains different (nearest to the shadow catcher ground plane):


There were 10 strips in total, here's where they aligned:


I've tried it with path tracing and using a saved UHD cache (lower res save) but it always produces this anomaly at the bottom.  The split at the upper portion of the car looks fine.

I've also tried the Rakete_BigRender script posted in another thread here about large render issues.  It also produces a different look but within each tile though.

Thoughts?

[arqrenderz]

I think you need to call for keymaster or some corona team guy...

[pokoy]

I had a similar issue when rendering tiles way back with Alpha already, but didn't come across the problem again later. It was as if the renderer wouldn't take into account GI from objects not visible to the camera/in frame which would be the case with strips. It rendered fine when the entire frame was rendered but not when tiled.

[JeffPatton]

^ Same deal here.  It renders fine at lower resolutions.  Starts throwing memory warnings around 15k, but still renders ok.  It will crash 3ds Max when I go this large in one shot though, so I was trying the scripts & strips options.

Might be helpful if there was that old bucket mode option, but I know that may generate its own world of issues.  That, or some kind of corona splitter option built-in that would properly calculate the scene.

I'll see if I can run it without a frame buffer.  Had to do that with VRay on the same scene.

[hoppergrass]

long shot and just a thought, can you use a pre-calculated GI solution from a lower res render? No idea if this will work or not :) 

[JeffPatton]

Tried that and unfortunately it didn't help. :/

[maru]

Hi Jeff, thanks for reporting this. We will definitely have to look into this. I am afraid we will need the exact scene you are rendering to reproduce the issue: https://corona-renderer.com/upload
If you can share it, please do, and let us know what's the file name.

I am really clueless about those different strips, but for running out of RAM you can try the "low mem" option in the devel/debug rollout, hopefully it will help: https://coronarenderer.freshdesk.com/support/solutions/articles/12000021288

[JeffPatton]

Thanks for the tip on the low mem option.  I'll give that a whirl too.  In the meanwhile, I've uploaded a stripped version of the scene (I verified that it still exhibits the issue).  The file name: "Multi-Render_Corona-stripped.zip"

Thanks!

[Xntric]

Not related to the problem.

Awesome render mate!

I'm curious about where you have found that HDRI+Backplate?

[JeffPatton]

Thanks.  The set is from https://hdri-locations.com/details/Mountain+pass+road/0108?fbclid=IwAR3GwN8YHysGvNdZZm8XeaHrgkh62Jaeysghpplzuzrf_-HbrX02CaCArxo

I added the motion blur effect to the backplate using the path blur tools in Photoshop.

[maru]

Thanks for the tip on the low mem option.  I'll give that a whirl too.  In the meanwhile, I've uploaded a stripped version of the scene (I verified that it still exhibits the issue).  The file name: "Multi-Render_Corona-stripped.zip"

Hmm, I can't find the file. Did you upload it through Dropbox or regular upload?

[Xntric]

Thanks a lot Jeff!

Looks awesome. I'm in the process of making a Mercedes c63 AMG aswell, and i'll be satisfied if mine could look half as good as yours.

Good luck, and i hope you find the solution :)

[JeffPatton]

Hmm, I can't find the file. Did you upload it through Dropbox or regular upload?

Regular upload.  It didn't fully complete the upload process last night apparently.  I just uploaded it again and received the successful upload message this time.  Should be there now (same file name).

[maru]

Regular upload.  It didn't fully complete the upload process last night apparently.  I just uploaded it again and received the successful upload message this time.  Should be there now (same file name).

It's here. Thanks. We will look into this.

[rowmanns]

(Internal ID=288116790)