Region Renders - quality

[james_g_brown]

Good morning,

We rendered a 5000px image over night with a time limit of 7hrs and it did 70 passes.

The next day, we needed to do a region render of a small section of the image.

We hit render and within 4hrs it had done over 400 passes, so when we put the rendered region in to the full image the quality was noticeably different (less noise, sharper etc....)

This is obvious to me. Smaller render  = higher quality....

Could anyone give me an example of workflow to render a region to the same quality as the main fullsize image?

I thought setting the number of passes the same as the full size image, so 70 in this case would work but it was still of a higher quality.

Am I missing something simple?

Cheers,

James

[maru]

I thought setting the number of passes the same as the full size image, so 70 in this case would work but it was still of a higher quality.

This is impossible unless some other settings were changed, or resolution. If you are using UHD Cache, then I think there might be a difference, too. Solution would be to render UHD Cache for the full image, save it, and then reuse it when region-rendering.

[james_g_brown]

Ok thanks Maru.

But would the saved UHD Cache from the full size image create odd effects if the region to render contains new lights, different geometry etc? It used to in Vray if I loaded in saved LCs and IMs and the lighting and geometry was different.

I haven't tried using saved UHDs so not sure how they work exactly....

Thanks as usual....

James

[maru]

Wait, I don't fully understand this.

You rendered scene A in high resolution, full frame.
You rendered scene A in high resolution, as a crop.

Right?

So how could there be any new lights or other objects in the crop? 

[james_g_brown]

So I rendered Scene A full size at 5000px. The architect then wanted a large concrete column removing so we deleted that geometry and rendered just the region where the column used to exist. So this changed the lighting in the scene as you can now see lights on the wall which the column hid before.

So when we drop this new region on top of the original full size image, the column has gone but the quality of the render is higher than the original image.

Hope that makes sense?

[maru]

I such case, the quality after the same amount of passes should be virtually the same unless there was something like direct light shining onto the column, or other thing that could be noisy when it was there. I don't know what this scene looks like, but this replacing some portion of the image with a crop might not make much sense as the column could have contributed to GI in the whole scene and removing it could change lighting a lot.

[steyin]

I thought setting the number of passes the same as the full size image, so 70 in this case would work but it was still of a higher quality.

I do this method, never have an issue. Can't notice if the quality is higher or not on my end when I do so, always looks about the same as the original image surrounding.

[juang3d]

Hi james_g_brown

Are you sure you are configuring Corona to stop after 70 passes instead of time?

I know it's pretty basic, but sometimes the most basic things are the ones we overlook :)

Cheers.

[pokoy]

I guess the geometry changes also changed the balance between direct and GI contribution, in that case sampling may change as well leading to a faster resolving. I'd say this would only happen when the change has a big impact on the overall lighting but who knows, maybe some optimizations would work better now than with the old setup. I guess this is something that only the devs can say, and probably not without the scene or images of both setups.

[danielmn]

I have done similiar test but never really had a problem with the render matching the quality as long as I limit the passes.

If it is a jump in quality a quick solution would be render a region about 20% larger than you need, and then feather the fullsize together with the croped render.  But again I have never seen the problem you are describing.

[moadr]

In general we use region renders a lot with exterior and interior images too when clients request some changes after the final rendering. In general we never have a problem with the region having different overall quality untill we use the same render settings and number of passes.

In this situation my guess is that the problem occurs from removing that coloumn and thus letting more light getting in to the scene. If it makes a big difference in the overall lighting of the scene, than the gi sampling will be of course different and that way it is harder to match the region's quality to the original one.

[atelieryork]

I have noticed on a couple of occasions during testing that if I render a full view at say 1600x1600 then render a region within it, sometimes the light "solution" is somehow better in the region (it appears to throw the same number of "samples" in that smaller region) and the result is therefore slightly different.

I might try to post an example if I come across it again.