OctaneRender™ 3.0 for LightWave™ - Alpha build 3.00.10.0

[abstrax]

When region rendering with F9 render does not stop when maximum samples are reached. Instead samples go on and on and on.

Capture.PNG

Yes, the behaviour was changed compared to v2. Since the max. samples per pixel are for the whole image and region samples are counted separately now (as requested by multiple users), the region rendering doesn't stop until it has integrated 256000 per pixel for the region. We don't include the region samples in the film samples/pixel statistics since it was mostly confusing people. The changed behaviour also allows you to refine regions after rendering has stopped because the max. samples per pixel was reached.

[BorisGoreta]

But how can I render animations with region rendering then if there is no maximum samples ?
What if I want to change a part of animation ? This happens when I need to change 10% of the frame throughout say 50 frames.

[abstrax]

But how can I render animations with region rendering then if there is no maximum samples ?
What if I want to change a part of animation ? This happens when I need to change 10% of the frame throughout say 50 frames.

You can't and you couldn't in the past either, because the samples per pixel you specified were for the whole image and not only for the region. If your region covered say 25% and you specified 1000 samples/pixel, it was actually rendering 4000 samples/pixel.

[BorisGoreta]

So was this an eye opener or you think it's fine as it is ?

You said I couldn't then said I could but I got higher samples then specified so which is it ?

Don't you think samples per pixel should be calculated differently meaning regardless of limited region we should get the same image quality as in the whole frame render.

I am not sure if we are talking about the region rendering tool which we can define and move about while the image is rendering.

I am talking about a static region rendering where only this part of the image is rendered and everything else is black.

Don't you think we should have a capability to render image sequences with limited region with defined image quality ?

[FrankPooleFloating]

Marcus, I am with Boris on this. We both (pretty sure he did too) used to do cropped chunks of animations without needing to change samples, the tweaked render would match perfect, and were able to drop them right into AE etc and be done... so a client could say "remove this little thing over here", and it was a simple fix, and rendered damn fast, because we were only doing a fraction of whole image. In a perfect world, we get both region rendering for cleaning up noise in dark corners AND region cropped, at matching samples as if whole image.

[abstrax]

So was this an eye opener or you think it's fine as it is ?

No, but it's really awkward to define a useful interface for this. If you've got a good idea how it could be done - especially in the Standalone - I'm all ears. Remember there are various use cases of region rendering - not only patching up existing renderings.

You said I couldn't then said I could but I got higher samples then specified so which is it ?

Don't you think samples per pixel should be calculated differently meaning regardless of limited region we should get the same image quality as in the whole frame render.

I am not sure if we are talking about the region rendering tool which we can define and move about while the image is rendering.

I am talking about a static region rendering where only this part of the image is rendered and everything else is black.

See, that's the thing, there is only one region render tool.

Don't you think we should have a capability to render image sequences with limited region with defined image quality ?

How about this approach:

When you draw a region while rendering, rendering stops when 256000 samples/pixel have been rendered for the region or when the region changed / got removed. The region samples are counted separately, the film samples/pixel don't change and max samples/pixel of the kernel node don't apply.

As soon as there is a film wipe (for example due to a change in the scene), the region samples will not be counted separately anymore and we render only up to the maximum samples that were specified in the kernel.

Usually, when there is a film wipe while a region is defined, only the region is wiped, which can be really useful if you want to see the rest of the scene while you are tweaking materials (for example). On the other hand, this would look kinda weird when you render an animation with a render region enabled. So we need a way to set the film outside of the region to black. One solution would be to do a full film wipe when you restart rendering.

A totally different approach to the above would be to allow you to specify the render region in the render target and always render everything black outside of that region. This would require the addition of a second region tool to define this region interactively and a restructuring of the render target node which doesn't have a film pin, where we could add these settings. It also opens a small can of worms, because what should happen, when the region is invalid or when you draw a region using the first region tool (for example to refine some area in the film) that is outside of the region defined in the render target. You see, this second solution requires quite a bit of work and isn't that obvious / easy to use either.

[juanjgon]

When region rendering with F9 render does not stop when maximum samples are reached. Instead samples go on and on and on.

Capture.PNG

As workaround you could render by time, not by samples. It should work also with the limited region enabled.

-Juanjo

[juanjgon]

3d model having vertex normal maps still don't seem to render properly... am i wrong?

It should work. Can you send to me a sample object with problems in the vertex normal map?

Thanks,
-Juanjo

[BorisGoreta]

Well it was quite easy to use region rendering in LW. Just pull the rectangle over the camera view and hit render sequence and I would get a sequence of the region.
This functionality is much more important for me then the interactive region rendering which I never use at all.

Why not lock interactive region rendering if the render sequence started as a region render. This way there is no need for an additional region tool, it is the same one but only it is locked for the duration of the sequence render. In this case the frame render would stop at the given samples per pixels ( but with correct calculation to get the quality of the region as if we were rendering the entire frame )

There is no point in region rendering a region render so I think this makes sense.

Also there could be a distinction between IPR rendering and final rendering. In IPR interactive region has more sense and in final rendering region sequence rendering has more sense.

Also when a still high res rendering is finished and I notice that some object was penetrating another one and I need to fix this I want to use region render ( I don't use interactive region render, I set the region rectangle prior to rendering ) and I want this region to be of the same quality as the full frame because otherwise I get different noise levels and I can't composite this fix over the full frame.

[Lewis]

Guys

How is the stability for you in new 3.0.10 beta ?

I've had so many blue screen or CUDA 700 error crashes in last 2-3 days that i can't even count them ?? I've not had any of crashes with Beta 3.0.6 in 30 days i used it. I've set tonemapping to only MB GPU(s) (and exclude dit form extender GPUs (i have 3 GPUs in MB and 2 in PIC-X extenders 1x) and that solves the slowness of tumbling in IPR (comparing to OR 2.x) but i didn't solve crashing part . I had it rendering for 2 minutes and Bluescreen or sometime 2 hours (same Still image scene with long 3-4h render per frame). I've even tired to turn off some GPUs and try but still crashed just few minutes ago.

Now I'm back to 3.0.6 to see is ti something gone wrong with my machine/GPUs or 3.0.10 is so unstable for some reason.

Do i need newer driver for 3.0.10 ??? I'm on 359.00 drivers and i see newest one is 356.10 on Nvidia pages but i have unwritten "rule" to not change/fix when it works (and it worked fine on 3.0.6) .