Deep Data Progressive GPU support

[beavertron]

Its not a show stopper but would the ability to render deep data with progressive rendering or CUDA be a thing that would be eventually supported. To switch over to CPU setup from GPU rendering a scene, then rendering another pass with CPU buckets isn't the end of the world, but it would be handy to be able to capture this data in the original render. Also by Render ID is great for speed and file size, however sometimes further accuracy is required and so with GPU's would speed this up? Thanks again for your support and amazing work and developments as always. It may not even be possible outside of my understanding with how deep data is generated.

Regards

[vlado]

I don't know yet. It's not trivial to implement (proper) deep support for progressive mode (CPU or GPU).

Can you clarify a bit what you mean by further accuracy than the render ID mode? We have a couple of other modes which might work better.

Best regards,
Vlado

[beavertron]

I don't know yet. It's not trivial to implement (proper) deep support for progressive mode (CPU or GPU).

Can you clarify a bit what you mean by further accuracy than the render ID mode? We have a couple of other modes which might work better.

Best regards,
Vlado

Hi Vlado, I notice a small edging artifact, between objects that are separate objects. When using the Render ID option and viewing the deep output in Nuke, there is a green jagged outline, which would appear to cause this? I have tried some of the other options but render times and file sizes become too big. When using the deep data to produce depth of field in Nuke, using Peregrine Labs Bokeh, this creates a slight problem on these edges, more noticeable when using a low F stop number on a Camera node with the Alembic camera. Its not entirely a show stopper as in animation its not noticeable and by increasing the F stop of the camera, can hide it even further. I will keep experimenting and testing to find a good balance.

Thanks Vlado.

[yolov]

Can you check if the alpha is a hard 1.0 at those deep samples where the problem occurs?

[beavertron]

Can you check if the alpha is a hard 1.0 at those deep samples where the problem occurs?

Hi Yolov,

The Alpha passes is hard 1.0 at those edge samples, all looks ok.

Appear to have cleaned this up. Originally I was rendering the Deep output, then in Nuke removing the RBG from the exr file by outputting this channel into another separate sequence of files, so it doesn't override the other elemental composite that is happening with AOV passes. This is when the problem occurs. However by using the setting in Maya Vray to exclude RGB pass from Deep files, this step is no longer required and the DOF effect looks much cleaner.

Thanks

[yolov]

Thanks for sharing.