GPU accelerated DOF in Frame Buffer

[WilleViz]

OK, this might be an crazy idea, but since I have been thinking about it, I might as well share it.

Instead of have to use a ton of AA samples to clean up the blurred parts of DOF, why not fake it good with a GPU accelerated filter in the Frame Buffer? You render a depth pass of course, and then use that to enable a feature like NUKEs ZDefocus:

NUKE filter_nodes/zdefocus

If we get really fancy, the feature might make it possible to change the focus point on the fly in the frame buffer and change the aperture/DOF too. It might require a different sort of depth pass, but I'm not clever enough to know any of that sort.

I understand that a proper rendered DOF is more realistic, but it might not always be needed, and we all love speed. Right?

[vlado]

I haven't seen good post-processed DOF yet (except maybe Richard Rosenman's plugin). VFB+ has a similar feature (not GPU as far as I know), but in my opinion it looks very unconvincing. In any case you can try it and tell me if a similar solution would work for you. I still think this is best left to other tools and not the renderer, but you never know...

Best regards,
Vlado

[WilleViz]

I haven't seen good post-processed DOF yet (except maybe Richard Rosenman's plugin). VFB+ has a similar feature (not GPU as far as I know), but in my opinion it looks very unconvincing. In any case you can try it and tell me if a similar solution would work for you. I still think this is best left to other tools and not the renderer, but you never know...

Best regards,
Vlado

Hi Vlado...

Thanks so much for taking your time to reply.

I know what you mean regarding the general quality of post-processed DOF, but you guys are geniuses, so I thought that I might as well ask.

Yeah, I was looking for the functionality of something like VFB+ DOF feature, but I can see why you prefer the output of something like Richard Rosenman's DOF PRO. That looks very good indeed.

But maybe people just prefer to use post-processed DOF in Photoshop/NUKE/Fusion, and then it would be a waste of your time. What do others think?

[bigbossfr]

Why you do not use shademap to render fast dof ?

[Moriah]

Unlike motion blur, at least for me, there's no really big difference to do DOF in the render. 95% of the time its done in post, alot more control and no need to re-render if the client doesn't like it.

[^Lele^]

+1 to both points made above.
If you want a really quick DoF/moblur/stereo RENDER, use the shademap (currently sitting in the stereo helper.).
And unless you need a very exact, multi-segment motion blur, you're probably a lot more effective setting yourself up a comp which deals with it (i suppose there are examples on the internet for whichever post-app you chose to use) and then feeding it the new material.
While i surely can't vouch for the effectiveness of a given solution, i can surely tell you that i hardly ever saw a shot come out with DoF and Moblur by default: rather, they were rendered with those effects as a very last resort when everything else failed, for whatever reason.
And even then, those shots would give some amount of grief to compers, and their AoVs treatment (read: most data channels become unusable...).

[WilleViz]

Thanks everybody!

I didn't even know about the shademap, and that looks like a great solution. There is so many features and tricks within Vray that it just takes time to learn about them all when you're a new user.