Bring back motion blur and DOF subdivisions (vray 3.0)

[Rens]

Hi guys! As far as I know the missing bucket issue happens when the cast samples aren't hitting anything in the bucket and VRay thinks the bucket is empty. This happens a lot with motion blurred edges and thin wires.

To get around it you can increase the bucket size as Dave mentioned, in effect including more in the bucket and lowering the chances of it rendering empty. Another option is to raise the min samples so the chances are better for a sample hitting something. Locking the noise might have switched the samples around so that when it's enabled the bucket would render.

I've seen this happen from time to time in all VRay versions, easy to work around though.

[Jussing]

OK, so I got the scene back from support with improved render settings (tweaking a total of three parameters) which work OK - but they still don't produce as good results as my 2.5 example.

I've made another test, where I've bumped up the settings even more to match the simple 2.5 example, and found that I have to increase minimum samples also to get close, but I end up with a 3 times slower render.
(see the attached image, once again with a sharpened image on the right)

Cons of the new motion blur in 3.0:
- You have to tweak a combination of four different parameters instead of one. (the official way is three parameters, but I've added the fourth because I found I had to increase minimum samples also, to get good results. If this is a bug or me not using it right, I take it back).
- Much slower (again, because of the minimum samples)
- If you, after mblur tweaking, turn off mblur, the non-mblur frame now takes long time to render (21 seconds versus 1 second!) (once again, because of minimum samples)

Pros:
- None that I have seen.

There's a pattern there about the minimum samples. Bug? I don't know.

I should add that I have V-Ray 2.5 on my 3dsmax 2014, and 3.0 on 3dsmax 2015.

Please, support, developers or users, provide me with render settings for my demo scene, that makes the V-Ray 3.0 render match the 2.5 render, and preferably a screenshot to boot. (so we're not comparing your 3.0 render with my 2.5 render, wondering about other differences such as CPU, 3dsmax versions, nightly build etc)

Thanks.

[joconnell]

I'd say a lot of things have changed in raytracing from vray 2.x to 3 and something like this mightn't be a good indication of it from a speed point of view since it's such a simple scene.

[Jussing]

I'd say a lot of things have changed in raytracing from vray 2.x to 3 and something like this mightn't be a good indication of it from a speed point of view since it's such a simple scene.

Good point. But I didn't want to clutter the test by including other potential grain sources like Vray lights or blurred reflections... then, for every grainy image I'd post there would be the potential "maybe you should bump up the light subdivisions", "have you tried to increase reflection settings", etc..

[Jussing]

Vlado & support - my point remains. I've been testing your motion blur for a week and found the following conclusion:

Cons with the new motion blur:
- We need to control FOUR parameters instead of one. Three according to Chaos Group, but I've found you also need to tweak minimum samples. Some of these are totally secret and mysterious (who would guess that you need to adjust "color threshold" to get proper motion blur?!) I think this is super bad user interface. I doubt anyone could like this.
- Render times are three times slower if you want to get close to the quality with good settings from Vray 2.5 (I know you disagree with this - but all my tests show this behaviour)
- AA settings don't revert back to non-mblur when you turn mblur off, so after you have turned it off, your non-mblur renders are 5-10 times slower than they need to be (not a showstopper, but an obvious, objective set-back in user friendliness)

Pros with the new motion blur:
- None

If you disagree, please show me where I am wrong, using the teapot test scene linkied to earlier. Thanks.

- Jonas

[vlado]

Yes, I tested the teapot scene. You are correct that for the same subdivs, because of the image filter, you will get slightly noisier results in V-Ray 3.0 (which happened in V-Ray 2.x with more AA subdivs as well, by the way).

I am not saying that getting the old settings back is not a possibility; but - first, no-one renders teapots only, and second, normally people usually use more than 4 max AA subdivs - in which case, V-Ray 2.x and 3.x produce identical results.

I am open to persuasion But if you further want to convince me, it would be best to look at an actual production scene.

Best regards,
Vlado

[48design]

I'm also no fan of the missing subdiv parameter of the VRayCamera. Most people I think are satisfied with some grainy motionblur or dof so one could just leave those subdivs as low as possible without raising a parameter that affects every other parameter in the scene and therefor the overall rendertimes...

[RockinAkin]

I'm also no fan of the missing subdiv parameter of the VRayCamera. Most people I think are satisfied with some grainy motionblur or dof so one could just leave those subdivs as low as possible without raising a parameter that affects every other parameter in the scene and therefor the overall rendertimes...

If your AA max was set to 16 subdivs, and your camera was set to 8 subdivs, your camera was effectively set to 16 subdivs.
Whichever Subdiv value was higher (AA Max -or- Camera Subdivs) was the one that was used by the Image Sampler.

DoF and MoBlur are strictly Image Sampler qualities - so it makes sense to control them through the Image Sampler only to eliminate redundancy.

[48design]

Sorry, but this makes no sense. What you say is that I could control the overall image sampler subdivs with the camera subdivs if those had been higher. This is exactly the point: I would never set the maximum subdivs to 16. ^^

[vlado]

I would never set the maximum subdivs to 16. ^^

And yet this is almost exactly what happened under the hood.

Best regards,
Vlado

[48design]

Okay, so you say: in VRay 2.x maximum subdivs of 8 was fine, VRay 3.x needs much higher but the rendertimes are even faster?

[vlado]

I am saying that in the majority of cases you can get the same or better results with shorter render times.

Best regards,
Vlado

[joconnell]

Vlado, just for the sake of general knowledge if we had aa of 1 / 8 and then set our dof subdivs to 12 and motion blur to 16 in vray 2.4, was vray using our max aa 8 value for the usual division routine but then using the 16 that motion blur was assigning for our top aa rate, thus leading to our material and light subdivs being reduced less than if we'd used 1 / 16 right away in the aa sampler?

So the apparently same settings between vray 2.4 and 3 would be:
Vray 2.4 - aa 1 / 4, motion blur and dof subdivs 8 and material samples 16 would be getting aa of 1 / 8 but with material and light subdivs of 4 (16 / 4 - incorrect maths but right idea)
Vray 3.0 - aa 1 / 8 and material samples 16 would instead get the same 1 / 8 aa but reduced material and lights of 2 (16 / 8 - again incorrect maths)

So there's a bit of ui stuff with the division routine that's changing sampling amounts between files and thus making vray 3 appear worse and slower for similar ui settings. By what you're saying they of course aren't the same sampling settings at all, vray 2.4 was in the background getting more AA for all the camera rays from our moblur / dof but it wasn't reducing the secondary rays as much. Vray 3 by setting the motion blur or dof using AA max samples IS dividing the secondary samplers more than what 2.4 would have, or did 2.4 take whichever aa value was highest from the aa sampler / moblur sampler / dof sampler and then use that number as it's value for the division routine?

[vlado]

Vlado, just for the sake of general knowledge if we had aa of 1 / 8 and then set our dof subdivs to 12 and motion blur to 16 in vray 2.4, was vray using our max aa 8 value for the usual division routine but then using the 16 that motion blur was assigning for our top aa rate, thus leading to our material and light subdivs being reduced less than if we'd used 1 / 16 right away in the aa sampler?

Technically, this should be equivalent to min. AA subdivs of 2 and max. AA subdivs of 16. and min. shading rate of 4. However there are some other factors involved in the 2.4 builds (which is what I was generally trying to avoid) so let me check more carefully and I will let you know what I find.

Best regards,
Vlado

[48design]

Thanks for clearing that up. So the suggestion is to upper the minimum subdivs to get better results, right?

By the way: sorry for "attacking" VRay that way but it is really hard to see any benefit if most of the stuff/settings you've learned over the last years seem to be not of use anymore and all the old scenes look like.... not so good like before.