Bitmap blur and filter

[glorybound]

I am assuming turning filter and blur off would be the best for render speed, correct? If a texture looks bad then I could adjust it as needed. Are you turning filter and blur off, or are you leaving it at default?

[vlado]

No, please NEVER turn off filtering. The default values are best for performance. You can decrease the blur a little bit, but it has the potential to increase render times as the renderer will need more AA rays to produce a good image.

Best regards,
Vlado

[glorybound]

Great, I am glad I asked. Thank you!

[dean_dmoo]

Is there a recommended setting for the bitmap blur? Standard it's 1.0 but that seems ridiculous to me. I always tend to set it to 0.1 or maybe lower for the diffuse maps. (reflection/glossiness is another matter).

[Alex_M]

I also tend to set the texture blur to 0.1-0.15. To me detailed textures look crisper that way, especially at longer distances and oblique angles.

[^Lele^]

I also tend to set the texture blur to 0.1-0.15. To me detailed textures look crisper that way, especially at longer distances and oblique angles.

What Vlado said, a million times over.
Further to that, lowering filtering will provide for apparent sharpness only in stills, making a veritable mess of spatio-temporal coherence in animation.
Want better results, use higher resolution textures, and/or a better filtering method (cfr. VRayHDRI loader.).

[DanSHP]

What Vlado said, a million times over.
Further to that, lowering filtering will provide for apparent sharpness only in stills, making a veritable mess of spatio-temporal coherence in animation.
Want better results, use higher resolution textures, and/or a better filtering method (cfr. VRayHDRI loader.).

But then do you lower the blur value in the HDRI loader too?

[vlado]

You could try the elliptical filtering mode instead of lowering the blur value.

Best regards,
Vlado

[Vizioen]

Sooooo I always put my textures in VrayHDRI, and set my filtering to elleptical. BUT for diffuse I always set blur to 0.01 and for reflection and such around 0.3, fur bump I leave it at 1.0.

So in short, this is wrong? Better to leave the blur at 1.0 when using elleptical?

[^Lele^]

Sooooo I always put my textures in VrayHDRI, and set my filtering to elleptical. BUT for diffuse I always set blur to 0.01 and for reflection and such around 0.3, fur bump I leave it at 1.0.

So in short, this is wrong? Better to leave the blur at 1.0 when using elleptical?

Well, what do your tests say?
Try rendering a textured plane with varying filter amounts, *to a specific N.T. and with plenty of max AA subdivs*, and the answer to your questions should be apparent.

[dean_dmoo]

Ah, time for tests and development. Those were the days.....

Need to update my library to vrayhdri anyway at some point. Will try to test the blur values aswell.

A question for the theory; is the blur value relative to the texture resolution or the render output/geometry? For some reason I always assumed it was based on render output / uvwmap size. So if you placed a texture at 1x1x1m the blur value would be relative to that size, and not whether the texture is 2000x2000 (visual blur is 1.0) or 8000x8000px (visual blur is 0.25).

[Vizioen]

Ah, time for tests and development. Those were the days.....

Haha hear hear.

[^Lele^]

here's a simple contact sheet, a scaled-down version of the texture used, and a fullsize crop of its detail.
I placed a 15000x7000 pixels image on a plane, and rendered it very slanted at 2048x1024 under a white domelight, to 0.004 noise threshold.
As the camera encompasses the whole texture in the furthest area of the plane, we see magnification in the bottom part of the frame, and minification (roughly a 7.5:1) at the top.

Notice that low filtering has no temporal coherence on top of everything else, and it will sizzle like mad in animations.
Further, notice how lowering filtering essentially delegates the (recursive!) job to the AA sampler, with potentially explosive results on rendertime (it'll depend on the texture contrast, its lighting conditions and so on).
Elliptical filtering, while being the slowest, is on par with no filtering, while preserving sharpness and maintaining temporal coherence without unduly, and situationally, taxing the AA sampler (see the sampleRate Re crop, in blue)

[^Lele^]

Ah, time for tests and development. Those were the days.....

Need to update my library to vrayhdri anyway at some point. Will try to test the blur values aswell.

A question for the theory; is the blur value relative to the texture resolution or the render output/geometry? For some reason I always assumed it was based on render output / uvwmap size. So if you placed a texture at 1x1x1m the blur value would be relative to that size, and not whether the texture is 2000x2000 (visual blur is 1.0) or 8000x8000px (visual blur is 0.25).

In max, it's much worse than that:
http://help.autodesk.com/view/3DSMAX...0-92646FC8A562
(Very bottom of the page).

I *assume* they really do trilinear filtering (ie. interpolate also across mipmap levels too) and that the Blur parameter biases which mip levels are being interpolated (consider it a power curve...).
So, set it to low and you get just the highest mip size being used, raise it, and it will progressively try and use the lower ones, closer to camera.
The Blur Offset is akin to a gain of said MIP levels being considered, on top of the blur value.

The problem with setting filtering to use the highest mip available (ie. the fullsize picture) at all times is that it turns essentially filtering off, with all the mathematical problems that that entails.

[DanSHP]

here's a simple contact sheet, a scaled-down version of the texture used, and a fullsize crop of its detail.
I placed a 15000x7000 pixels image on a plane, and rendered it very slanted at 2048x1024 under a white domelight, to 0.004 noise threshold.
As the camera encompasses the whole texture in the furthest area of the plane, we see magnification in the bottom part of the frame, and minification (roughly a 7.5:1) at the top.

Notice that low filtering has no temporal coherence on top of everything else, and it will sizzle like mad in animations.
Further, notice how lowering filtering essentially delegates the (recursive!) job to the AA sampler, with potentially explosive results on rendertime (it'll depend on the texture contrast, its lighting conditions and so on).
Elliptical filtering, while being the slowest, is on par with no filtering, while preserving sharpness and maintaining temporal coherence without unduly, and situationally, taxing the AA sampler (see the sampleRate Re crop, in blue)

If Elliptical filtering is using less AA, why does it take longer to render than max bitmap Blur 0.01?Just curious!