Transparency versus geometry comparison

[romullus]

Did a little test to see what's the best method to make foliage in Corona. Results are self-explanatory.
Opacity texture used, was 8bit RGB tiff. Should i use grayscale image as opacity map, with method B i can get very similar render times to method C, so B is a clear winner for me. Of course, your mileage may vary ;]

[Ondra]

That makes sense. Embree can chew through any number of geometry, as long as it does not include many transparent surfaces. That involves switching over to texture evaluation, calling slow 3dsmax bitmaps, then switching back, etc.

[romullus]

Speaking about slow max's bitmaps, during this test i played a bit with filtering. I was shocked how rapidly performance is droping when one lowers blur parameter in coordinates rollout or add aditional bitmaps to material o_O

[Ondra]

lowers blur? Dont you mean increases?

[maru]

This is interesting, I would never expect such differences in rendering time.

[romullus]

lowers blur? Dont you mean increases?

No, i mean lowering. If i set texture bluring to 0.1 for diffuse, translucency and normal maps, i get about 66% performance in rays/s compare it to default bluring.

[Ondra]

omFg....

[maru]

And what if you disable filtering from Corona's ui? Does it make rendering faster or slower?

[romullus]

Strangely enough, it makes rendring slower than with filtering on and default (1.0) bluring of textures, but faster than filtering - on and bluring - 0.1
Oh my... must be my mistake somwhere.

[Juraj]

I keep the 'B' for my foliage. It's the best of both worlds, gives edge detail of mapped alpha, but still reasonable performance.

The filtering dillema, is just something I hate. Sometimes it gives of superior result, sometimes too shitty artifacts. It's odd how it affects performance.

[steyin]

So then the days of setting blur to default low value are to be gone? Interesting. I always followed Ludvik's advice on that one.

[maru]

I always disable filtering globally. :)

[Juraj]

So then the days of setting blur to default low value are to be gone? Interesting. I always followed Ludvik's advice on that one.

I always found interesting how differently people set this up.

For example:

Bertrand: Filtering OFF, but Blur kept default 1.0
Grant W: Filtering ON, but Blur adjusted to 0.01

Then there is of course, OFF + 0.01, which should in theory give the best result most close to original texture, but that often just produces artifacts, and downright doesn't work for sensitive maps (bump, glossy,etc..).

Now I am interested if this data is something handled purely by 3dsMax, or is it interpreted differently by renderers as this thread suggests ?

[Stan_But]

Bertrand: Filtering OFF, but Blur kept default 1.0
Grant W: Filtering ON, but Blur adjusted to 0.01

Interesting. But if Filtering OFF is not matter what mult. of texture blurring. Cause it is OFF

I use same setup as Maru - Filtering OFF.

[romullus]

Interesting. But if Filtering OFF is not matter what mult. of texture blurring. Cause it is OFF

I use same setup as Maru - Filtering OFF.

I am under the same impression as well.
I think, i gonna do some filtering tests soon and split the thread to separate topics.

[Juraj]

I also thought the same, that if filtering is OFF, blur doesn't matter. But it seems to be it does. Just check the preview ( I have small scene in Vray opened right now and it still makes difference).
I don't therefore think the filtering and blur is connected.

I used filtering OFF for quite some time too, it was the only way how to get good clarity in 2k renders. But when I started rendering more hi-res, and used high-res texture (4k+ with lot's of detail),
filtering off would give me often lot's of noise and artifacts and some materials with reflection/glossiness driven by these textures would appear somehow odd/noisy/artifacty.

And for animations, I have to keep filtering ON otherwise I sometimes get micro-flicker in very detailed textures.

[lacilaci]

But you guys do use filtering on bump maps right? Cause whenever i tried unfiltered bitmaps for bump it's just lot of noise... On other maps i don't turn filtering off, just lower filtering/blur value to 0.3+-.

But I guess I should note here that most of the time i turn off image filtering for corona in vfb rollout/render settings.

[Ondra]

image filtering and texmap filtering are 2 COMPLETELY different things.

BTW: yes, bump mapping needs always some filtering to work, the algorithm is based on it.

[greysheep5]

image filtering and texmap filtering are 2 COMPLETELY different things.

BTW: yes, bump mapping needs always some filtering to work, the algorithm is based on it.

so keymaster, for bitmaps in the bump slot one should always use filtering like pyramidal or summed area instead of none? and what about the blur values?

it was my impression with vray 2.0 that everything worked quite well with almost every map set to filter=none and blur 0,01 so we tried to apply that to corona as well. but as people like juraj stated, the matter is probably a bit more complicated.


the attached image was rendered with corona (only a little post color/contrast/glow) to illustrate a room of an old house we are planning/converting currently (architectural design not visualization mostly)

it´s not meant to be something special or a nice visualization on it´s own but just part of the planning process... anyway, the thing is, in this scene (because it had to cost no time) every single map was set to filtering=none and blur 0,01. even the bumpmaps... we rendered it in 3000px wide and with about 100 passes, turned out quite fine i think, no? i mean especially the (quite dramatic and not very well mapped) bump mapping looks like it works to me?

besides this filtering issues the scene uses every conceivable sin like color correct maps (they are supposed to be slow too?!) for single bitmaps to alter brightness and contrast for using them in different slots of the materials (diffuse, bump, reflection). plus it is lit via an HDR light outside and a few smaller visible and invisible spheres and planes inside... rendertimes for those 100 passes were not very high and i think to get the same out of the box result with vray in reasonable time we would have had a hard time.

so thanks, conclusion again, corona is awesome. just my 2 cents to the topic, not very scientific. ;)

we will experiment a bit more with this whole bitmap filtering stuff i guess, always a pleasure to read how other people do it.

best regards,

chris.

[Ondra]

I dont know what happens internally in 3dsmax when you use bump mapping without filtering. I just know that you cannot evaluate bump map from a point sample - you need to look into its neighbourhood, which is very similar to what filtering does.

[greysheep5]

I dont know what happens internally in 3dsmax when you use bump mapping without filtering. I just know that you cannot evaluate bump map from a point sample - you need to look into its neighbourhood, which is very similar to what filtering does.

thank you keymaster! sounds convincing, even if i am no programmer by any means. ;)

i will look into that and re-render some scenes for more comparisons. anyway, corona bump mapping works very well compared to what i was expecting from bumps coming from vray.

[bouhmidage]

How do you converted the geometry ( with alpha map ) to an object  ??

[druwee]

So then the days of setting blur to default low value are to be gone? Interesting. I always followed Ludvik's advice on that one.

I always found interesting how differently people set this up.

For example:

Bertrand: Filtering OFF, but Blur kept default 1.0
Grant W: Filtering ON, but Blur adjusted to 0.01

Then there is of course, OFF + 0.01, which should in theory give the best result most close to original texture, but that often just produces artifacts, and downright doesn't work for sensitive maps (bump, glossy,etc..).

Now I am interested if this data is something handled purely by 3dsMax, or is it interpreted differently by renderers as this thread suggests ?

interesting indeed. - i have been battling with texture blurriness and concluded that high res textures need the Grant approach and smaller resolution textures can do quite well with Betrands approach

the measure being high res 3000x plus and low res being anything below 2500x

interested to hear the technical logic behind this phenomena. but common sense would tilt towards pixel density and how much work the blur&filtering would need to do per relationship between pixels.

[Demogorgon]

I've always used a summed area filter and 1.0 blur :(

[macrog]

Hi,

is there any script for mass change textures filtering (max 2015) ? I've tried setBitmapParams and massFiltering - but they seems to don't work with Corona materials

thanks in advance

[Bzuco]

If you dont have option to change leaf geometry shape(example B,C from first post), keep leafe geometry as it is and just decrease color depth of leafs alpha image from e.g. 8bit to only 1bit B/W and save it as png. It has only tiny visual impact but same speed improvement like in example B.