[Hervé]

Hello...

This is a 20 samples
25 bounces
14 hours


Well, what I am concerned about is the bad AA... really bad for 20 samples....

BTW, it serves nothing to have a nice beta if plugins are not fixed in LW and aparently also in C4D....

Imagine if this was for a client... I'd be in deep s....t

[Hervé]

what I was trying to do...

[hdesbois]

I'm puzzled...
Till now, I thought AA was rather excellent with Maxwell. The frame looks like a 4 point poly also. This AA issue seems to be linked with high contrast speculars. We need some tests. My time is short just now, maybe I'll try later.
And thanks for the apple textures.
HD
edit : maybe this AA issue would be cleared by rendering the picture at higher res since it affects only the narrow elements. Then, beta can help with higher speed meaning you can render higher res. Have you tryed a Fprime render?
HD

[Hervé]

you're welcome... yes Frame was a strange error... it is working right now with UV's... (cubic and planar still act strnage sometimes....

But I did not care too much about the frame...

But look the AA.... it is terrible... I'll try other tests..

I hope AA is better in the beta...

BTW everything is subpatched in that scene...

[Hervé]

another try, this time just with an emitter "replacing" the window.

8 bounces
12 samples in 1h50...

still AA is pretty bad...

Hdesbois, do you remember that F_Prime had the same problem with highly contrasted speculars....

[Hervé]

I really believe this is an issue with the LW plugin....

I mean look here... also hard lighting and round poles....

http://www.maxwellrender.com/forum/viewtopic.php?t=3416



and AA is perfect... so I really think the LW plugin has room for improvments...

BTW looking at the above renders... both AA on the bed structure look the same... first one with 20 samples or last one with 12 samples.... same thing...

Hello Victor... is it fixed in the beta..?

[Hervé]

some tests without the house

[Para]

Just to make sure, these aren't resized images?

[Hervé]

no, unfortunatly, they are not post processed nor resized.... the geometry AA lokks ok, but the reflection AA does not..

I am rerendering it bigger now just to see if maybe it is too small render...

Did someone mentionned something about filtering image not present in alpha version...?

[hdesbois]

Too small to render seems likely. I think I've seen something like that in real life digital photography with high contrast linear details less than a pixel wide. If you look very closely to your reference pic, you can see some bad AA on the curved bar of the bedframe. It's faint and barely visible, but only due to low contrast :
reference :


Render :



My guess is you sometime have this kind of artefacts at low res digital photography.
HD

[Frances]

The geometry on that picture frame looks really messed up in the 3rd to the last image.

[tom]

where is your emitter exactly? if your reflections see emitter edges directly, this may occur.
also how many watts is it and what are details for the aperture?

[hdesbois]

I made some tests :

On this picture, emitters are 0.3mm wide are put on a 800*600 mm black diffuse rectangle. So, they are 1/3 pixel wide, and you can see aa problem. Fprime render is exactly the same. But I can't see how it could be otherwise. Here is what is happening :


The bad thing here is not so much the jagged edge than the lighter pixels where the thin line "change column" in the middle of the drawing. It makes the line appear discontinuous, and I think it would be the same with any digital imaging system with regular square pixels. So, maybe after all your particular AA problem is "physically correct" .
HD

[tom]

but i mean this is a problem with emitter edges only, not the other objects...
because they emit light equally on their geometry, thus you can't smooth out the edges.
and the direct relections of the emitters themselves make nonAA edges appear.

[Hervé]

HDesbois, thanks a lot for your tests ! well indeed if you have detail , the motto is render big... (one can always resize it later)

I understand what you say about pixels changing columns to fake a diagonal, but I thought there would be some math to smooth out the process a bit.

I have the original photo in a magazine printed I guess at 300 DPI, while screens are lower res (72 or 96 dpi) that maybe why...

I have made a bigger render (I'll rerender with the beta, because so far 18 samples in about 17 hours...) andoff course it is better now...

Tom, I have tried with an emitter to see if that would change something, but no. In the first render post, there are no emitters, just an open window nearby. and no sun activated...

No Frances, my geometry is perfect... he he... I have tripled checked. the entire bed structure is one closed mesh. LW all subpatched.

Here is a new render I think a bit closer. I made that one with only 8 bounces. It looks like 8 bounces (default) is the real thing.. no..? I mean the most physically correct...?

BTW, the bump on frame painting is a bit strong... my bad...