Chaos Corona Forum
Chaos Corona for 3ds Max => [Max] Bug Reporting => Topic started by: romullus on 2017-10-13, 21:20:08
-
It always bothers me why Corona does so bad job when rendering wire map on grazing angle surfaces. Is it because of filtering bluring? If so, can we have exposed that parameter in wire map GUI, so we can tweak it and try to find optimum value?
-
Ouch. Does changing the image filter to other modes help with this in any way?
-
I don't think that image filtering could help here a single bit, but i'll try it anyway.
Edit: nah, change between various filtering methods and even no filter at all, is barely perceptible.
-
I thought since it's not a texture map it's filtered through AA which would be done in the image sampler... Could be wrong, of course.
-
yup, play with AA or set the resolution higher then downsample
-
Nope and nope. It's wire map's fault, it can't be fixed by image filtering nor by downsampling.
-
Ah, OK.
So to clear things out... Do you mean thicker lines on the sides (grazing angles) of your object?
Is this looking fine? (from C4D latest daily, 1.7 RC8)
(https://i.imgur.com/gS8YJaU.jpg)
-
The only "solution" I've found so far, is to use real world size instead of pixels. Have you tried that?
-
@burnin, yes extremelly thick lines at grazing angles is the thing that bothers me. Most of the time it's barely noticeable, but in certain scenarios it sticks like a sore thumb. Your exmple looks fine, but try to place the camera directly above the bottle and then move it slightly to the side, so side wall of the bottle would be visible at very grzing angle - you should get similar effect as in my example.
@PROH, thank you, world size indeed looks like a solution in that case. It's less convenient to use and in some scenes probably won't give acceptable results, but at least now, i'm pretty happy :]
-
With px size it is expected that the further lines will appear thicker, because they will be the same thickness as the closer ones.
Other than that I think rendering a bit more passes would improve quality because of AA.
-
Those lines are clearly NOT the same thickness. And i don't think that such result is expected. I've attached zoomed in version, so you can better see what's going on. Pix size in this render was set to 1 px, but in output, wireframe came out 9 px wide. No amount of AA would ever deal with this.
-
Ah, yes.
(https://s1.postimg.org/6m3cdqjlwv/c_Wire_Shader-_Width.gif)
Pixel width size is screen space relative. Feels like some kind of border case bug.
Next example shows camera moved by a small margin (~0.1mm)
(https://s1.postimg.org/4g3xryt0qn/wire-samples0000.jpg)
(https://s1.postimg.org/7eg7vh12i7/wire-samples0001.jpg)
I wonder if it's precision issue? Can it be solved? How then & is it worth it?
Had occasionally observed similar in blenders freestyle, bypassed by doing larger viewport render and composited back. Since i never used for more than an informational value (no artistic or other meaning).
PS
It's also the reason why i prefer working with World units (also for bump&displace), as i know better what to expect - i often switch through different previews (formats, angles, resolutions) during workflow. But as it can be seen, both approaches have their pros & cons.
-
I don't think it's borderline case, I'm doing quite a few wireframe rendering and this behaviour bothers me a lot. Issue is less noticeable when choosing sub-pixel line width, but is very evident (and ugly) when line width is ≥ 1 V-Ray can render perfect wireframes in both modes and i believe that Corona should be able too.
-
Yup, i agree & support your effort for improvement. The more i think about it, this should be reported as a bug... & hopefully VRay tech will come to the rescue ;)
-
Ok, something is definitely wrong here.
In the image from the 1st post I thought the thick lines are actually multiple edges.
-
And please:
https://forum.corona-renderer.com/index.php?topic=12432.msg80188#msg80188
Which tried to be a gentle reminder for:
https://forum.corona-renderer.com/index.php/topic,10093.msg64388.html#msg64388
Good Luck
-
And please:
https://forum.corona-renderer.com/index.php?topic=12432.msg80188#msg80188
Which tried to be a gentle reminder for:
https://forum.corona-renderer.com/index.php/topic,10093.msg64388.html#msg64388
Good Luck
hah, good to see that this one gets some upwind :)
-
I know why this is happening - the pixel to world ratio used to "normalize" the width to 1px changes on grazing angles, because you see more of the grazing surface in one pixel. But you would need to compensate for this based on direction, which is currently not possible - the line width changes for all directions, even those not needing it.
So currently it works correctly for perpendicular lines, but not for those parallel with the perspective projection. Looking at the results, it will be probably best to change the way this works.
-
Would it be possible to compute anisotropic pixels depending on angle, or is it too complex? If not possible, then yeah, i would prefer constant width parallel lines at the expense of perpendicular ones, as often those are not visible at grazing angles anyway.
Or maybe we need a swith - favor parallel vs perpendicular? Just kidding :]
-
No changes still.
Any plans?
-
nothing changed so far
-
Logged.
(Internal ID=475481973)