BUG? Displacement black dots

   14169   9   1
User Avatar
Member
9380 posts
Joined: July 2007
Offline
I can't get rid of black dots in render, color and alpha (see picture)



they only appear when I use displacement shader with True Displacements enabled
the displace bounds parameter has no effect on this
dots appear even if displace amount is set to 0 (see hip file)
what can it be?

Pixel samples seem to affect this, but I must set it to like 10x10 to get rid of almost all dots. In the example file it's no big deal, but in my project I render many passes with the same displacement and increasing pixel samples rapidly increases render time

Any ideas if this can be avoided by some parameter or something?

Attachments:
displace_black_dots.hipnc (107.7 KB)

Tomas Slancik
CG Supervisor
Framestore, NY
User Avatar
Member
4140 posts
Joined: July 2005
Offline
Just took a quick look - your subdivide SOP made the problem worse - it's arguable you need it at all, but even if you do, don't use that and instead go to the Render Tab/Geometry of the object and tag Polygons as Subdivisions. Doing it in the render tends to give better results. I've noticed that setting that and bumping Pixel Samples to 6(which I wouldn't go below as a rule) looks pretty good.

Cheers,

J.C.
John Coldrick
User Avatar
Member
9380 posts
Joined: July 2007
Offline
it still doesn't solve my problem, because the black dots are still there and they are randomly scattered on every frame (because of camera anim) like noise
and when I render some 16 bit or float passes like Depth (in my project i have thickness) and then color correct them, the dots are more than clearly visible

I need to get rid of them completely
is there a reason for them to appear?

there is really no need for subdivide sop in this example, but in my project i have a dense mesh which is like the subd. was already applied, i use however poly to subd. but it doesn't clear things up
Tomas Slancik
CG Supervisor
Framestore, NY
User Avatar
Member
4344 posts
Joined: July 2005
Offline
I think this is a bug.

Looks like the micropolys aren't being stitched properly?

Here is a stripped down scene where P = P; and N = N; in the displacement shader and the geo is being rendered as SubDs.

Attachments:
black_dots.hipnc (172.5 KB)
h95.jpg (14.8 KB)
h82.jpg (14.1 KB)

if(coffees<2,round(float),float)
User Avatar
Member
1002 posts
Joined: July 2005
Offline
I believe the stitching is working (if you turn it off, there are considerably more artifacts).

The problem in wolfwood's scene is that the shading quality has been decreased to 0.1 - this will produce a really coarse displacement that is difficult to stitch. Setting it back to 1 gets rid of most of the artifacts.

In the original scene, the sop supdivision produces a large amount of small polygons - this case is also difficult to stitch. If you're using displacement, it's better to allow mantra to perform the subdivision.

Andrew
User Avatar
Member
9380 posts
Joined: July 2007
Offline
thanks andrew, but why they are there anyways?
when Wolfwood showed result from mantra H8.2 without the dots, with the same settings as in H9.5 with them?

i have a dense geometry with bad topology (many tris, different densities), but with consistent normals and without polygon errors (overlapping, cracks, duplicities …). i am using poly to subdivs and still getting those dots
i am also using glossy refractions and occlusion in some passes so the pixel samples and render quality are a big rendertime hit for me

the sample scene i provided, has 0 amount of displacement, very clean geometry and when you turn off the subd SOP and turn on poly to subd, you will still get the black dots

so can you please confirm that this is normal error with micropoly displacements and the only way to minimalise the dots visibility is increasing shading quality and pixel samples? thanks
Tomas Slancik
CG Supervisor
Framestore, NY
User Avatar
Member
4344 posts
Joined: July 2005
Offline
I attached both hip files.

Both have a Pixel Samples of 1x1 and a shading rate of 0.1. (I set the shading rate to be low in order to make the holes more apparent.)

Increasing the Pixel Samples to 3x3 and the Shading Rate to 4 still causes holes but they are no where near as visible, but they are still there.

Attachments:
blackdots_h95.hipnc (165.7 KB)
blackdots_h82.hipnc (113.7 KB)

if(coffees<2,round(float),float)
User Avatar
Staff
2540 posts
Joined: July 2005
Offline
These holes are a common issue with all micropolygon based renderers. Mantra, RenderMan, and quite a few RenderMan compliant renderers have to wrestle with various coving strategies. RenderMan still generates these holes on occasion fyi. So does Mantra.

When the geometry gets diced in to micropolygons (literally cut apart) there are many cases to handle that are not trivial. It's like sewing up a quilt. You have “T” intersections, multiple edges offset, etc. These conditions are a byproduct of dicing the geometry.

Setting the shading quality as low as 0.1 (approximately 10 pixels per microplygon) you can have some pretty large seams forming on the curved surface that coving may not be able to hide.

Here's a thread for example from AQSIS to give you a hint as the complexity and issues that lie here.
http://72.14.205.104/search?q=cache:vnR18G1yFqsJ:www.aqsis.org/xoops/modules/newbb/viewtopic.php%3Fforum%3D2%26post_id%3D5759%26topic_id%3D809+coving+renderman&hl=en&ct=clnk&cd=2&gl=ca&client=firefox-a [72.14.205.104]

Do a google search using coving micropolygon as your key words.

We want to see all cases where you get holes in Mantra to see if we can solve them. Sometimes it is crazy displacements with discontinuities in the algorithms or in this case, simple geometry but with very low shading quality setting that should only be this low for previs and artefacts are to be expected.
There's at least one school like the old school!
User Avatar
Member
1002 posts
Joined: July 2005
Offline
Jeff is totally right that stitching is a difficult problem.

However, Wolfwood was also right that there was a bug that worsened the issue in H9.x - a fix should be available in the next build of 9.5. Basically, mantra was internally making a slight adjustment to P that would produce displacement cracks even though the displacement shader was not moving P at all.

Andrew
User Avatar
Member
9380 posts
Joined: July 2007
Offline
i just tried 232 build
and the renders are much cleaner now even on lower settings

thank you so much Andrew for such a quick support
Tomas Slancik
CG Supervisor
Framestore, NY
  • Quick Links