[Radiance-general] ripples in my pools...

Fri May 5 11:36:44 PDT 2017

Ah! Nothing better than new source code to fix your problem... Just installed the latest HEAD and everything looking great with rpict version 5.1a<https://www.dropbox.com/s/41rx9ii17qwepgi/51a.jpg?dl=0>.

Thanks for the feedback everyone.

I had originally run with version 4.3a (eesh - old...). Tried with 5.0 official release, still no good<https://www.dropbox.com/s/cx83dihv0uucyf0/50.jpg?dl=0>.

Stefan - glad I wasn't the only one...

Andy - yep, using pcond but as Stephen mentioned, it was visible any way you looked at it.

John - I've been using more and more -aa 0 runs for models with more and more detail. This one doesn't require it, but is a set of tests trying to fine tune settings for -aa 0 runs to be accurate enough and fast enough compared to a 'traditional' run.

Thanks again,

Mike Martinez | Integral Group
Associate Principal | Daylighting
d: 510.457.0107<tel:510.457.0107>
p: 510.663.2070 x2007<tel:510.663.2070%20x2007>

From: Greg Ward [mailto:gregoryjward at gmail.com]
Sent: Friday, May 05, 2017 10:48 AM
To: Radiance general discussion <radiance-general at radiance-online.org>
Subject: Re: [Radiance-general] ripples in my pools...

This artifact appears to be fixed in the CVS head.  It must have been related to sample collision avoidance, which I was doing in a rather ham-fisted way in the first Hessian implementation.  (And yes, it affected results with -aa 0 more noticeably.)  The newer version of ambcomp.c uses a more subtle and successful rejection sampling method, which doesn't bias the original samples.

I put this fix in back in October, but we've been waiting on some unrelated code changes before building the changes into the latest Windows distro.  So, if you're working from that, or you haven't installed the more recent binaries for Linux or Mac, you should do that first.

Cheers,
-Greg

From: Stephen Wasilewski <stephen at coolshadow.com<mailto:stephen at coolshadow.com>>

Date: May 5, 2017 10:02:52 AM PDT

I've gotten this error on multiple projects, It's visible in rvu, rpict results and images made with rtrace.  It is most apparent on -ab 1 so it comes from either the light through the skylight direct from skyglow (with a clear material) or the first ray through the trans material from a direct source.  Small openings have more pronounced patterning compared with larger openings (large opening at top small opening at bottom: https://goo.gl/photos/stsFoix8s4cE3wk46).  It looks like a diffraction pattern and kind of behaves like one (smaller opening, more prominent), does something in the ray sampling explain why the apertures act like diffraction slits?  would it go away with a really high -ad?  is it also there with ambient caching, but the interpolation smooths it out?

Stephen Wasilewski

On Fri, May 5, 2017 at 8:41 AM, Greg Ward <gregoryjward at gmail.com<mailto:gregoryjward at gmail.com>> wrote:

Hi Michael,

Adding back original message, which somehow got lost in my spam folder...

Which version of Radiance are you using?  (I.e., what does "rpict -version" say?)  I found a couple of minor sampling issues in the interreflection code, though I'm not sure if they could cause this effect.  The last fix was a couple of weeks ago, so pretty recent by Radiance standards.

-Greg

P.S.  If you send me your scene in a private e-mail, I'll have a closer look when I get a chance.

From: Andy McNeil <mcneil.andrew at gmail.com<mailto:mcneil.andrew at gmail.com>>

Date: May 5, 2017 8:21:36 AM PDT

Are you using pcond on the image? I had a similar looking issue caused by pcond : https://unmethours.com/question/10506/how-do-i-avoid-banding-when-im-tone-mapping-radiance-images-using-pcond-and-phisto/

Andy

On Fri, May 5, 2017 at 4:12 AM, John Mardaljevic <J.Mardaljevic at lboro.ac.uk<mailto:J.Mardaljevic at lboro.ac.uk>> wrote:
Hi Michael,

Shouldn't really happen, nonetheless perhaps it is possible that there's some weird alias like effect happening with this geometry and the very small ad.

Do your renderings have to be aa 0, and have you tried larger ad values?

Cheers
John

Vote for climate-based daylight modelling:
http://www.lboro.ac.uk/enterprise/enterprise-awards/impact/daylight-modelling/

John Mardaljevic PhD FSLL

----------- Original Message -----------
From: Michael Martinez <mmartinez at integralgroup.com<mailto:mmartinez at integralgroup.com>>
Date: May 4, 2017 at 12:27:18 PM PDT
To: Radiance general discussion <radiance-general at radiance-online.org<mailto:radiance-general at radiance-online.org>>
Subject: [!!SP: THIS MESSAGE IS SPAM] [Radiance-general] ripples in my pools...
Dear list,

I've managed to create unwanted diffuse "ripples" within diffuse pools of light. I'm working with -aa 0 renderings, and having a tough time smoothing things out.

In this example<https://www.dropbox.com/s/fr23tpnv5uqlhxg/clear_section.jpg?dl=0> <https://www.dropbox.com/s/fr23tpnv5uqlhxg/clear_section.jpg?dl=0>,<https://www.dropbox.com/s/fr23tpnv5uqlhxg/clear_section.jpg?dl=0> the diffuse bit of sky component directly below the skylight has subtle ripples of darkness in the pool of light, and also in the bright ceiling above the patch of direct sun...

With a trans material in the skylight<https://www.dropbox.com/s/ptco2powrlhy48x/diffuse_section.jpg?dl=0> <https://www.dropbox.com/s/ptco2powrlhy48x/diffuse_section.jpg?dl=0>, the pattern is even more noticeable, and you can see it mapped onto the walls of the scene as well...

Settings for these runs (which are oversampled x12):
-dp 4096 -ms .14 -ds .1 -dj .9 -dt .05 -dr 3 -ss 16 -st .01 -ab 3 -av 0 0 0 -ar 0 -ad 64 -as 0 -aa 0 -ps 1 -pt .04 -lr 12 -lw 1e-5

Things I've tried with no luck:
- big and small -ar values;  300, 800, 3000 (I recall an old tip from Mark Stock to crank -ar with -aa 0 to get rid of halos?)
- various -ad settings (16, 32 64, 200) while keeping -as 0
-turned on -as with -ad 64 and -as 32
- tried a negative value for -lr
- tried making -lw super small; 1e-10

I'm using the rad program to turn on penumbras - maybe there's something in the -dx, -sx, or -px settings I'm unaware of?

Many thanks for any ideas...

MM
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.radiance-online.org/pipermail/radiance-general/attachments/20170505/d57a6989/attachment-0001.html>