PDA

View Full Version : Hot RGB Pixels Driving Me Bananas



spherical
10-21-2015, 09:16 PM
OK, I've been chasing this for two days and counting.... I think I've found a solution - and then it isn't. Getting hot color pixels like confetti on edges of objects. Very evident on trees, mostly because there are a lot of edges, but they appear on modeled objects, too. The pattern is random; changes with every render - both in frequency and location. However, it is evident that it is occurring in distinct horizontal bands across the entire image. The bands shift vertical placement with each render. The first in this comparison was the first limited region render today. It seems clear, but there are 11 hot pixels there:

130476

Stuff I've tried:
Opened scene in 11.6.2 (started in 2015.3) seems less but not gone
x64/x86 no difference
Limit Dynamic Range on/off
GI on/off
AS on/off
Fog on/off
SkyTracer2 on/off
Raised Ray Limit to 64 (shouldn't make any difference, as using clip maps, but I thought I'd try it)
Removed G2 and FPrime master plugins and G2 shaders
Certified that clip map is 1-bit
Changed clip map file format (doesn't apply to modeled geometry, but I thought I'd try it)
Raised default shading and light samples from 1/1 to 8/8
Tried all reconstruction filters

Not a memory problem - using only ~1/5 of 32GB
Not a CPU heat problem
Happens in eight versions of this scene. The first in this series started as another project; swapping out the primary models and keeping the outside environment. Going to try starting a scene from scratch and loading items from the other project.

Thoughts?

OFF
10-21-2015, 10:06 PM
Try to remove all UV maps...

djwaterman
10-21-2015, 10:27 PM
The images aren't clear, hard to diagnose because I don't know what I'm looking at. Can you post a scene with a single problematic element in it, like a leaf or some such?

raw-m
10-22-2015, 01:22 AM
I had something very similar to this and found that changing the Cameras Reconstruction Filter sorted it. It was a while back so might be a different issue now, worth a try, though?

spherical
10-22-2015, 03:32 AM
The images aren't clear, hard to diagnose because I don't know what I'm looking at.

All you need to see are the bright speckles that are not in the same position in each frame but are constrained within horizontal bands across the image that also change their vertical position with each render. Very odd. The band shifts almost looks like a TV raster paint that isn't in synch with the camera that is filming it.

danielkaiser
10-22-2015, 04:17 AM
Could you render out a larger image? Hard to pick-out details.

spherical
10-22-2015, 08:09 PM
Easiest just to sample it up.

130496

Every4thPixel
10-23-2015, 04:21 AM
Can you post a screen shot of all your render and camera settings?

MarcusM
10-23-2015, 04:50 AM
Maybe use "Compositing Buffer Export" and export all buffers kinds to check.

Snosrap
10-23-2015, 06:50 PM
Try removing any HDRI images from the scene and see if they go away.

spherical
10-24-2015, 06:48 PM
Maybe use "Compositing Buffer Export" and export all buffers kinds to check.

I'll try that.


Try removing any HDRI images from the scene and see if they go away.

None in there. Was the first thing that popped into my mind, though, as we've seen this many times.

The biggest conundrum is, whatever the cause, why/how does the pattern change with each F9? That makes no sense. Being constrained to horizontal bands just adds to the confusion. We've seen a lot of odd stuff over the years, but this is just nuts.

djwaterman
10-24-2015, 11:45 PM
The biggest conundrum is, whatever the cause, why/how does the pattern change with each F9? That makes no sense. Being constrained to horizontal bands just adds to the confusion. We've seen a lot of odd stuff over the years, but this is just nuts.

Best to post a scene and see if anyone else gets the same results.

gerry_g
10-25-2015, 03:44 AM
one point poly's especially if radiosity is in use, seen similar in my own images, may be wrong

spherical
10-25-2015, 02:21 PM
In the list... GI on/off makes no difference.

gerry_g
10-25-2015, 05:07 PM
ok but you have some sort of foliage there, I am guessing it is imported meaning maybe you didn't clean it up and check it too well, have you looked for stray one pointers is numerics, I had seen similar with unattached points, Otherwise I don't know, maybe you are looking through the wrong side of geometry in front of the camera.

spherical
10-25-2015, 09:17 PM
Well, I've used those trees since 2006. They were generated in Vue5 and repaired/optimized by myself after import. Took a lot of work to get them right & I looked at everything at least five times. Never had an issue until this. This past year I've moved over to Xfrog foliage, as it is way better. Of course, a lot has happened with the technology since 2006, so it stands to reason that current stuff would be better. I'd have changed these out by now, but with the arcane download set up that Xfrog has (no customer login to a dedicated area -- you have to ask for replacement links when the ones you have expire -- and they are supplied in no particular order (changes every time), so you have to reconcile what you have and resume the tedium... not good, but they don't listen). Beside the point anyway. Thanks for looking at this.

Axis3d
02-11-2016, 02:06 PM
Did you ever solve this problem? I've been getting the same thing. Trying to track it down with no luck. Tried all the above.

Greenlaw
02-11-2016, 02:28 PM
I don't know if this will help but way back in my 'Box Days', we would occasionally find hot pixels in our renders. The cause almost always turned out to be a tiny geometry error.

As a test, try rendering with a tripled version of the mesh and see if it still happens. If the hot pixels go away, check the original mesh--there may be some non-planar polygons or some other issue causing problems.

Greenlaw
02-11-2016, 02:33 PM
As a side note, we'd sometimes see the 'hot-pixels' error only in certain channels, like motion vectors for example, which would cause random streaks to appear in our motion blur. In this case, we just clamped the crazy values in comp. This doesn't work for all channels of course and depending on the severity of the problem, we might have to re-render the whole scene with the fixed mesh.

Sometimes, changing the element from float to integer in compositing might fix the error too. Not an ideal solution but it might work in a pinch. As my former boss at R+H might say "Perfection is good but done is better."

G.

MonroePoteet
02-11-2016, 04:52 PM
It might also be an artifact from PhotoReal motion blur. If you're using Motion Blur, try changing it to Classic or Dithered. If the hot pixels go away, then go back to PhotoReal but bump up the Motion Blur passes.

mTp