Latest path tracer testing

[Peter Milner]

I ran this one over night (15 hours and 2000 passes) and noticed a couple of issues.

1. If you look closely you'll see that the whole image is covered in little black dots.

2. The monitors are set as area lights and colour by object. They are on a seperate lighting channel but no matter how high I increase the level of the channel in the editor, the decal remains at the same brightness. The light given off though does increase illuminating the surrounding objects.

 

Peter

 

[Peter Milner]

Another point is that the balance between the bright sunshine coming through the window and the amount of reflection on the glass seems wrong.

[Roy Hirshkowitz]

Looks like a bug. For some reason, I'm not getting it with your last desk + window model that you sent me. Could you send this model + decal?

http://www2.mcneel.com/upload/Upload.aspx

If you've got an .nXtImage of the thing send that as well-- it may help diagnose the monitor problem. If you want to look at this yourself, you can use the read-out at the cursor to see if any of the luminance values change as you manipulate the channels. Try setting the channel to 0, also.

[Peter Milner]

All sent.

Peter

[Roy Hirshkowitz]

I see the problem with the black pixels-- we've got an odd # of scanlines (1300 x 945)-- change that to 946 and they'll go away. How did you set the resolution? (You're not supposed to be able to set either the x or y res to an odd #.)

[Roy Hirshkowitz]

The monitor thing may be a real bug-- I've got time this morning.

[Peter Milner]

I just typed it into the box in the Walkabout window.

Peter

[Roy Hirshkowitz]

Thanks-- that needs to be fixed to round to an even number.

Thinking about the glass reflection now-- I'll check the model.

[Roy Hirshkowitz]

Here's where I am on your issues:

The monitor-- typical flat screens produce around 250-350 cd/m^2-- your estimate of 20 incandescent watts is off by a factor of 100 or so. To see this in the image editor, turn off the first three channels to prevent any reflected light from striking the monitor surface. What's left is the monitor emmissivity. Use the cursor to check the values. Again, a multiplier of ~100 will produce more accurate data-- you can then see what nXt thinks the monitor will look like when backlit by the window. You can also turn the daylight off (channel 0) and look at what it thinks the monitor looks under artificial conditions.

The glass reflection-- still working on this. What's got me a little confused is why the alpha channel seems so low (139). This may be a problem-- don't know yet.

[Roy Hirshkowitz]

May have overstated the 100x a little-- area lights are still directional so if you view the monitor head on you should get the full on 350 cd/m^2. Viewed from off to the side they'll be less bright. You're still underestimating substantially, though.

[Roy Hirshkowitz]

Something still isn't right w/ the reflections of the monitor-- need to do some more digging around here.

[Eric B]

Peter - I have not had to do this in a while so I do not know if Roy and gang have modified the way in which AR5 nXt works with decaled light sources like monitors but when I was asked to have a glowing Plexiglas ceiling (see the CAT scan image here in the example images, 3rd from the end) I had to not only set the cd/ft^2 but also the saturation of the decal image. A shift in image gamma of about x3 or x4 with around a 25 cd/ft^2 created the soft glow effect, i.e. not enough to act like a true light but still give that overall wash of color.

I recall at the time that Roy was going to look into why the image had to be manipulated so much as well. It would seem normal just to place the decal and then assign an intensity of illumination but doing so always bleached out the image excessively. Only by saturating / darkening the image considerably was the wash out countered, even though it had no effect on the intensity of the projected color of light.

I may want to reload that old model and run it through the newest builds to see what happens.

[Roy Hirshkowitz]

That's all still true Eric-- it has more to do with the difference between reflective and emissive color-- emissive color really requires HDR-- trying to do what you did with a decal will require some monkeying around to get it right.

OTOH that's not what's going on here in Peter's model. Still investigating this one.