New Build (304)

[Roy Hirshkowitz]

Path Tracer Problem

Fixed a problem in which a complex object (one which generates a lot of faces at render time) tagged as a light source can slow the path tracer down substantially.  This slowdown is also observed in the new engines, although not to the same degree. 

AutoCAD ACA 2013 - Known limitation

The standard ACA mechanism for assigning materials to sub-components is not working properly in 2013.  I am actively pursuing a solution to this, including attempting to communicate with Autodesk.  Until this is fixed, avoid using ACA 2013 with nXt if you use ACA objects (walls, windows, etc.).  Other AutoCAD objects, including native AutoCAD objects and custom objects appear to function fine in 2013.

Work continues on engine 4.  This build includes fixing a few crash bugs and resource leaks.  Color backgrounds are now supported.  Alpha objects are supported, although shadows will not be retained in this engine.

[Roy Hirshkowitz]

They resolve nicely on a lot of interiors.  Your curtains are only letting a small percentage of the sunlight in-- so yours is going to take much, much longer.  Open the curtains and you might have something interesting.  Put a picture in the frame, or let it be a real mirror.  304 might make a difference-- this thing is a real moving target right now.

[Garret Diduck]

The window is 2/3 completely unblocked in the model, so there should be plenty of light coming in. I'll play around with it some more and update to 304 as well. 

[Roy Hirshkowitz]

Hmmm... I'm happy to take a run at it if you're willing to upload it.

[Garret Diduck]

Uploaded!

Roy Hirshkowitz said:

Hmmm... I'm happy to take a run at it if you're willing to upload it.

[Roy Hirshkowitz]

Just some messing around.  Lights off, without curtains.  I rendered the second example without the wood material to see the caustics a little more clearly.  Resolving nicely with no worrisome fireflies to speak of.  I'll have a chance later to play with this one some more.  Still unsure of the source of your fireflies but either curtains, lights, or build 303 are likely culprits.

[Roy Hirshkowitz]

I left this one going all day while I've been doing other things.  The caustics from the vase is nice.  The sunlight caustics in the picture frame is still not completely resolved after >11000 passes (!)  Looks pretty good, though.

[Marc Chaumier]

Beautiful processing of the light, nevertheless. The E4 is promising. The contrast and the depth of the shadows was better with E1(comparison with 303). And, after consultation of my wife, the yellow dominant of colors, tarnished them and impoverished the image (on the 2 pictures from Garret).

Now, it seems that you have done a tone correction in the 304, this one look much better! The caustics are so wonderful, even in the picture frame! And the shadows and the colors looks like those of the E1.

The glass of the coffeepot is a little bit dark, not enough transparent.

[Garret Diduck]

That is looking much better with more passes. Besides the mirror not resolving, there are still fireflies on the mug as well. My concern is that, while the overall image is quite good, the render will need to cook a lot longer (than the PT) to get the fireflies to resolve enough to make the image acceptable to a client. To get, say, 15000 passes would take quite awhile to resolve and even then, the image may still have fireflies. 

[Marc Chaumier]

I don't know if I have to open a new thread, but they are 2 or 3 points I have noticed, in the difference between E1 et E4. One of them is very curious. It is about the deal with curved faces. Look at the two pictures, made with the same model, exactly, one with E1, the other with E4.

[George Ioannidis]

Which one is better for you Marc? I think E4 looks more "natural".

[Roy Hirshkowitz]

Anything with transparency may look a little different (sometimes a lot different.)  In general, e4 will be the more accurate engine in these cases since it doesn't rely on a trick called "transparent shadow rays" when dealing with refractive surfaces.  On the other hand, it has a much more difficult problem to solve so convergence may be an issue.

I'm pretty sure the problem labeled "curved faces" is similar to the problem labeled "why" and related to e1's incorrect treatment of rough transparency.  Not completely sure, but I doubt if it really has anything to do with curved surfaces.

Marc Chaumier said:

I don't know if I have to open a new thread, but they are 2 or 3 points I have noticed, in the difference between E1 et E4. One of them is very curious. It is about the deal with curved faces. Look at the two pictures, made with the same model, exactly, one with E1, the other with E4.

[Roy Hirshkowitz]

Careful-- I haven't run the thing with curtains and lights.  Some of the contrast will likely go away.  I can appreciate that you (and your wife) might prefer e1-- but my job is to make the simulation more accurate-- not to make the picture better (that's your job.) 

Marc Chaumier said:

The contrast and the depth of the shadows was better with E1(comparison with 303). 

...

Now, it seems that you have done a tone correction in the 304, this one look much better!