BillSpradlin
06-30-2004, 01:27 AM
I've seem to have run across a little quirk in Dirtmap, well a quirk in one method of using it. As you all know there are several methods for applying the dirtmap and using it, you can plug it into the material shader slot of the SG for the material of the surface, you can pipe it through a surface shader, you can pipe it through an hsv->rgb.v node into the diffuse of your base shader, or you can just pipe it into the ambiant/incandescence component of your base shader.
This issue that I came across was when applying the Dirtmap into the material shading slot of the SG node. I typically have used it this way and ran a complete pass and then comping it together in post, which works very well, however one thing that I've seem to haven't noticed before is now popping up more blatently. What happens is when using this method is that when an object is translated/rotated across the frame that what seems to be happening is that dirt is being "projected" onto the model and you can see the model pass through this dirt layer. The main occluded areas stay occluded as expected, but a thin layer of noise (even after cranking up the samples to 90+) would translate across the objects surface. I'm guessing this has to do with the way dirtmap calculates occlusion (through ray tracing) since the rays are cast from the camera.
So the solution to this is to pipe the dirtmap through the out color of a surface shader and apply that to your objects. This seems to negate the issue of the filmy noise layer being projected across the surface from the camera.
When I get home I'll try to post some quick test renders to demonstrate what I'm talking about in case any of you are having difficulty visualizing what I'm referring to.
This issue that I came across was when applying the Dirtmap into the material shading slot of the SG node. I typically have used it this way and ran a complete pass and then comping it together in post, which works very well, however one thing that I've seem to haven't noticed before is now popping up more blatently. What happens is when using this method is that when an object is translated/rotated across the frame that what seems to be happening is that dirt is being "projected" onto the model and you can see the model pass through this dirt layer. The main occluded areas stay occluded as expected, but a thin layer of noise (even after cranking up the samples to 90+) would translate across the objects surface. I'm guessing this has to do with the way dirtmap calculates occlusion (through ray tracing) since the rays are cast from the camera.
So the solution to this is to pipe the dirtmap through the out color of a surface shader and apply that to your objects. This seems to negate the issue of the filmy noise layer being projected across the surface from the camera.
When I get home I'll try to post some quick test renders to demonstrate what I'm talking about in case any of you are having difficulty visualizing what I'm referring to.