Hi Craig,
The depth map is not an image map, it is a data map. Each single pixel represents one depth information. This map should NEVER be rendered in Antialias quality, because it will render then two depth values, or more, into one value, and that is pretty much certainly something that does not exist in that depth. If AA was written into it, a seam will appear in Ae with a product of the, e.g., Background and foreground object. To an critical eye, that looks like CGish.
As each pixel represents a depth value, the typical suggestion is to oversample, to render out the depth information at least twice in size, and apply it to the scaled up footage in Ae. Then scale down.
The set up of the camera is critical as well, especially when you render in anything but floating point precision. Worse case here, 8bit/channel, as you get nearly no quality representation of the depth. If the Depth Map is set very large, and the parts in question would be in, e.g., 10% of that “space”, you would get only ~26 values for that part, which might not work if that is everything you like to have from “in focus” to the Bokeh part.
In After Effects this information needs to be set to RGB Values, if integer, e.g. PNG 8 or 16 bit was used. Otherwise, and gamma is applied to it and perhaps destroys any precision left. (Interpret Footage, Preserve RGB, Colorspace none).
The problem with any DOF Post blur is given that no partially transparency can be calculated, without a lot of work (taking the scene apart).
Another major problem is given in the 2D information, once the 3D space is flattened, the DOF calculation misses the invisible parts, which the Aperture Sensor/Film combination would have seen. So the edge part of the blur has no complete information and has to repeat those pixels. See example (and the page before!) here:
http://old.cineversity.com/forum/forum_posts.asp?TID=2720&PN=1&TPN=12
Here will see the other problems with DOF, no reflections, difficult transparence handling, with 2D post approaches. Sorry that I get so long on all of that, but DOF is since decades a favorite theme of mine, I own a Patent Pending for a specific Aperture, the “Dual Aperture”. So forgive my detailed response ;o)
Note that any Ae blur combination will fail, e.g., post DOF Blur and Motion Blur. It can be set up, but as both needs to be calculated at the very same time, Ae will not deliver, it calculates on or the other first. Here is the major skill and advantage in the C4D approach, it just works!
The buffer question and the curve, I’m not sure what you have set up, typically I would try to use an Adjustment Layer, which has the Buffer Applied to it.
Buffer-Tunnel, if the camera can’t see the Tunnel, it will see the Spheres. I would expect exactly that. You might have to render out two versions. Normally I would suggest to work with a “Seen By Transparency” set up with a 100% transparent tunnel, but with DOF, I might be more cautioned about that. Switch off all textures and Lights, and the second pass should be fast, especially since you have buffered it.
Let me know if there is any other question.
My best wishes for the project
