Harold Baize wrote:

> more difficult to use than just rendering two
> separate images. To see what I mean, try
> rendering the standard benchmark scene
> "skyvase". 

I must admit, that I didn't consider this problem
while designing StereoPOV. Thanks to your Feedback,
Harold, and to the Feedback of some further useres,
that reported similar problems with comparable scenes,
I now have a more clear notion what users may want.

I for my part have the habbit to allways use real world
units in my scenes. In accordance with this, I define
    stereo_base 0.065
and define my objects with 1POV unit = 1 meter

Of course I don't want to force other users into this.

 > Hermann suggested that the problem
 > with skyvase.pov is that the original author
 > chose the camera location by trial and error.

Not the camera location.
The problem when tracing this scene in stereo is,
that the camera is defined at about 1 Unit
while the objects are one order of magnitude larger
(about 50 Units) and as a consequence, are placed at a very
large distance compared with the "direction" vector of the
camera.

In a mono trace, this won't show up, because there only
the aspect ratio and the image angle (focal length) matters.

But in stereo, this is quite different.
When you try to make stereo images of very large objects like
large buildings, mountains or landscapes, you will encounter
some problems. To get the whole view onto the image, you have
to use a large distance. At large distances, the stereo system
won't give much depth reolution anymore, because the separation
between the two eyes, i.e. the "stereo base" is small compared
with the distance.
One common trick is to bring in some foreground objects (trees,
flowers, a person) to show the relations.
Another solution is to increase the stereo base. This will reduce
the visual size of the object but at the same time increase the
depth resolution.

The problem here arises, because I designed all of my stereo cameras
in the Patch in a way, that the stereoscopic window is identical with
the image plane.
As a consequence, if we increase the stereo_base, but let the image
plane at the original dimensions, the both halfimages won't match
anymore. So we have to scale the whole camera as well.

 >
 > Hermann tried and also had difficulty using
 > Stereo POV with the skyvase.pov scene,
 > but with some effort he found settings that
 > worked.

The next step of course is to derive a rule
and than build a "convienience shortcut" into
StereoPOV, that allows to re-adjust the
placement of the window by internally scaling
all camera vectors.

 > I imagine a stereo patch to POV
 > that only requires the user to calculate (or guess)
 > the stereo base separation and provide a
 > location for the stereo window.

I think, the most simple aproach (for the user)
would be:
- you set the distance of the window
- SteroPOV derives the window adjustment
   and sets the stereo_base according to the 1:30 rule.

More advanced users should be able to manualy
set all properties of the stereo camera anyway.


 > I think Stereo POV should work as easily as
 > my old way, no matter what units are used for
 > the camera.
The main reason for me to do the patch was, that
some things are not possible in the "old way".
I mean: Crand,
         Area Lights with jitter
and:    Panoramic and Fisheye camera.

My patch supports this features.

Post a reply to this message