remigagne nous apporta ses lumieres en ce 2007/09/17 14:32:
> Hi all,
> 
> I have a scene that renders correctly at 1000 x 1000 with this true
> orthographic camera:
> 
> camera { orthographic
>  location <-28, -14.5, -28> * 1000
>  right    1000 * x
>  up       1000 * y
>  look_at  <0, 0, 0> }
> 
> I'm now trying to create new 100 x 100 sub-portions of this scene.  For
> example, if I render at 100 x 100 using this camera:
> 
> camera { orthographic
>  location <-28, -14.5, -28> * 1000
>  right    100 * x
>  up       100 * y
>  look_at  <0, 0, 0> }
> 
> I get a 100 x 100 image that's *identical* to the (450, 450) to (550, 550)
> center of the original 1000 x 1000 image, which is perfect.  But how can I
> generalize this to render any portion of the scene?  For example, how can I
> generate a 100 x 100 image centered on the (30, 400) portion of the
> original?
> 
> I think I can translate the camera and look_at point, doing something like:
> 
> camera { orthographic
>  #declare dx = ?;
>  #declare dy = ?;
>  #declare dz = ?;
>  location (<-28, -14.5, -28> * 1000) + <dx, dy, dz>
>  right    100 * x
>  up       100 * y
>  look_at  <dx, dy, dz> }
> 
> I just can't wrap my head around the math required to calculate dx, dy & dz,
> given a center pixel location of (30, 400).  Any advice?
> 
> A little more background: the scene is in a predefined pov file with no
> camera.  I'm generating new pov files programmatically; these simply
> include the original scene and a dynamically calculated camera (based on
> user input).
> 
> Thanks in advance!
> 
> 
> Remi
> 
> 
> 
It would be easier if your camera was parallel to an axis, like the z axis.
You first need to evaluate how many pov-units fit in your 1000 original pixels.
It can be evaluated empiricaly using a plane with a checker pattern. It can be 
calculated from the direction vector that default to 1. It's about the same as 
the distance between the location and the look_at point, or about 42169 
pov-units whide! The field of view extend 0.5 units left and right from the 
camera axis at 1 unit in front of the camera, multiplied by the distance to the 
look_at point.
So, for each 1 pixel of movement, you need to move the camera 42.17 units. You 
could use the vrotate(<Udx,Udy,0>, Your_Angle) macro on the <Udx, Udy, 0> vector 
of the user to get the <dx,dy,dz> translate vector you need.
You can add translate<dx,dy,dz> to your camera after everything else. This way, 
your camera won't change it's orientation: the location and look_at will change 
in exactly the same direction and distance. Easier to maintain and less chance 
of making a typo.

-- 
Alain
-------------------------------------------------
You know you've been raytracing too long when you have ever "Hand-Coded" a 
bezier patch.
Stephan Ahonen

Post a reply to this message