Taking very large screenshot from scene

Question

I want to take very large screenshots from my application in OpenGL like 20000x20000 for printing on the banner. First of all, I am not able to create such big framebuffers because of the maximum GPU texture size limitation. Anyone can help me how to capture the framebuffer in different chunks?

Answer 1

As you already noted, capturing it in multiple passes is the way to go. In the simplest form, you can use multiple passes, each rendering a part of the scene.

To get the properr image sub-region, all you need to do is applying another transformation to the clip-space positions of your vertices. This boils down to simple translations and scalings in x and y:

When considering the euclidiean interpretation of the clip space - the normalized device space - the viewing volume is represented by a cube [-1,1] in all 3 dimensions.

To render only an axis-aligned sub-region of that cube, we have to upscale it so that only the sub-region fits into the [-1,1] area, and we have to translate it properly.

Assuming we want to divide the image into an uniform grid of m times n tiles, we could do the following transformations when rendering tile i,j:

1. Move the bottom left of the tile to the origin. That tile position will be at (-1 + 2*i/m, -1 + 2*j/n), so we have to translate with the negated value:

   x' = x + 1 - 2*i/m, y' = y + 1 - 2*j/n

   This is only a helper step to make the final translation easier.

2. Scale by factors m and n along x and y directions:

   x'' = m * x' = x * m + m - 2*i, y'' = y' * n = y * n + n - 2*j

3. The tile is now aligned such that it's bottom left corner is (still) at the origin and the upper right center is at (2,2), so just translate it back with (-1, -1) so that we end up in the view voulme again:

   x''' = x'' - 1 = x * m + m - 2*i - 1, y''' = y'' - 1 = y * n + n - 2*j - 1

This can of coure be represented as simple affine transformation matrix:

(  m     0     0   m - 2*i - 1)
(  0     n     0   n - 2*j - 1)
(  0     0     1        0     )
(  0     0     0        1     )

In most cases, you can simply pre-multiply that matrix to the projection-matrix (or whatever matrices you use), and won't have to change anything else during rendering.