* Take the transform of the ImageFrame into account when blending
The implementation computes the axis-aligned bounding box after we transform
the corners of the source image, and then iterates through that box
and computes the inverse of the affine transform of the source image.
The samples are taken based on the u/v coordinates, so that the differences
in size/aspect ratio between the images don't matter.
This makes for a much simpler implementation, and gives us the flexibility
to add different filtering methods in the future, for example.
Signed-off-by: Ică Alexandru-Gabriel <alexandru@seyhanlee.com>
* Name the parameters for the blend node properly
This avoids confusion between which one of the images is the `source`
image and which one is the `destination`.
Signed-off-by: Ică Alexandru-Gabriel <alexandru@seyhanlee.com>
* Remove rendundant computation for u/v coordinates
Signed-off-by: Ică Alexandru-Gabriel <alexandru@seyhanlee.com>
* Rewrite the sampling/clamping logic
* Add image frame transform node
* Move transform node to transform module
* Fix a few issues with our transformation logic
* Fix math + do cleanup
---------
Signed-off-by: Ică Alexandru-Gabriel <alexandru@seyhanlee.com>
Co-authored-by: Dennis Kobert <dennis@kobert.dev>
