Camera Transformations

These morphs were inspired by James' reccomendation that we take as many of our own photos as possible to see what the algorithms are good at doing and what they're bad at doing.
In that vein, I tried morphing between three different camera transformations:

  1. Translation
  2. Rotation around the object
  3. Still-camera rotation
The hope was that the algorithm would correctly interpolate the points inbetween the starts and ends of each transformation. I full expected it to work in each case.
However, it did not.
The still rotation clearly fails, quite hilariously. As said in class, as long as somebody doesn't have two ears one side of the face, poisson filling would work. However, in the case of still rotation, we are essentially moving some features from side_A to side_B of the image, and other features from side_B to side_A. This criss-crossing of offsets confuses the morph.
A solution to this would be calculating offsets in terms of polar coordinates.

Each morph is only morphing between the left-most image and the right-most image in the row. All in-between images are morphs.

Translation
Camera Shift
No Gradient