Convergency tests with white-noise initialization
The following experiments show how the optimization converges even when it starts naively from white noise.
Grayscale input
Gradual sequences
Click to enlarge