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

 

 

Click to enlarge
 

 

Click to enlarge