| Input + Scribbles | Mask on RGB | Result |
 |
 |
 |
| Lambda = 0.98 | Lambda = 0.90 | Lambda = 0.60 |
 |
 |
 |
| Mask (t=2) | Mask (t=16) | Result (t=16) |
 |
 |
 |
Here we demonstrate the advantage of using diffusion distances in the context of local edit propagation. Note how the yellow scribble did not propagate over all the buildings as they contain many strong black-and-white gradients. With an appropriate value of t, the diffusion distances prove to be more robust to these gradients while still being able to differentiate the sky and the water from the buildings.
Instructions: Roll your mouse over each thumbnail to see a larger version of the image appear below. This makes it possible to easily flip back and forth between the images. Click on a thumbnail for a full resolution version of the image.
Input + Scribbles Mask on RGB Result Lambda = 0.98 Lambda = 0.90 Lambda = 0.60 Mask (t=2) Mask (t=16) Result (t=16)