Smartblend Experiment #1 by The Gigapanographer Currently Known as "Kilgore661"

GigaPan Beta + Sony DSC-H7

Autopano has a blending algorithm called 'smartblend' that is good at removing ghosts from gigapans (it is very slow though and only works on small gigapans). I wanted to test the idea that rather than removing the odd ghost, I could use smartblend to remove all the people from a busy street scene. This is the result.

It's a 9x4 which I shot in rows from bottom to top. After taking the whole pano, I re-shot the bottom two rows four times. If someone walked into the frame as a photo was taken then I didn't bother to re-take the shot, I just let the imager carry on and hoped that I'd have another version of the same shot without a person in it.

I chucked the original 36 images and all four sets of the bottom two rows into AP and smartblended. The result was not what I had hoped - it seems smartblend (or its implementation in AP) is not as sophisticated as I thought. However, I made a new gigapan from the top two rows of the original set and then added the most people-free shots from the four sets of the bottom two rows. This is "cheating" as far as my original idea was concerned, but the result is quite good given how busy the scene was.

I will upload a non-smart-blended version of the same scene for comparision in due course.