Recently I've played around with the Seam Carving algorithm for the content-aware image resizing.

It turned out that the algorithm is pretty powerful, elegant but yet simple. One of the interesting parts was that it might be optimized with the Dynamic Progrming approach. Also, it may perform a simple object removal from the image without even modern ML algorithms.

I've tried to describe my experience here in the interactive article:

  • 3
    Avidan is a national hero :) You can't take a CV course in IL without implementing this paper.
  • 3
    Wow, that's cool (and even cooler to see TS in it).

    Now imagine if it could be multi-threaded.
  • 4
    Dude that's some awesome work right there! Amazing!
  • 2
    wow! it just blew my mind :O
  • 1

    Tested it out on a few pictures, results vary..

    Sometimes things end up too short, or too narrow..

    Has a slight Google Streetview vibe to it.

    Legs for example, end up shorter.

    Cars end up narrower.
  • 1
    That is super cool!! I’d love to implement something similar myself.
    Great work!
  • 2
    This has great meme potential
  • 1
    @Nanos Content-Aware is a bit of strong phrase to use here. If I recall correctly, the algorithm removes pixels that are in the "path of less resistance" - basically low color/intensity gradients. So if the pants are single color (not Ali G style), there are no gradients and it will pick those pixels to be removed. Same for the car.
Add Comment