I don't think the baby's face is that important

Apropos of my passing mention of that brilliant Hays/Efros scene completion technique, here's "Seam Carving", a very crafty image resizing technique:

PDF with more info here, home page with MOV version of video here.


9 Responses to “I don't think the baby's face is that important”

  1. Rask Says:

    Wow, I want this in my next Photoshop!

  2. Joachims Says:

    My first thought: Wow, cool technique!
    My second thought: Another step away from the idea that images (photos) are representations of observable reality, and toward images as art that only takes inspiration from reality.
    My third thought: Wow, now everybody* can be their own Ministry Of Truth (or Nazi- or Stalin-) picture forger and add or remove people or objects from pictures.

    * as opposed to: only people who have the time and knowledge and brains to do it.

  3. Stark Says:

    The removal method there made my jaw drop. Having done this sort of work in Photoshop on more than a few occasions... just wow. That is amazing.

    Just think... no more vacation snaps with some random person in them! ;)

  4. Microfrost Says:


  5. Arlo Says:

    Mind-dropping and jaw-blowing stuff. Maybe you've already seen it, but there's yet another crazy image research project out there. Check out the Ted Talk on Photosynth: Blaise Aguera y Arcas: Jaw-dropping Photosynth demo

  6. magetoo Says:

    Thank you, Coral Cache, for the mov version.

  7. dvayn Says:

    Certainly an interesting and subtly clever application of gradient information! Interesting, but the fact that it's not completely automatic in terms of it recognizing psychologically important areas of the image is a little dose of reality for those impressed by today's computer vision research projects. Many of the fancy and impressive algorithms that seem to do EVERYTHING, really just do it well for the select choice of scenes that were made for the demonstration. After all, with a digital projector and a digital camera, you too can make a fully automatic 3-D scanner (given the right scene, of course):


    Check the last thesis - tooting my own horn a little, but I thought it would go along with Dan's image research trend :)

  8. Daniel Rutter Says:

    Many of the fancy and impressive algorithms that seem to do EVERYTHING, really just do it well for the select choice of scenes

    The Hayes/Efros scene completion technique also fails badly in certain situations; their PDF has several such cases.

  9. dvayn Says:

    Indeed. Surprisingly enough I actually went through most of the paper when you posted it earlier.

Leave a Reply