Slashdot Mirror

← Back to Stories (view on slashdot.org)

Facebook Guesses What's In Pictures To Help Visually Impaired (cio.com)

Posted by msmash on from the good-guy-Facebook dept.

Reader itwbennett writes: Taking the issue of bad image metadata into its own hands, starting today, Facebook will tell users of screen readers what appears in the photos on their timeline. Jeremy Kirk explains: "To describe the images, Facebook built a computer vision system with a neural network trained to recognize a number of concepts, including places and the presence of people and objects. It analyzes each image for the presence of different elements, and then composes a short sentence describing it that is included in the web page as the 'alt' text of the image."These users are often neglected by technology companies. Which is why it's encouraging to see Facebook address the issue. Twitter also recently took a step to improve the user experience of visually impaired people on its social networking website.

19 comments

  1. Source code leaked by Anonymous Coward · · Score: 5, Funny

    if image.has('eyes')
        return 'cat';
    else if image.has('plate')
        return 'food';
    else if image.compressionArtifacts > 0.5
        return 'motivational poster';
    else
        return 'selfie';

    1. Re:Source code leaked by GrumpySteen · · Score: 1

      else if viewer.is('female')
              return 'dick pic';

  2. What could possibly go wrong? by Prien715 · · Score: 3, Informative

    Google tried this a while ago and came up with hilariously racist results.

    --
    -- Political fascism requires a Fuhrer.
    1. Re:What could possibly go wrong? by blueshift_1 · · Score: 1

      I had forgotten about that. That is the danger with naive association in complex social environment.

    2. Re:What could possibly go wrong? by Anonymous Coward · · Score: 0

      An algorithm can't be racist.

    3. Re:What could possibly go wrong? by Rosco+P.+Coltrane · · Score: 2

      I wonder what's hilarious in those 6 photos: the algorithm correctly interprets 5 out of 6 lo-res photos. For anyone involved in artificial vision research, that's an amazing achievement.

      As to the 6th photo, seriously, political-correctness aside, it doesn't seem far-fetched that a computer would interpret a close-up photo of black people as gorillas goofing around in front of a camera. Hell, I bet a human being with bad eyesight could've made that mistake too. There's nothing racist about it: human beings do look like apes sometimes because, well, that's what they are, And black people can look like gorillas to a computer, because gorillas usually sport a dark skin tone. Me, I'm amazed that the computer recognized an ape at all - be it a real ape or an evolved ape.

      --
      "A door is what a dog is perpetually on the wrong side of" - Ogden Nash
    4. Re: What could possibly go wrong? by Anonymous Coward · · Score: 0

      A) Yes, an algorithm can be racist. For instance:

      While (suspect = suspects++)
      {
            If (suspect.race == black) arrest();
      }

      B) This algorithm wasn't racist but it produced racist results.

    5. Re:What could possibly go wrong? by AtlasEnder · · Score: 1

      Great point. This is important but must be handled carefully.

    6. Re:What could possibly go wrong? by JustAnotherOldGuy · · Score: 1

      An algorithm can't be racist.

      Correct, but the output can appear to be racist or provide racist results, and in many cases that's enough to cause a problem.

      But you're right, mathematics itself isn't racist. It's the product of the computation that can be interpreted as being racist.

      --
      Just cruising through this digital world at 33 1/3 rpm...
    7. Re: What could possibly go wrong? by Anonymous Coward · · Score: 0

      while (suspect = suspects++)
      {
                  If (suspect.race == african_american) arrest();
      }

      FTFY

  3. hmm by mschoolbus · · Score: 1

    Someone run Goatse through it and see what it says.

    1. Re:hmm by Anonymous Coward · · Score: 0

      segmentation fault (core dumped)

  4. "Help the visually impaired" my foot by Rosco+P.+Coltrane · · Score: 2

    All this tells me is that Facebook is actively developing new, innovative ways to invade your privacy, and this particular bit of data mining technology has become reliable enough that they felt it would be good PR to create a feel-good, help-the-disabled feature out of it.

    Somehow Facebook being able to interpret the contents of photos doesn't make me warm and fuzzy...

    --
    "A door is what a dog is perpetually on the wrong side of" - Ogden Nash
  5. Burden on screen readers vs authors by Anonymous Coward · · Score: 0

    W3C made the opposite decision by making the ALT attribute required for the IMG element when all that was needed was a system doing image identification like Facebook's.

  6. Trees? by jandersen · · Score: 1

    Has it been tried on these http://www.funnyordie.com/arti...?

    1. Re:Trees? by Anonymous Coward · · Score: 0

      Could no understand number 10.

  7. Gosh but I so love these NSA/FB collaborations! by WillAffleckUW · · Score: 1

    It is ever so keen to assist our Friend the Computer in finding those who do not worship workplace and crowd image surveillance!

    The Computer is our Friend!

    --
    -- Tigger warning: This post may contain tiggers! --
  8. Using Google to Foil Facebook by idontgno · · Score: 1

    Someone needs to post some Google Deep Dream images and watch Facebook's image recognition algorithm collapse.

    --
    Welcome to the Panopticon. Used to be a prison, now it's your home.
  9. If it works as well as their translator ... by YoungManKlaus · · Score: 1

    this is gonna hilarious!