Slashdot is powered by your submissions, so send in your scoop

 



Forgot your password?
typodupeerror
×
Software

Seeing the Forest For the Trees 64

swframe writes "A new object recognition system developed at MIT and UCLA looks for rudimentary visual features shared by multiple examples of the same object. Then it looks for combinations of those features shared by multiple examples, and combinations of those combinations, and so on, until it has assembled a model of the object that resembles a line drawing. Popular Science has a summary of the research. I've been working on something similar and I think this accomplishment looks very promising."
This discussion has been archived. No new comments can be posted.

Seeing the Forest For the Trees

Comments Filter:
  • by bezenek ( 958723 ) on Tuesday May 11, 2010 @04:30AM (#32166858) Journal
    David Marr proposed the idea of a primal sketch as the first stage of converting the two-dimensional image on the retina to a full understanding of what is being looked at. This work culminated in a paper published in 1980 called "Theory of edge detection."

    Marr was a faculty member at MIT, so it is appropriate for this work to have been done there.

    For more information, see:

    http://en.wikipedia.org/wiki/David_Marr_(neuroscientist) [wikipedia.org]

    and

    http://www.amazon.com/Vision-Computational-Investigation-Representation-Information/dp/0716715678 [amazon.com]

    -Todd
    • ...will it sort my porn collection according to the models?

  • by djupedal ( 584558 ) on Tuesday May 11, 2010 @04:35AM (#32166890)
    I'VE been working on a similar project as well!

    Maybe if enough of us with the same project interests get together, we can create an accurate summary of the parent!

    You see? We could look at each other's projects for combinations of features shared by multiple examples, and combinations of those combinations, and so on!!??

    This is amazing!
  • by Rogerborg ( 306625 ) on Tuesday May 11, 2010 @04:41AM (#32166916) Homepage
    Honda just gets on with implementing it [youtube.com]. Oh, look, it's even got an automobile analogy: Asimo just did a drive-by on your research.
    • I was impressed the asimo guessed, "Maybe toy car?" when looking at the hand-sized mini-cooper model.
      I did want to see them work on overlapping categories, like: is-a(toy) and is-a(car).

    • Yeah! Fuck the academics! What has research ever brought us, amirite?

  • I've heard of various other approaches -- to two different things, and I'm not sure which one the researchers are mainly going for. Is the goal here to produce a useful vision system for AI, or to get a better understanding of how the brain works? It seems like while these are compatible goals, it's helpful to distinguish them and decide which you care more about.
  • by Viol8 ( 599362 ) on Tuesday May 11, 2010 @04:59AM (#32166990) Homepage

    Even neural nets have to be programmed at some level to exhibit behaviour that the programmers think will allow them to learn the task at hand unless these guys used some sort of genetic algorithm. The article doesn't mention it. Does anyone know?

    Also it doesn't explain whether the system just recognises similar pictures to what its seen before - eg this picture looks like object type 123 (which to a human would be a horses rump) or whether it can combine all views of an object and recognise them all as that object , eg this picture looks like a horse. If its the latter how does it do it - does it have to be shown the object from a large number of angles or can it just infer from a couple of angles what the object would be like from many others?

    • Re: (Score:1, Informative)

      by Anonymous Coward

      ?There's nothing particularly unique about genetic algorithms with respect to learning? All systems operate within the constraints set by their programming.

    • The system is trained with labelled views of pre-categorized images, i.e., "this is one view of a watch, " "this is another view of a watch," etc. What makes it novel is that it is not told what the identifying features of a watch (or any other object are). It figures out for itself that the circular face, the stem, etc. are the distinctive features of a watch.

    • by wurp ( 51446 )

      Some neural nets can learn on their own, without training. It shocked me too the first time I read of it.

      Dammit, I can't find a reference now. The example I read about was classifying plants in broad and more specific types - the only input was data describing each of the plants.

  • dogs etc (Score:4, Interesting)

    by jrraines ( 1808940 ) on Tuesday May 11, 2010 @05:37AM (#32167122)
    I certainly haven't worked in this area but for years have wondered how people including fairly young children recognize a dachshund, a bulldog and a great dane as dogs and other things as goats, cats, etc. Dogs are amazingly varied in shape and size and color. It seems like a VERY hard problem.
    • by Viol8 ( 599362 ) on Tuesday May 11, 2010 @06:14AM (#32167310) Homepage

      Theres no way just from looking at pictures of dogs that you could tell they're all the same species. There are some characterstics that some breads have in common with others (other than the obvious 4 legs etc) but they don't all overlap. With something like this its more than a simple case of pattern recognition - its aquired knowledge.

      • With something like this its more than a simple case of pattern recognition - its aquired knowledge.

        Except pattern recognition *is* acquired knowledge.

        • by Viol8 ( 599362 )

          No it isn't. If someone tells you "that is a dog" then you learn a simple fact without any patterns coming into it apart from being able to recognise what the particular breed of dog looks like in the future.

          • by Creepy ( 93888 )

            And of course you didn't watch the video. What they said is vision has a lot of background processing, whether you realize it or not (maybe it happens during REM sleep? I don't know). It is why we can identify the differences between, say, a dog and a badger or know a Terrier is a dog and a Pug is a dog despite body differences.

            If you walk into a room you've never been in before, how many items can you identify given 5 seconds? 15? 30 trips into the room for 5 seconds? My guess is the more exposure,

      • Learning is multi-sense. We recognize dogs as much from the sounds they make and their behavior as from pure visual appearance.

        Attempts to do reductionist recognition (i.e., one sense at a time) are doomed to mediocrity. This system doesn't even categorize it's own training set - it has to be fed labelled (i.e., pre-categorized) images.

      • There are some characterstics that some breads have in common with others (other than the obvious 4 legs etc) but they don't all overlap.

        And this, my friends, is exactly why you should not use plutonium for baking.

      • Acquired versus earned knowledge could well be a huge factor in getting something like this to work.

        Perhaps rather than using one A.I. system to recognize everything, what's needed are hundreds or thousands of specialized A.I. clusters, all working on specifically recognizing one particular kind of object by gathering as much property data as possible.

        Then, hand that data down to the next tier of A.I. clusters charged with recognizing several kinds of similar objects using pattern recognition on the data fr

    • by sznupi ( 719324 )

      It probably largely relies also on observation of behavioral patterns, most dogs have pretty similar ones. And this is the time when child soaks in, at tremendous speed, the "rules" of social life.

      • by RobVB ( 1566105 )

        I [...] have wondered how [...] young children recognize a dachshund, a bulldog and a great dane as dogs

        It probably largely relies also on observation of behavioral patterns, most dogs have pretty similar ones.

        Or, in layman's terms, dogs go "woof".

    • Interestingly enough, very young children will typically generalize and call anything with 4 legs a "doggy" until they are corrected or shown the distinctions. That ability to generalize is one of the things that makes the human mind what it is. So yeah, it's amazing we can take an exemplar and be able to understand that other similar things of varied shapes sizes and colors are related but yet still understand species distinctions. My guess is that we're naturally very good at recognizing 3 dimensional
    • It's also well known that children tend to have trouble with categorizing animals. Frequently they'll overgeneralize - calling every 4-legged animal a dog, for example - and it's only with constant correction that their category boundaries become adjusted properly.

    • If you were to only identify things by how they look, yes.

      For example, dogs bark and cat's meow.
  • Okay, program-- this is a quark. Do you see it? Do you recognize it? Great. Get to work, and I'll be back in a few weeks to see how you're doing.
  • by jipn4 ( 1367823 ) on Tuesday May 11, 2010 @08:00AM (#32167920)

    Hierarchical models of object recognition are decades old, as are attempts to implement them. This work doesn't yet work better than other engineering solutions, and it isn't obviously any more plausible than other approaches. So, it's a nice start, but it isn't a breakthrough.

  • So now we've got an AI that can classify Justin Bieber as a girl too!
  • by S3D ( 745318 ) on Tuesday May 11, 2010 @09:35AM (#32169090)
    Which should have been included into TFA from the start:
    http://people.csail.mit.edu/leozhu/paper/RCM10cvpr.pdf [mit.edu]
    The main achievement claimed is that no image labeling or any additional data like viewport position was needed, the learning process was completely automated.
  • One obstacle toward progress in this field is how to define an object. Are electrons, atoms, molecules, proteins, cells, leaves, trees, forests and planets all considered to be objects? And who gets to decide - a bunch of undergrad test subjects who draw lines around pictures and give names to each image segment? This algo separates objects and parts, but (from what I can tell, having read the article but not the paper), there's no big reason to say one thing is an object and another thing is a part. Seems

Keep up the good work! But please don't ask me to help.

Working...