Awesome computer vision conference

[nasu_dengaku]

I went to a computer vision conference today.  It reminded me of how much I like computer vision.  It has hard math.  It requires maintaining an interesting dichotomy between the exactitude of computer science and the fuzziness of the real world.  It is a crucial stepping stone on the path to artificial intelligence.   It bridges the gap between

Comments

[easwaran]

That's pretty impressive if they can recognize the scene fairly well from out of 10,000 hours of video! Most of these things that are very hyped recognize what the person is looking at or thinking about from one of ten choices or something similarly restricted. Of course, reconstructing really open-ended images (as would be required for dreams) is still a far more difficult task. But even something not that good would tell us a lot more about dreams probably - I suspect they're difficult to study since we only have memories of them, and don't really have access to introspective responses while they're going on. Even for lucid dreamers, it's hard to rule out the possibility that there are other dreams going on that they're not conscious of at the time.

As for labeling objects - do you know what leads them to think that a mouse can do that? Clearly mice interact with objects in a similar way to humans, but conceiving of them as objects and "labeling" them are somewhat more refined tasks that aren't really necessary for lots of

[factitiouslj]

covet.com looks like it could be neat, but you're right, it definitely needs a neither button. I also got a couple non-clothing images. Not sure what effect those have.