• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Will we live to see computer AI that can search images for objects or specific details?

There's a lot of research going on, I know at my school (the MIT AI Lab) they're doing a lot of projects working with this. There's an interest in it, so progress will be made in the coming years. I think we'll definetly see it in our lifetimes
 
Originally posted by: Chronoshock
There's a lot of research going on, I know at my school (the MIT AI Lab) they're doing a lot of projects working with this. There's an interest in it, so progress will be made in the coming years. I think we'll definetly see it in our lifetimes

Oh, excellent. Seems like such a near-impossible thing to accomplish though. Good luck to them.
 
Originally posted by: xboxist
Originally posted by: Chronoshock
There's a lot of research going on, I know at my school (the MIT AI Lab) they're doing a lot of projects working with this. There's an interest in it, so progress will be made in the coming years. I think we'll definetly see it in our lifetimes

Oh, excellent. Seems like such a near-impossible thing to accomplish though. Good luck to them.

I dont agree. I think for the most part this is already in play. Software exists now that can scan imagery and highlight manmade structures. Its not a big leap to go to different scales of analysis. In fact its probably.................
 
Originally posted by: Ogg
Originally posted by: xboxist
Originally posted by: Chronoshock
There's a lot of research going on, I know at my school (the MIT AI Lab) they're doing a lot of projects working with this. There's an interest in it, so progress will be made in the coming years. I think we'll definetly see it in our lifetimes

Oh, excellent. Seems like such a near-impossible thing to accomplish though. Good luck to them.

I dont agree. I think for the most part this is already in play. Software exists now that can scan imagery and highlight manmade structures. Its not a big leap to go to different scales of analysis. In fact its probably.................

I read the same thing not long ago. I believe there is already a search engine out there that does this.
 
Yeah, probably, I think this may be accomplished by using a neural network model.. basically start off by giving the computer an image and telling it that this pic has a "phone", "flowerpot", "window", etc. Give it as many images you can with as much description as you can, then start showing it pics and have it guess what objects are in it. Tell it "yes" or "no" and it will eventually learn with a large enough set of images and metadata. I'm vastly simplifying this, and I'm sure someone here will school me pretty bad, but it's a very interesting concept and I believe that it's possible.
 
Originally posted by: Kev
we can bareley get text search engines to give accurate results...

That's because the people whose information is being searched aren't interested in the accuracy of the search. It seems that something like this would be restricted/controlled enough such that it wouldn't have every porn site entering meta information about their images saying they're flower pots or tea kettles...

-geoff
 
Originally posted by: ggavinmoss
Originally posted by: Kev
we can bareley get text search engines to give accurate results...

That's because the people whose information is being searched aren't interested in the accuracy of the search. It seems that something like this would be restricted/controlled enough such that it wouldn't have every porn site entering meta information about their images saying they're flower pots or tea kettles...

-geoff

they could just as easily fill their images with unrelated crap like they do on text sites
 
Originally posted by: xboxist
Topic Summary: Like if you want to find a picture of a person wearing red sunglasses...
That actually wouldn't be extremely difficult - the color red has its own digital signature. Then you could try to fit a general pattern of what sunglasses would probably look like, and parse through pictures until the computer finds one that's a decent match.

Incorporate that into a knowledge-based system and keep teaching it new things...to search for other objects, landscapes, etc etc. It wouldn't learn on its own but teaching it would eventually make it pretty damn accurate at simple image searches.
 
I recently saw a keynote speech by Steve Jobs, I believe, for Apple. He demonstrated the search capabilities of the newest Mac OS. One thing I found interesting is that he did a search of say, "Bolivia," and one of the results was a map. The cool part about it was that "Bolivia" wasn't in the filename and the map wasn't a text file, obviously. But, the word Bolivia could be found in the picture. I think that's close to what the OP was saying.
 
Originally posted by: AgentEL
I recently saw a keynote speech by Steve Jobs, I believe, for Apple. He demonstrated the search capabilities of the newest Mac OS. One thing I found interesting is that he did a search of say, "Bolivia," and one of the results was a map. The cool part about it was that "Bolivia" wasn't in the filename and the map wasn't a text file, obviously. But, the word Bolivia could be found in the picture. I think that's close to what the OP was saying.


Yeah, that's pretty much what I was getting at. Searching for content within an image that isn't part of the file name. That Bolivia example is pretty good.

Searching for a picture that contains "four people playing volleyball at sunset" is probably way beyond text-finding though. Cool stuff.
 
Back
Top