Slashdot Mirror
Google Announces Image Recognition Advance
Rambo Tribble writes Using machine learning techniques, Google claims to have produced software that can better produce natural-language descriptions of images. This has ramifications for uses such as better image search and for better describing the images for the blind. As the Google people put it, "A picture may be worth a thousand words, but sometimes it's the words that are the most useful ..."
...is an offline app that compares two images, and if they scale-match, keep the higher resolution one and ditch the smaller one. And runs the comparison over several thousand files (or even hundreds of thousands, or millions) - time is not a factor.
(a scaling deduplicator, if you will).
Is there already such a beast? Anyone?
Political debates have me rolling my eyes so much I think I got optical whiplash. I should sue. - Foamy The Squirrel
Pretty nifty ... wondering if I describe a scene such as from the TFA:
... rather than just display an image from the library that meets those criteria ...
"Two pizzas sitting on top of a stove top oven (with a glass of wine)"
if it can generate an image algorithmically
Hulk SMASH Celiac Disease
Actually, pass.
You are on a beach with a marvelous and intriguing view of the ocean. You see a hut to the north. You see a shell.
I wonder if you could make a Google Glass assistant for the blind using this technology? Like a little earbud that describes stuff in front of you, and distances, and whatnot.
"Describe my surroundings."
"There is a lamp post directly ten feet in front of you. A lovely pizza parlor is off to your right (four out of five stars). There is moderate foot traffic, seven people in the immediate vicinity. There is a man walking towards you smiling. It looks like your friend Greg. There is heavy traffic to your left."
http://xkcd.com/1444/
Especially considering a 1 mega-pixels image in 8 bits gray-scale. That's 1 MB worth of information. Considering 8 letters in average per word (including the various punctuation characters) and 250 words per page in whatever-16-bits character encoding, the image weighs the same as a book of 200 pages.
Not as "advanced" in image recognition as advertised.
Basically they took the output of a common object classifier and instead of just picking the most likely object (which is what a typical object classifier looks for), it leaves in in a form where multiple objects are detected in various parts of the scene. Then they train a neural network to create captions (by giving it training pictures with associated captions).
According to the paper, it sometimes apparently generates a reasonable description. Other times it reads in picture of a street sign covered with stickers and emits a caption like "refrigerator filled with lots of food and drink".
Actually the most interesting thing about it is the LSTM-based Sentence Generator that is used to generate the caption from the objects. LSTM's are notoriously hard to train and they apparently they borrow some results from language translation techniques to attempt to form intelligible sentences.
This is all very googly-researchy in that they want to see what the limits of pure data driven machine learning are (w/o human tuning). This is not a however much of an advance in image recognition as it is an advance in the language for caption construction.
Right now the technology is well behind human cognition. It may say, "two men playing tennis" where a human would notice a few things, do some research, and say "Roger Federer practicing with his trainer for the upcoming Davis cup"
But the cool thing is that the machine will eventually reach, then surpass the human. The computer of tomorrow will say "Federer practicing for the David cup, but his injury will prevent a win. He also needs to start using a nitrogen fertilizer on his lawn."
If video games influenced behavior the Pac Man generation would be eating pills and running away from their problems.
The picture shows a naked man stretching his anus with both hands, to approximately the width of his fist. The inside of his rectum is also clearly visible. Below his gaping anus, his dangling penis and scrotum are visible, as well as a golden ring on the ring finger of his left hand.
Next they need an AI that can describe the smells in a image.
Then you can finally tell if someone on the internet is a dog.
the preceding comment is my own and in no way reflects the opinion of the Joint Chiefs of Staff
One word equals thousand images, for the readers imagine their own versions of the subject.