Google Makes Huge Leap in AI Image Recognition Advance..

[Tech_Dreamer]

A picture may be worth a thousands words..

 

 

Scientists at Google have created artificial intelligence software that can describe the contents of photographs far more accurately than ever before.

 

Usually the AI software's description of pictures are faulty or very awkward description which is mostly a false description of the given image,  But the later software advances in combination with  language interpretation cluster have made huge advances in describing images more accurately..

 

 

A picture may be worth a thousands words

 

As well as making it easier to search for images, the software could be used to help blind people understand pictures better, Google said.

 

 

Here's an example of an accurate description by the software & how it works...

 

 

Neural networking is a computational model that mimics some of the same architecture used in the brain. Such systems have a series of interconnected neurons which can take information from a variety of sources and are also capable of learning.

 

 

Why is this a huge thing you ask?

 

We as a human being can accurately describing a complex scene with ease because of our brains are made to do exactly that , Our occipital cortex is exactly there for this function primarily & it requires a deeper representation of what’s going on in the scene that a human brain can interpret from years of training & learning through trial & error,

 

Automatically describing the content of an image is a fundamental problem in artificial intelligence that connects computer vision and natural language processing

 

indicated red area is your visual processing unit (Occipital lobe/cortex)

 

 

"A picture may be worth a thousands words,But sometimes it's the words that are the most useful - so it's important we figure out ways to translate from images to words automatically and accurately."

                                                                                                                                                               - Google researcher

 

even when you're reading this your brain is mostly scanning for visual cues & grammatical error inorder to interpret it & respond to the subject (same rules apply to much subtler things like a simple well made advertisement, that's why they make it so good so when we see a product with a good ad & we see it in stores physically or visually like as in online shopping we co relate with the good nature of the product with the ad & make the immediate assumption that it's good even tho we never tried the product before which leads us to the buy) But all this happens in fraction of a second because of the consistent training & the complex other system that co-evolved with the brain works together, in short human brain is not made of silicon.it changes & adapts to the inputs , physically,overtime like acquired talent, the idea is same with practice & practice makes everything work, now you know why..

 

 

That's why the current system could not work with recognition & accurate output because it tried to do all those things single handed,it can process the vast amount of data that can be put in , But to interpret those image & breaking it down & building a natural description of images is a really hard thing to do , The machine-learning software developed by Google used two neural networks  one which deals with image recognition, the other with natural language processing.Still this is a incredibly complex task for these behemoths.

 

 

 

Our experiments with this system on several openly published datasets, including Pascal, Flickr8k, Flickr30k and SBU, show how robust the qualitative results are  the generated sentences are quite reasonable. It also performs well in quantitative evaluations with the Bilingual Evaluation Understudy (BLEU), a metric used in machine translation to evaluate the quality of generated sentences.

 

                                                                                                                                                          - Google  Researchers

 

 

Here's an example of the software 's different version of describing a given event (image below)

 

 

Neural networking is a computational model that mimics some of the same architecture used in the brain. Such systems have a series of interconnected neurons which can take information from a variety of sources and are also capable of learning.
 

The neural network developed by Google was the work of four scientists  Oriol Vinyals, Alexander Toshev, Samy Bengio and Dumitru Erhan.

 

The Below image was fed into the system & was automatically captioned "Two pizzas sitting on top of a stove top oven"

 

 

 

Two years ago Google researchers created image-recognition software and showed it 10 million images taken from YouTube videos. After three days the program had taught itself how to pick out pictures of cats.

 

How it works:

 

This idea comes from recent advances in machine translation between languages, where a Recurrent Neural Network (RNN) transforms, say, a French sentence into a vector representation, and a second RNN uses that vector representation to generate a target sentence in German.

 

if we replaced that first RNN and its input words with a deep Convolutional Neural Network (CNN) trained to classify objects in images, Normally, the CNN’s last layer is used in a final Softmax among known classes of objects, assigning a probability that each object might be in the image. But if we remove that final layer, we can instead feed the CNN’s rich encoding of the image into a RNN designed to produce phrases. We can then train the whole system directly on images and their captions, so it maximizes the likelihood that descriptions it produces best match the training descriptions for each image.

 

EG:

 

The model combines a vision CNN with a language-generating RNN so it can take in an image and generate a fitting natural-language caption

 

This is quite a great achievement in the field of machine learning & artificial intelligence Just keep your fingers crossed like Elon Who's afraid a AI terminator overtake

 

Please Post your thoughts & rants on this down beloooooooooooooow...

 

 

 

Original Paper:

http://arxiv.org/abs/1411.4555

News Blog:

http://googleresearch.blogspot.co.uk/2014/11/a-picture-is-worth-thousand-coherent.html

News link:

http://www.bbc.com/news/technology-30100184

 

Bonus:

 

TLDR caveman edition: Show.Picture.to.Computer...Computer.Talks..Computer.Good...

[kwljunky]

-snip

 

[Tech_Dreamer]

- disturbing graphic gif-

 

Lel! You should've read the TLDR edition at the bottom! :lol:

[kwljunky]

Lel! You should've read the TLDR edition at the bottom! :lol:

My mind was blown by the amount of awesome!

[ganondorf eve]

nice news post B)

[Guest]

It's cool, but it doesn't look like something we're going to see have an effect on Google image search anytime soon.

I remember watching an Nvidia keynote where this (with a lot less data) was being done on a GPU. Doing it on a large scale might take major data centre upgrades from Google.

[LukeTim]

This is awesome. Neural Networks are amazing... the brain is amazing... it's all so amazing!

[Rank]

 

Two years ago Google researchers created image-recognition software and showed it 10 million images taken from YouTube videos. After three days the program had taught itself how to pick out pictures of cats.

 

 

ahahahahahahahaah. people really like cat vids.

[StingBull]

This is awesome. Neural Networks are amazing... the brain is amazing... it's all so amazing!

[smjpl]

 

You've just sent me on a random Small Faces buzz. Thank you

 

OT: AI is cool. Neural networks are amazing. I wonder how many neural parameters this sort of AI requires? I'd say its a lot.

[Constantine]

Well what else would you do when you datamined the majority of people and the internet?

 

Machine Learning of course.

[Beskamir]

Google makes finding high res versions of images that you already found and intend to use for something like a school project quick and easy as it is. This can only get better!

[patrickjp93]

It's cool, but it doesn't look like something we're going to see have an effect on Google image search anytime soon.

I remember watching an Nvidia keynote where this (with a lot less data) was being done on a GPU. Doing it on a large scale might take major data centre upgrades from Google.

You mean Google Brain? That amounts to a honey bee...