Google Blogoscoped

Forum

Google’s Internal Company Goals Revisited  (View post)

Niraj Sanghvi [PersonRank 10]

Friday, June 1, 2007
17 years ago5,129 views

>>Google’s paper mentioned launching many user interface updates. Not too surprisingly, they did launch quite a few variants of their homepage indeed in 2007.

Not just their homepage, but the new iGoogle page and themes, as well as the new analytics interface.

Roger Browne [PersonRank 10]

17 years ago #

> we don’t know if they have the world’s top AI lab by now

If it turns out that "imgtype:face" is being done by AI rather than by the Image Labeller game, then Google has a strong claim to have the world's top AI.

> paid content results section

I wonder if that refers to News Archive Search, where many of the older articles are subscription only or pay-per-view, and have the prices clearly shown.

Ionut Alex. Chitu [PersonRank 10]

17 years ago #

<< If it turns out that "imgtype:face" is being done by AI rather than by the Image Labeller game, then Google has a strong claim to have the world's top AI. >>

There are many face detection algorithms that work very well (>99%).

http://www.google.com/search?hl=en&client=firefox-a&rls=org.mozilla%3Aen-US%3Aofficial&hs=r8s&q=%22face+detection%22+algorithm&btnG=Search

Roger Browne [PersonRank 10]

17 years ago #

Google's face detection seems to work with white skins, black skins, toys, cartoon characters, side profiles, small faces, groups, caricatures etc. There are very few false positives (we don't get to see the false negatives).

I can't get seem to get anything like that quality out of any of the online demonstrations for face detection algorithms.

Rob [PersonRank 0]

17 years ago #

"Google's face detection seems to work with white skins, black skins, toys, cartoon characters, side profiles, small faces, groups, caricatures etc. There are very few false positives (we don't get to see the false negatives)."

Um, Roger, you surely don't mean Google's face detection, do you? It would be more accurate to say "the face detection algorithm that Google purchased, through their acquisition of Nevenvision".

Purchasing an algorithms is very different from having the world's top AI lab.

Mysterius [PersonRank 10]

17 years ago #

[put at-character here]Rob: What do you think a simple AI would look like?

Rob [PersonRank 0]

17 years ago #

Mysterius: I have no idea what a simple AI would look like. Do you mean a simple strong AI? A simple weak AI?

My only point was to say that, just because Google bought an algorithm that does face detection, a problem that, as Ionut points out, is largely solved, does not mean that they have the world's best AI lab. Yes, you might now have bought/hired a few people that are the best in the world at solving one particular sub-problem (face detection) in one particular domain (image analysis). But there are hundreds of other sub-problems and hundreds of other domains (for example: what about the tougher problem of face recognition, after you've done the face detection? Is Google the best in the world at that? Better than homeland security? Better than casinos, who regularly use this technology to detect/track unwanted gamblers?)

Even if the Nevenvision algorithm is indeed the best face detection algorithm in the world, achieving 99% accuracy on all kinds of faces where other algorithms only achieve, say, 80% accuracy, that does not mean Google has the best AI lab in the world.

That is my only point. Google's growing AI lab is very good. But to conclude that it has the best lab in the world, because it bought an algorithm that does face detection really well is a bit, well, silly.

Ionut Alex. Chitu [PersonRank 10]

17 years ago #

I think this is one of the best AI labs in the world: http://www.ifi.unizh.ch/ailab/ .

Sebastien Marcel [PersonRank 1]

17 years ago #

I disagree with Rob, Face detection is NOT largely solved. Frontal face detection is almost solved but there are still problems with illumination and occlusions. BUT non-frontal face detection also called Multi-View Face Detection is still very challenging and a hot topic these days in the research community.

Indexing images by faces is a logical step for Google Image and should appear also soon in Picasa, unless automatic face finding is not yet fast enough! As a matter of fact, as all images retrieved by Google have been processed off-line while for indexing my personal photos using Picasa I might need real-time stuff!

Here is another demo of face image indexing (and tagging) using Google Image and a on-the-fly face detection:
http://www.idiap.ch/googleportrait/

You can also search image results for Paris Hilton: http://www.idiap.ch/googleportrait/index.cgi?query=Paris+Hilton
or even Angelina Jolie (http://www.uberreview.com/2007/02/google-portrait-face-detection-system.htm).

But I agree that face recognition is even more challenging.
We plan next to use automatic face recognition using data collected with Google Portrait (image + tags), applying the same technology we transfered to Bananasecurity (www.bananasecurity.ch) and to its free biometric lock screen (BananaScreen).

Forum home

Advertisement

 
Blog  |  Forum     more >> Archive | Feed | Google's blogs | About
Advertisement

 

This site unofficially covers Google™ and more with some rights reserved. Join our forum!