Face recognition + the entire original series of Star Trek (w/ vids) | Hacker News Reader

Face recognition + the entire original series of Star Trek (w/ vids)

facemining.pittpatt.com

92 pointslbrandy34 comments

There was this one guy in The Original Series who appeared quite often as an extra. He was a roman guard in "Bread and Circuses" and also a red shirt in "Trouble with Tribbles." He was also in other episodes as an extra. I'd like to see a full face index, including the extras, so I could figure out how many episodes he was in.

(And, no I'm not talking about David Ross/Lt. Galloway, who appeared in 9 TOS episodes. The extra I'm talking about was uncredited.)

EDIT: I've been browsing Memory Alpha. There are quite a few actors who appeared in a bunch of episodes, some are in more episodes than major characters!

64 eps - http://memory-alpha.org/en/wiki/William_Blackburn

32 eps - http://memory-alpha.org/en/wiki/Roger_Holloway

59 eps - http://memory-alpha.org/en/wiki/Eddie_Paskey

I was going to reply earlier with a google search coming up with a rough count for William Blackburn, but I couldn't figure out if he was your guy.

http://tinyurl.com/cbqrfo (TinyURL'd to preserve formatting)

He's not. In terms of episodes, I think he's even more prolific!

It's only a matter of time before someone creates an algorithm and app that takes publicly available and tagged photos of people (think FB profiles), and then uses the info to identify all of the people in a photo at a political rally, crime scene, or amateur porn site.

It's already under development by several startups, including ours (except ours isn't for gov't uses, but to map your personal photos with names using facebook pictures).

The cool thing about facebook's photos is that with their API, you can pull the name->XY locations of the picture which has a person tagged to it (making the facial detection portion super easy). Then with the set of name -> faces[], training the facial recognition database becomes even easier.

Facebook it self has a even more sinister potential. Imagine they train a facial recognition database for everyone already tagged in photos on facebook. Then have a iPhone app, which you can take a picture of anyone... which would spit out the name/info/school/interests/everything. There is nothing technologically stopping this, and it will be just a matter of time.

(puts on my tin foil hat)

http://www.ted.com/talks/pattie_maes_demos_the_sixth_sense.h...

6:47 minutes into the demo :)

Will this be a net positive or negative for society? Stalking-type use becomes possible, but maybe people behave better if they feel less anonymous.

This will be a huge change for people who live in cities, but are currently effectively disconnected from one another. Imagine even just knowing the name of each person you pass on the street, and knowing that they know your name. Huge social change. I think it's positive.

I wouldn't want a bunch of unknowns taking photographs of me. I don't want to imagine a society that take photos of each other without your permission.

Just go ask politely their name or something if you defenetely need to know.

And the CIA is an investor in facebook.

(Hey, where did my tin foil hat go!)

"""except ours isn't for gov't uses, but to map your personal photos with names using facebook pictures"""

Like http://www.polarrose.com does?

Yeah kind of. Another place that does facebook face mapping is http://www.face.com.

The facial recognition isn't our main feature, just a decoration/extra on top of our actual product/service which is related to photo organization and synchronization.

Definitely. Privacy in general is dead (in the same sense that Microsoft is dead).

Microsoft is dead in the sense you don't have to worry about them.

Privacy is dead in the sense that you definitely have to worry about it. Wearing a small IR LED on a cap should be enough to block most CCTV photos of yourself. Most people won't, but if you care, you can take action.

Both are dead in the sense of inevitability. How much longer do you expect the IR LED trick to work?

Why should I worry about losing my privacy? Certainly there are secrets I need to keep, like my bank account details, but those are of a different nature. I won't be surprised if in a couple of decades, anyone on the planet will be able to find out everything I've done all day, complete with high-definition movies of it happening if they want. And, to be honest, I'm quite ok with that.

At the same time, it'll be trivial to fake high res movies of you doing anything - "photoshop" for movies mapping your physical form onto others etc

Wow you robbed a bank today? Nah it was a 'shopped movie.

I think that lessens the need for privacy even more...

>> "Privacy is dead in the sense that you definitely have to worry about it."

>> "Most people won't,..."

Seems like a contradiction. Fact is, most people don't care.

Yeah, those that don't do anything probably assume privacy as the default, so they don't act at all.

In-Q-Tel, the venture capital arm of the CIA, has invested in a number of companies specializing in facial recognition, and has also been closely involved in Facebook's funding. If you set aside all the conspiracy theories you can generate from that fact, it's actually kind of cool.

Comparing a person's face against a homogeneous collection of millions other faces is an intractable problem. If you place the search in the context of the social network, however, it becomes much simpler - assuming it's not some random face in the background, you probably only have to travel out a couple degrees of separation to find a probable match. Even if it is a random face on the street, you're aided by the geographical context.

I think Facebook is generally more hype than substance, but they do leave the door open for some pretty interesting ideas.

Great marketing, timed for the release of the new Star Trek movie. (Assuming that it's not an old demo, and its just proximity to the movie responsible for it being posted and up-voted here.)

This is just my opinion but I think the polygonal outline (which mimics your logo?) you use to show the face id makes the technology you've developed appear less professional than it should. Perhaps, a more processing-intensive desaturation/alpha blend (like they do in 20/20 with surveillance videos) where you highlight the head area?

The arrowed side of the polygon indicates the direction of motion of the head.

http://www.pittpatt.com/technology/

Right, the polygon makes sense, but I agree with wallflower - it's not very aesthetically pleasing. I think a good designer could come up with something that is both functional and nicer to look at.

Video editor here. I much prefer it to a generic box or circle. Maybe an egg shape would be equally functional and prettier, but I liked the immediacy.

and the moon on a stick too please

People who make amazing technical products, but fail to present them as well as they could, are often disappointed by the outcome.

Anyone guess which algorithms they're using?

Our own.

Our company was founded as a CMU spin-off of the Robotics Institute to commercialize the work of one of our founders. See: http://pittpatt.com/about/

Is there any way to contact you guys in regards to licensing your product? (I tried the contact form on the site, but no dice from it)

I just checked and the web form appears to be working correctly. Someone mentioned we had a glitch recently. Feel free to submit it again.

>Our own.

That doesn't help.

Luckily I also provided a link, which, should you visit, mentions a paper. It might help.

We also have an entire website describing the technology stack built on top of the basic algorithm that allows it to be extended to entire videos for large scale data processing and large scale reasoning. We explain most of the pieces of the puzzle.

See: http://pittpatt.com/technology/

fyi

Some relevant research papers by Dr. Henry Schneiderman

Face View Synthesis Across Large Angles http://www.ri.cmu.edu/publication_view.html?pub_id=5137&...

Semi-Supervised Self-Training of Object Detection Models http://www.ri.cmu.edu/publication_view.html?pub_id=4875&...

A histogram-based method for detection of faces and cars http://www.ri.cmu.edu/publication_view.html?pub_id=3530&...

Probabilistic Modeling of Local Appearance and Spatial Relationships for Object Recognition http://www.ri.cmu.edu/publication_view.html?pub_id=543&m...

Real-time Model-based Visual Tracking http://www.ri.cmu.edu/publication_view.html?pub_id=1034&...

Pittpatt definitely has the best face-detection algorithm I've seen in any commercial/opensource system. The algorithm they use can be found in these papers:

http://www.cs.cmu.edu/~hws/

Maybe someday they'll get ported into openCV to replace the current detector which doesn't handle frontal features well.

Reply on news.ycombinator.com