Hololens facial recognition prototype | Hacker News Reader

Hololens facial recognition prototype

github.com

63 pointsadamloving29 comments

Perfect, I'm looking forward to a world where everyone has their name floating above their heads like in World of Warcraft. Saves you having to remember stuff.

Link it to other information from my contacts as well.

"Hello ... Dave. How is your ... wife ... Kelly? And your ... dog?"

I would love this

Just reading those slight pauses makes me think of Black Mirror

There was a scene in Black Mirror 3x01 with exactly that situation.

And see their status score which is a result of their credit score, Facebook likes, linked in profile and Google searches.

Then you get a whole generation who is obsessed with this score and go to extremes to manipulate them.

Yeah it's going to be awesome /s

Until then we can hopefully agree to a Universal XP and Attribute Protocol, so you can also see peoples' level and skills.

That must be why Microsoft bought LinkedIn.

This detects a face and displays the attributes it calculated about it. That's pretty interesting though from the title I thought it was something that looked up people and displayed their name, etc.

Pretty cool though ultimately it just takes the Halo Lens input, sends it to the Microsoft Face API[1] and displays its results.

[1] https://www.microsoft.com/cognitive-services/en-us/face-api

Yeah, name and other attribute lookup is feasible (and seems inevitable). Also, this prototype only claims to take a snapshot and submit it. Needs to perform the check in realtime against video input.

It does try to identify names, actually. Looking at the source[1], once it has used the Oxford API to find where faces are in the frame (and their approx age and gender, etc), each individual face is then also queried against the MS emotion API and an OpenFace[2] instance. Given the source, it looks like that instance is meant to be trained on staff members and report their organizational hierarchy.

What would be super neat would be ripping out the OpenFace query, and replacing it with a lookup via Facebook[3], so it could actually just show the names for anyone I was FB friends with.

[1]https://github.com/UoA-eResearch/hololens_facial_recognition... [2]https://cmusatyalab.github.io/openface/ [3]https://news.ycombinator.com/item?id=11836884

Looking at the picture, I love "Mustache: 0" haha

That's what caught my eye as well. Wondering if it is boolean or perhaps decimal. Mustache: 0.13

What I want to see is a .05 next to 1.0 mustache and then try to reverse engineer the criteria from there

Who on earth are you to judge what a 1.0 mustache is?

</snark>

It is not my call, I defer entirely to The Machine. We can not understand its reasoning, only attempt to understand how it became what it is. Serious question on this topic though, does anybody know if this facial recognition library has been completely open sourced? Apologies if the answer is glaringly obvious or has already been answered in this thread.

New mustache style: "machine 1.0"

https://www.microsoft.com/cognitive-services/en-us/face-api isn't open source, but https://cmusatyalab.github.io/openface/ is

I want to see a "Moustache: 3".

It's decimal, out of 1

I wonder if any of the researchers evaluate the ethical decisions behind this. Right now it looks naive.

We have a president who is very anti-muslim and Pro-surveillance. What if this falls in the wrong hands? What if it classifies the wrong person?

Snapchat filters are already racist with dark color tones.

I'm arguing that widespread use of advanced facial recognition will be detrimental to society.

I'm honestly astounded that this works. The color camera on the Hololens is 1280x720, making it useless for most tasks. I love this device so much, but man, such a misstep.

Super neat prototype though.

What "most tasks" are you referring to? Just curious since the applications of Hololens don't seem very solidified to begin with.

Well, "most tasks" here is really just the things I've wanted to do -- not exactly an unbiased sample. Things like tracking objects, reading QR codes, matching more complicated hand gestures (than the built-in ones). Unfortunately, a lot of the really interesting things that could be done involve the high-res IR cameras, which aren't exposed in any way to userspace.

Object tracking[1] and QR code reading are definitely possible at that resolution. For QR codes it's going to depend on the distance from it though.

[1] http://introlab.github.io/find-object/

The color camera on the Hololens is 1280x720, making it useless for most tasks

This seems.. somewhat extreme.

I built a (prototype) hand-tracking system using a 800x600 webcam[1]. In Javascript. Using a Flash->Javascript bridge. In 2009. I'm surprised that one can't find useful things to do with something that appears somewhat more capable.

[1] https://github.com/nlothian/FlashCamShim

<sarcasm>Yes, I'm glad people panicked about Google Glass possibly allowing for this, but by the time Hololens gets to try it, it's inevitable and cannot be stopped.</sarcasm>

I think perhaps you didn't click on the link? This calls the Microsoft Face API and simply returns back attributes about the face you're looking at.

Google specifically disallowed apps from doing this (though you could sidelide them I guess), and lots of people complained that it was a useless gesture.

Reply on news.ycombinator.com