Lambda Labs Debuts An API For Face Recognition | Hacker News Reader

Lambda Labs Debuts An API For Face Recognition

techcrunch.com

29 pointssinzone12 comments

Cool.

It may be true that Face.com's API is theirs, and they can do what they'd like with it - however, it's a little disconcerting to see a company say on one hand "We love our API developers and we intend to support you in the future" to shutting down their API with haste.

I've already told my friends who relied on Face.com about this service, it's great to see someone saving startups the hassle of reimplementing functionality that will be complex and costly to solve in a fast-paced environment.

I hope more services like this one proliferate in the future!

If this really is "open source," as the TechCrunch article says, where's the source code for their face recognizer? As a computer vision researcher, I'm interested in seeing how it works.

The only thing that's open source here is the client libraries.

EDIT: I misread, looks like the lack of source code is a temporary issue:

  Lambda Labs co-founder Stephen Balaban ... explained to me that the
  company wanted to go the open source route, but planned to eventually
  monetize by charging larger customers for support, as is a standard
  practice for open source-based companies.

Yea, we're talking with a few people who've done this before. Once we get the the license choice squared away, right now we're discussing MIT vs AGPL, we'll publicly launch the source. If you, or anybody else reading this, wants early access, send me an email s@lambdal.com.

http://api.lambdal.com the direct link

I've been using lambda labs' face detect api for a few weeks now. The API calls essentially return the same structured JSON as face.com which has been a huge plus (considering I'm using this as a face.com replacement and I didn't have to change much code to get my old code up and running).

Stephen (sabalaba) was nice enough to give me a sneak-peak of some of his team's upcoming facial recognition features. I can't speak to how the underlying algos compare to face.com's but their APIs have been accurate enough for my needs and they have some really cool ideas in development.

Deleted comment

It looks like it's selecting from only 6 predefined celebrities (none of which are Stallman), so among those, Arnold matched best.

You can train your own album with an arbitrary number of people. And who's stallman? That's St IGNUcius of the Church of Emacs!

> Face.com was closed off, and didn’t share the details of how it operated.

Not so true, at least they gave a talk outlining their methods (I remember they used single-point classification with very complex distance and 3D model fitting) as well as quality numbers (~95% recognition) at Yet Another Conference'2011 in Moscow.

Stephen, great to meet you at Langton Labs a few months ago. Glad to see that you solved this problem.

I'm curious about your strategy of open-sourcing this. (In general, I'm curious about business models for ML.) Do you mind sharing your thoughts?

We're talking with a few people who have more experience than us in managing open source projects, they'll be able to help us optimize it. Since it's a web API, we're venturing through unknown territory.

I think it's probably going to look similar to something like Automattic as far as business model, where the techno-elite are always able to run their own install for free while those looking for more of a seamless 'hosted' experience, will pay.

Nice work Stephen, looks like you guys have gotten a lot done since we talked @Stripe a few weeks ago. Congrats!

Reply on news.ycombinator.com