As a Nine year old, in early 90’s , an attempt was made to kidnap me. Although I was saved by a good samaritan the culprit was never caught.
Every 8 minutes a child goes missing in india. I want to take this spirit of helping others and really scale it .
Help me understand the tech being used/built. Seems it would difficult to do facial recognition from old photos of kids given how much their appearance can change in just 1-2 years.
Report
Maker
@Percival
* We use Ruby on Rails for API Server and Dashboard : http://helpingfaceless.com
* Python for Analytics (Numpy / Scipy / timeseries / stats modules )
* We are trying to incorporate BigML into our pipeline ( https://bigml.com/ ) to really scale our operation .
* We use Apache storm to create a pipeline for information/photo handling ( cropping, face detection, moderation, match-verification-by-human )
* We have an version one android app and are currently building an iOS app.
* Currently we use openbr to of face-recognition but will be moving to something better soon.
Our Face recognition model for children works best within age group of 9-19 . We verify matches manually and also crowdsource the verification to increase our confidence in the match.
So the process goes like this
Face Recognition ---(match)-----> crowdsourced verification through app -----(top 20%) ------> manually verify ------> create automatic report to be sent to NGO's
You can read more about our tech here:
www.http://machinelearningmaster...
You're tackling a real and big problem so I salute you for that. However once you have collected data how can it be used to help these kids?
Report
Maker
@Percival Thanks for checking us out. We have tie up with Non-governmental organization like S.O.C.H who have ~4 year experience in child rehabilitation . Child rehabilitation is a complex issue, specially mired with cultural and social roadblocks, S.O.C.H help us navigate around them and really provide them help.
*S.O.C.H : https://www.facebook.com/sochfor... .
Moreover we have broken our mission into two:
* Short term goal ( next 1-2 years ): Help missing and rescued children and keep collecting "publicly available" data with strategic partnerships with Government of India and other Government organizations.
* Long term goal ( 3-5 years) : Combat child trafficking using face recognition technology and data analytics. We will provide technology solutions to organizations who work in Anti-Trafficking. The solutions will help them collect and analyze data, over time, from photographs and locations, resulting in pattern recognition.
@Percival " We know we are uniquely placed product, and its has been a challenge to figure out a viable business model around it. Our idea is to create a SAAS business model for organization working in anti-trafficking (250+) to share data 'safely' and use our face-recognition/child-tracking pipeline. In your viewpoint does it sounds like a viable plan? "
@shashank_singh I think the challenge with those organizations is they might not have much budget and potentially move verrrrrry slow. I think it's worth pursuing but in parallel I would also think about how you can get some community behind this to raise $$$. Charity Water is probably the best example of this. I would also wonder how microtransactions can exist in this app. Maybe that's a virtual currency to incentivize users and create liquidity for you.
Report
Maker
@Percival Thanks , we agree . Not sure about crowdsourcing because that would be one-shot thing?
But lately, we have been toying with an concept of virtual currency called "karma" and were thinking of tying up with social networks/ e-commerce portal to back it up. Sort of like "commodity virtual currency".
Moreover our tie-up with Kairos API ( http://www.kairos.com/ ) , provides us unique opportunity to help them make their face-recongiton even better through human feedback.
@shashank_singh Yes not crowdsourcing but an ongoing way for people to contribute. Perhaps subscription based or in-app purchases.
For virtual currency tools check out Koinify or CounterParty.
Report
Maker
Thanks @Percival Checking it out. This discussion has really a correct direction to think into. Would it be okay if we can run our ideas by you ?
Wow this is amazing - crowdsourcing millions, maybe billions of people towards a really important cause. Goodluck!
Report
Maker
@sievakozinsky Thanks man. We really need exposure on global scale and we have been leveraging social media as much as we can with our limited manpower.
if you are interested to know more, "Youth Ki Awaaz" gave us a nice coverage .
* www.youthkiawaaz.com/2014/08/nex...
Or if you want to see how we are using Ruby on Rails + Python to fight trafficking
* http://machinelearningmastery.co...
This is really great. Any idea about whether this could be expanded outside India? Both child trafficking and woman trafficking are big problems in a lot of other parts of the world.
I feel like face recognition alone won't work very well in this situation. There may be several years of age difference between the last official photo of the child and the one captured through the app. I would build a crowdsourcing interface where real people would go match the photos which are sorted by the system via face recognition cues and geo location.
Good luck with it
Report
Maker
Hello @erenbali , sorry for the delayed response, we just got stuck in something rather urgent.
1. Yeah, our long-term vision is to take helping faceless international but before we do scale to 7.1 billion people :) , we need to have working small scale pilot projects as proof of concept. And the one running in Mumbai/India is exactly that. A way to prove to governments and organizations that we are serious and this model works.
Our initial training data only included only children but nothing really stops us from expanding our training data, So we can surely expand to combating human(woman)-trafficking.
2. Its a three layered system : Our current Face-Recognition system has only accuracy in range 50-70% . So we bolster our confidence in matches by passing the similarity-test through our app users. Ex this screenshoot. https://lh5.ggpht.com/2me6WPI5Wr...
So the whole process goes like this
Face Recognition ---------(match)-----> Crowdsourced verification through app ----------(top 20%) ------> Manually verify ---------(success)----> Create automatic report to be sent to NGO's
Thanks for checking us out, and these much needed feedbacks from users keep us honest :)
Furthermore, if you want to share more on email or facebook . We are also available
1. community-manager@helpingfaceles...
2. https://www.facebook.com/helping...
Sorry disregard this question. I see the app is live now.
Report
Maker
@Percival :) . We are continuously iterating over our technology . Our biggest roadblock was low face recognition accuracy and right now we are working with Kairos API to fix that.
I should probably mention, we have been able to save 3 children by now. I know its a small number but we aim to scale the operations next few months, which should be easy considering we operate like a very lean startup.
Ghosted.boo
Ghosted.boo
Ghosted.boo
Ghosted.boo
Ghosted.boo
StudySoup
Carbon Health
Ghosted.boo