Show HN: Automatic image moderation and nudity detection
nullnude.com
nullnude.com
Only odd thing I noticed was that male torsos were flagged for nudity. Is that intentional?
It's semi intentional. We are constantly working on the deep learned models twitching and improving them as we go and the one in the API workers right now is stable, accurate, but doesn't have few major categories learned yet. The newest one knows the difference between female and male torso, and will flag a male torso accordingly - probably increasing a "covered nudity" confidence as it appears.
Example: http://images2.gazzettaobjects.it/methode_image/2014/11/25/V...
returns:
{ ... "data": { "covered_nudity": { "confidence": 1, "result": false }, "nudity": { "confidence": 1, "result": false } } }
That's a great find, I'll add it to the test images. It might be b/w, it might be that shadow above the breast, I'll check and let you know. Usually b/w images passed the tests well, I might have not have enough in the test sets.
I tested it with a bunch of pictures and it was perfect.
On a side note, what programming language would you be using to communicate to the API? We have Python and PHP libs on github already, NodeJS in the works.
TypeError: $ is not a function $( document ).ready(function() {
Maybe jQuery is loaded too late?
That's a valid point on the documentation, we have only included the raw json response for each endpoint, I'll write descriptions for each field in them. Thanks!
I do seldomly check the sandbox queue of images looking for those situations. But I do not know about the user taught neural models. You can't trust the internet to teach your AI as microsoft lately shown us. But I will implement some kind of reporting system. Thanks ilovecocacola!