Got some recurring irritating video to process? Got a slack channel? This solves your dumb problem.
We have a set of Piper cameras. Piper is one of those IoT security systems where it'll just notify you on a phone or something if it detects something. They've got a handful of ways to detect things: noise, movement, IR, and I think a few other things we never bothered to hook up.
We have a Slack channel for our house. Piper integrates with IFTTT, so we have Slack notifications from our security cameras when something goes wrong. It posts a video to the channel, which is pretty handy, actually.
We have a Roomba. The Roomba trips the motion detector every day. Usually a couple times! And you know how it is with false-positives; you see a bunch of them and the true positives get lost in the noise. The videos being posted to the channel started getting ignored, which makes the whole setup not that useful.
I work at Google in Research and Machine Intelligence. Computers are really good at repetitive boring-ass tasks, and recognizing whether a video contains a Roomba is totally repetitive and boring.
Dumb-ass-problem = Dumb-ass-solution.
Oughtta be that all you have to do is copy Dockerfile.example to Dockerfile, then replace the bunch of tags with your assorted keys and client ids and secrets and whatnot.
Then you fire it up with docker run just like everything else that uses Docker.
It's going to respond to any URL that starts with ift.tt in your channel [that's stupid; just sort of happens to work in this case], with an image from ~10s into the video, then tries to categorize it.
Haha, why are you looking at this section; I'm the guy who made the shitty robot respond to urls that start with 'ift.tt'; what kind of testing do you think I set up here?