Yahoo's repo using Caffe:cpu as neural network, to recognize Not Safe For work images. Ported to a web interface
- Clone repo
- Build the image
docker build -t github.com/stefanolsenn/nsfw . - Start the container
docker run -d -p 8080:8080 github.com/stefanolsenn/nsfw - Create a POST-request against localhost:8080/ with payload:
{"path": "URL_TO_IMG"}response will be a json object{"Score": string}with the score from 0-1. Where 1 is 100% NSFW
There is currently no validation of the url nor the http stream's datatype. If the url isn't pointing to a image, the program wil most likely break. Feel free to submit a pull request :)
It was just a little side-project and isn't tweaked to production. The response time is quite heavy (700-1500ms) on the vm with 1 core assigned. The pretrained model could be optimized by using /~https://github.com/TechnikEmpire/NsfwSqueezenet to decrease the response time, or by scaling the vm vertically.
Copyright 2016, Yahoo Inc. /~https://github.com/yahoo/open_nsfw
Copyright (c) 2014-2017 The Regents of the University of California (Regents) /~https://github.com/BVLC/caffe