opennsfw-standalone 0.0.6

Stand-alone wrapper for Yahoo's OpenNSFW model using ONNX.

opennsfw-standalone

A small library for intergrating yahoo/open_nsfw model directly into an application.

This library uses the ONNX Runtime to run inference against Open NSFW. The conversion from the original Caffe model provided by yahoo/open_nsfw was done as following:

1. Convert from Caffe to Tensorflow using ethereon/caffe-tensorflow.
2. Convert from Tensorflow to ONNX using onnx/tensorflow-onnx.

The image pre-processing routine was approximated and is implemented using Pillow.

Prerequisites

• ONNX Runtime 1.x
• NumPy 1.x

Installation

$ pip install opennsfw-standalone

Usage

Python

import sys

from opennsfw_standalone import OpenNSFWInferenceRunner


inference_runner = OpenNSFWInferenceRunner.load()

for image_filename in sys.argv[1:]:
    with open(image_filename, "rb") as fp:
        nsfw_score = inference_runner.infer(fp.read())
        print(image_filename, nsfw_score)

CLI

$ python -m opennsfw_standalone myimage.jpeg mysecondimage.jpeg
myimage.jpg 0.9997739
mysecondimage.jpg 0.9984438

License

Although this repository is MIT licensed, the model itself is subject to the BSD 2-Clause license Yahoo provided.