Skip to content

Latest commit

 

History

History
67 lines (64 loc) · 3.12 KB

README.md

File metadata and controls

67 lines (64 loc) · 3.12 KB
Raw

This repository shows how to use Yahoo OpenNSFW pretrained model inside a browser to check user uploaded media.

TL;DR

To run the model inside a brower do as follows

git clone https://github.com/zaghaghi/nsfw-webdnn
cd nsfw-webdnn
emrun --no_browser --port 8081 .

Requirements

Emscripten

If you don't have emscripten install it as follows. You can read more about it here.

git clone https://github.com/emscripten-core/emsdk.git
cd emsdk
export EMSDKPATH=$(pwd)
./emsdk install latest
./emsdk activate latest

Webdnn

WebDNN is required if you need to use you own model instead of OpenNSFW.

conda create -p env python=3.6
source activate env/
pip install numpy setuptools chainer
git clone https://github.com/mil-tokyo/webdnn
cd webdnn
export WEBDNN=$(pwd)
python3 setup.py install
wget http://bitbucket.org/eigen/eigen/get/3.3.3.tar.bz2
tar jxf 3.3.3.tar.bz2
ln -s $(pwd)/eigen-eigen-67e894c6cd8f/Eigen $EMSDKPATH/emscripten/*/system/local/include

Convert OpenNSFW model

I used these commands to convert OpenNSFW pretrained model to WebDNN equivalent. You can read more about WebDNN supported models in their documentations

git clone https://github.com/yahoo/open_nsfw
cd open_nsfw/nsfw_model
python $WEBDNN/bin/convert_caffe.py  --input_name data --input_shape '(1,3,224,224)' --output_names fc_nsfw --out output resnet_50_1by2_nsfw.caffemodel

OpenNSFW Image Transformer

We should preprocess every image before feeding it into network. These lines are the preprocess section in OpenNSFW using caffe.

caffe_transformer = caffe.io.Transformer({'data': nsfw_net.blobs['data'].data.shape})
caffe_transformer.set_transpose('data', (2, 0, 1))  # move image channels to outermost
caffe_transformer.set_mean('data', np.array([104, 117, 123]))  # subtract the dataset-mean value in each channel
caffe_transformer.set_raw_scale('data', 255)  # rescale from [0, 1] to [0, 255]
caffe_transformer.set_channel_swap('data', (2, 1, 0))  # swap channels from RGB to BGR

And these lines are WebDNN equivalent which you can find in script.js

let image_array = await WebDNN.Image.getImageArray(img_src, {
    dstW: 224, dstH: 224,
    bias: [104, 117, 123],
    color: WebDNN.Image.Color.BGR,
    order: WebDNN.Image.Order.CHW,
});

How to server

You can serve this example using this command

emrun --no_browser --port 8081 .

or any other web server that set correct mimetype for webassembly files (.wasm). Note that python SimpleHTTPServer can't be used here because it returns incorrect mimetype for .wasm files, read here if you insist using python SimpleHTTPServer.