openface-tornado

Source code for openface-tornado docker image: https://hub.docker.com/r/evxu/openface-tornado/

Description

A face detection API based on OpenFace and tornado web server. It can help you detect faces, extract face features, draw bounding boxes on the detected faces, and compare faces between two images.

Installation

$ docker pull evxu/openface-tornado:latest

Run the server

Run the docker container by:

$ docker run -i -t -p 9000:8000 evxu/openface-tornado --name pinkring /bin/bash
$ python ./oft/demo/faceserver.py

Or run the default demo in background:

$ docker run -p 9000:8000 -d --name pinkring evxu/openface-tornado

Now, you have built a container named ‘pinkring’ within our API running.
To mount your own file, use -v option.

Let’s test your installation on your host:

1. Test the face matching server:
In command:

curl 'http://localhost:9000/matchface/?url1=http://images.idspicturedesk.com/images/161/lowres/abexs/abexs411.jpg&url2=http://images.idspicturedesk.com/images/161/lowres/abexs/abexs411.jpg'

Or visit the hTTP address in your browser.
It should return you {"distance": 0.0} in JSON data format, because url1 and url2 are the same image.

2. Test the face detection server with url:
In command:

curl 'http://localhost:9000/detectface/?url=http://images.idspicturedesk.com/images/161/lowres/abexs/abexs411.jpg'

Or visit the hTTP address in your browser.
It should return you the data in JSON format starting with {"shape": [350, 600, 3], "results": [{"box": [380, 109, 130, 130], "feature": [-0.0076855164952576, ...
response.shape returns the the shape of the image.
response.results gives you the information of locations and face features of detected face,
where response.results[0].box represents the bounding box of first detected face in the format of [left, top, width, height], and response.results[0].feature is the feature in format of list(128,) extracted from the first detected face.

3. Test the face detection server with image file:
in command:

curl -F image=@xxxx.jpg http://localhost:9000/detect/

It will return data in the same format as that when you post image in URL.

4. Show the detected face in bounding boxes

Open your browser and visit:

http://localhost:9000/boxface/?url=http://images.idspicturedesk.com/images/161/lowres/abexs/abexs411.jpg

It will display the image with bounding boxes on detected faces.

Usage

To request face matching server, request the URL on the host:
curl 'http://localhost:9000/matchface/?url1=[url of first image]&url2=[url of second image]'

To request face detection server:
post image with URL:
curl 'http://localhost:9000/detectface/?url=[url of image]'
post image with a image file:
curl -F image=@[image path] http://localhost:9000/detect/'

To draw bounding boxes on faces:
The URL is http://localhost:9000/?boxface/?url=[url of image]
If you are running this API on a remote server. Map the port of your local machine to the remote server by ssh -L 90009000 [address of remote server] . Then you should be able to check the results with you local browser.

in Python

To request face detection server:

import request
# if use url
url = 'your image url'
resp = requests.get('http://localhost:9000/detectface/ulr='+url) 
# if use file
imgpath='path of your image file'
resp = requests.post('http://localhost:9000/detectface/', files = {'image': open(imgpath, 'r')})
features = []
boxes = []
rst = resp.json()['results']
if rst: # if there are faces detected
      features += [b['feature']]
      boxes += [b['box']]