Wrap face analysis algorithms in Docker

.env

@@ -1,5 +1,35 @@
+# these values configure the build of the docker images
 BUILD_TAG=latest
+DEVTOOLS=false
+
+# the port on which the api will be available
 APP_PORT=80
+
+# paths to the directories where data will be persisted on disk
 DATA_DIR=./persisted_data/prod/images
 DB_DIR=./persisted_data/prod/database
-DEVTOOLS=false
+
+# queue configuration
+IMAGE_PROCESSOR_QUEUE=faceanalysis
+
+# configuration values for mysql
+MYSQL_USER=faceanalysisrw
+MYSQL_PASSWORD=some-secure-string
+MYSQL_ROOT_PASSWORD=some-very-secure-string
+MYSQL_DATABASE=faceanalysis
+
+# allowed values are DEBUG, INFO, WARNING, ERROR and CRITICAL
+LOGGING_LEVEL=DEBUG
+
+# separate multiple extensions with underscores
+ALLOWED_IMAGE_FILE_EXTENSIONS=JPG_PNG
+
+# api access token configuration
+TOKEN_SECRET_KEY=some-long-random-string
+DEFAULT_TOKEN_EXPIRATION_SECS=500
+
+# maximum distance between two face vectors for them to be considered the same person
+DISTANCE_SCORE_THRESHOLD=0.6
+
+# docker image name of the algorithm to use for face vectorization
+FACE_VECTORIZE_ALGORITHM=cwolff/face_recognition

.travis.yml

@@ -1,7 +1,11 @@
 language: generic
 
 services:
-  - docker
+- docker
+
+before_install:
+- openssl aes-256-cbc -K $encrypted_e08e3836e1cc_key -iv $encrypted_e08e3836e1cc_iv
+  -in ./.travis/secrets.env.enc -out ./secrets.env -d
 
 script:
-  - ./run-test.sh
+- ./run-test.sh

README.md

@@ -4,7 +4,7 @@
 1. Create an Azure VM (preferably Ubuntu 16.04)
 2. Install Docker and Docker Compose
 3. Clone this repo
-4. Replace default environment variables in environment_variables.yml
+4. Replace default configuration values in `.env` and `secrets.env`
 5. To run tests type './run-test.sh' from within the top level directory
 6. To run in production type './run-prod.sh' from within the top level directory
 7. If you would like to clear the production database, run './delete-prod-data.sh'

algorithms/face_recognition/Dockerfile

@@ -0,0 +1,41 @@
+FROM python:3.6-slim-stretch
+
+RUN apt-get -y update && \
+    apt-get install -y --fix-missing \
+        build-essential \
+        cmake \
+        gfortran \
+        git \
+        wget \
+        curl \
+        graphicsmagick \
+        libgraphicsmagick1-dev \
+        libatlas-dev \
+        libavcodec-dev \
+        libavformat-dev \
+        libgtk2.0-dev \
+        libjpeg-dev \
+        liblapack-dev \
+        libswscale-dev \
+        pkg-config \
+        python3-dev \
+        python3-numpy \
+        software-properties-common \
+        zip && \
+    apt-get clean && \
+    rm -rf /tmp/* /var/tmp/*
+
+RUN cd ~ && \
+    mkdir -p dlib && \
+    git clone -b 'v19.9' --single-branch https://github.com/davisking/dlib.git dlib/ && \
+    cd dlib/ && \
+    python3 setup.py install --yes USE_AVX_INSTRUCTIONS
+
+WORKDIR /app
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY vectorize.py .
+
+ENTRYPOINT ["python3", "vectorize.py"]

algorithms/face_recognition/requirements.txt

@@ -0,0 +1 @@
+face-recognition==1.0.0

algorithms/face_recognition/vectorize.py

@@ -0,0 +1,35 @@
+import json
+import face_recognition as fr
+
+
+def get_face_vectors(img_path):
+    img = fr.load_image_file(img_path)
+    face_locations = fr.face_locations(img)
+    face_vectors = []
+    for top, right, bottom, left in face_locations:
+        cropped_img = img[top:bottom, left:right]
+        cropped_features = fr.face_encodings(cropped_img)
+        if cropped_features:
+            face_vector = cropped_features[0]
+            face_vectors.append(face_vector.tolist())
+    return face_vectors
+
+
+def _cli():
+    from argparse import ArgumentParser
+    from argparse import FileType
+
+    parser = ArgumentParser(description=__doc__)
+    parser.add_argument('image', type=FileType('r'))
+
+    args = parser.parse_args()
+    image = args.image
+    image.close()
+    img_path = image.name
+
+    vectors = get_face_vectors(img_path)
+    print(json.dumps({'faceVectors': vectors}))
+
+
+if __name__ == '__main__':
+    _cli()

app/Dockerfile

@@ -3,40 +3,9 @@ FROM python:3.5-slim
 RUN apt-get -y update \
     && apt-get install -y \
         mysql-client \
-    && apt-get clean \
-    && rm -rf /tmp/* /var/tmp/*
-
-RUN apt-get -y update \
-    && apt-get install -y --fix-missing \
-        build-essential \
-        cmake \
-        gfortran \
-        git \
-        wget \
-        curl \
-        graphicsmagick \
-        libgraphicsmagick1-dev \
-        libatlas-dev \
-        libavcodec-dev \
-        libavformat-dev \
-        libboost-all-dev \
-        libgtk2.0-dev \
-        libjpeg-dev \
-        liblapack-dev \
-        libswscale-dev \
-        pkg-config \
-        python3-dev \
         python3-numpy \
-        software-properties-common \
-        zip \
-        mysql-client \
     && apt-get clean \
-    && rm -rf /tmp/* /var/tmp/* \
-    && cd ~ \
-    && mkdir -p dlib \
-    && git clone -b 'v19.7' --single-branch https://github.com/davisking/dlib.git dlib/ \
-    && cd dlib/ \
-    && python3 setup.py install --yes USE_AVX_INSTRUCTIONS
+    && rm -rf /tmp/* /var/tmp/*
 
 WORKDIR /app
 

app/faceanalysis/api.py

@@ -4,14 +4,15 @@
 from http import HTTPStatus
 import werkzeug
 from werkzeug.utils import secure_filename
-from azure.storage.queue import QueueService
 from flask_restful import Resource, Api, reqparse
 from flask import Flask, g
 from .models.models import Match, Image, User, ImageStatus
-from .models.database_manager import DatabaseManager
+from .models.database_manager import get_database_manager
 from .models.image_status_enum import ImageStatusEnum
 from .log import get_logger
+from .queue_poll import create_queue_service
 from .auth import auth
+from .settings import IMAGE_PROCESSOR_QUEUE, ALLOWED_EXTENSIONS
 
 app = Flask(__name__)
 app.config['UPLOAD_FOLDER'] = os.path.join(
@@ -20,10 +21,8 @@
     'images')
 app.url_map.strict_slashes = False
 api = Api(app)
-queue_service = QueueService(account_name=os.environ['STORAGE_ACCOUNT_NAME'],
-                             account_key=os.environ['STORAGE_ACCOUNT_KEY'])
-queue_service.create_queue(os.environ['IMAGE_PROCESSOR_QUEUE'])
-logger = get_logger(__name__, os.environ['LOGGING_LEVEL'])
+queue_service = create_queue_service(IMAGE_PROCESSOR_QUEUE)
+logger = get_logger(__name__)
 
 
 class AuthenticationToken(Resource):
@@ -46,7 +45,7 @@ def post(self):
         args = parser.parse_args()
         username = args['username']
         password = args['password']
-        db = DatabaseManager()
+        db = get_database_manager()
         session = db.get_session()
         query = session.query(User).filter(User.username == username).first()
         session.close()
@@ -71,7 +70,7 @@ def post(self):
                             help="img_id missing in the post body")
         args = parser.parse_args()
         img_id = args['img_id']
-        db = DatabaseManager()
+        db = get_database_manager()
         session = db.get_session()
         img_status = session.query(ImageStatus).filter(
             ImageStatus.img_id == img_id).first()
@@ -81,8 +80,7 @@ def post(self):
                 return ('Image previously placed on queue',
                         HTTPStatus.BAD_REQUEST.value)
             try:
-                queue_service.put_message(os.environ['IMAGE_PROCESSOR_QUEUE'],
-                                          img_id)
+                queue_service.put_message(IMAGE_PROCESSOR_QUEUE, img_id)
                 img_status.status = ImageStatusEnum.on_queue.name
                 db.safe_commit(session)
                 logger.info('img successfully put on queue')
@@ -99,7 +97,8 @@ def post(self):
 
     def get(self, img_id):
         logger.debug('checking if img has been processed')
-        session = DatabaseManager().get_session()
+        db = get_database_manager()
+        session = db.get_session()
         img_status = session.query(ImageStatus).filter(
             ImageStatus.img_id == img_id).first()
         session.close()
@@ -111,8 +110,6 @@ def get(self, img_id):
 
 class ImgUpload(Resource):
     method_decorators = [auth.login_required]
-    env_extensions = os.environ['ALLOWED_IMAGE_FILE_EXTENSIONS']
-    allowed_extensions = env_extensions.lower().split('_')
 
     # pylint: disable=broad-except
     def post(self):
@@ -125,7 +122,7 @@ def post(self):
                             location='files')
         args = parser.parse_args()
         img = args['image']
-        db = DatabaseManager()
+        db = get_database_manager()
         if self._allowed_file(img.filename):
             filename = secure_filename(img.filename)
             img_id = filename[:filename.find('.')]
@@ -151,12 +148,12 @@ def post(self):
         else:
             error_msg = ('Image upload failed: please use one of the '
                          'following extensions --> {}'
-                         .format(self.allowed_extensions))
+                         .format(ALLOWED_EXTENSIONS))
             return error_msg, HTTPStatus.BAD_REQUEST.value
 
     def _allowed_file(self, filename):
         return ('.' in filename and
-                filename.rsplit('.', 1)[1].lower() in self.allowed_extensions)
+                filename.rsplit('.', 1)[1].lower() in ALLOWED_EXTENSIONS)
 
 
 class ImgMatchList(Resource):
@@ -165,7 +162,8 @@ class ImgMatchList(Resource):
     # pylint: disable=assignment-from-no-return
     def get(self, img_id):
         logger.debug('getting img match list')
-        session = DatabaseManager().get_session()
+        db = get_database_manager()
+        session = db.get_session()
         query = session.query(Match).filter(Match.this_img_id == img_id)
         imgs = []
         distances = []
@@ -182,7 +180,8 @@ class ImgList(Resource):
 
     def get(self):
         logger.debug('getting img list')
-        session = DatabaseManager().get_session()
+        db = get_database_manager()
+        session = db.get_session()
         query = session.query(Image).all()
         imgs = [f.img_id for f in query]
         session.close()

app/faceanalysis/auth.py

@@ -1,15 +1,16 @@
 from flask_httpauth import HTTPBasicAuth
 from flask import g
 from .models.models import User
-from .models.models import DatabaseManager
+from .models.database_manager import get_database_manager
 
 auth = HTTPBasicAuth()
 
 
 @auth.verify_password
 def verify_password(username_or_token, password):
     user = User.verify_auth_token(username_or_token)
-    session = DatabaseManager().get_session()
+    db = get_database_manager()
+    session = db.get_session()
     if not user:
         user = session.query(User).filter(
             User.username == username_or_token).first()

app/faceanalysis/face_vectorizer.py

@@ -0,0 +1,39 @@
+import os
+import json
+import docker
+from .log import get_logger
+from .settings import MOUNTED_DATA_DIR
+from .settings import HOST_DATA_DIR
+
+logger = get_logger(__name__)
+
+
+def _format_mount_path(img_path):
+    return '/{}'.format(os.path.basename(img_path))
+
+
+def _format_host_path(img_path):
+    # volume mounts must be absolute
+    if not img_path.startswith('/'):
+        img_path = os.path.abspath(img_path)
+
+    # adjust the path if it itself is a mount and if we're spawning a
+    # sibling container
+    if MOUNTED_DATA_DIR and HOST_DATA_DIR:
+        img_path = img_path.replace(MOUNTED_DATA_DIR, HOST_DATA_DIR)
+
+    return img_path
+
+
+def get_face_vectors(img_path, algorithm):
+    img_mount = _format_mount_path(img_path)
+    img_host = _format_host_path(img_path)
+    volumes = {img_host: {'bind': img_mount, 'mode': 'ro'}}
+
+    logger.debug('Running container %s with image %s', algorithm, img_host)
+    client = docker.from_env()
+    stdout = client.containers.run(algorithm, img_mount,
+                                   volumes=volumes, auto_remove=True)
+
+    face_vectors = json.loads(stdout.decode('ascii')).get('faceVectors', [])
+    return face_vectors

app/faceanalysis/log.py

@@ -1,8 +1,9 @@
 import sys
 import logging
+from .settings import LOGGING_LEVEL
 
 
-def get_logger(module_name, logging_level):
+def get_logger(module_name, logging_level=LOGGING_LEVEL):
     logging_levels = {
         'DEBUG': logging.DEBUG,
         'INFO': logging.INFO,