This little project enables using the PRNet library in a WSGI-compatible web server and provides a docker image with one such fully functional web-server.

# Clone this repository
git clone https://github.com/veloman-yunkan/prnet_http_server.git

cd prnet_http_server/

# This script clones the PRNet repo and gets the trained model
# data from Google Drive
./setup_PRNet 

# Install required dependencies - see Dockerfile.base

# Run the tests (1 of them may fail because of slight numeric differences)
./run_tests

./build_docker_image

This step takes a while and creates a 1.5GB sized docker image prnet_http_server:latest.

If anything changes in this repo or PRNet, then - after updating them - just run

./update_docker_image

Updating the docker image is much faster - the new image is built on top of the prnet_http_server:base image (containing just the dependencies) that was created by the ./build_docker_image script.

./run_tests --use-docker

Start the server with

docker run --rm -d -p 8000:8000 prnet_http_server:latest

Send an image to it with

curl -OJF image=@test_data/single_face.jpg http://localhost:8000

When starting the server, you can specify the upper limit on the image file size (in bytes) through the PRNET_MAX_IMAGE_SIZE environment variable:

# Make the server reject images exceeding 5MB
docker run --rm -d -p 8000:8000 -e PRNET_MAX_IMAGE_SIZE=5000000 prnet_http_server:latest

The container runs the gunicorn web server inside and allows to specify its settings through the command line. For example, to make the web-server use 4 worker processes for handling requests, provide the --workers=4 option:

docker run --rm -d -p 8000:8000 prnet_http_server:latest --workers=4
#                                                        ^^^^^^^^^^^