REST API for Curious AI

Project involves to create a RESTFUL API with use case:

• Upload pictures and store them
• Upload pictures and predict from a model

This project uses PyTorch densenet161 as example. Does the image proccesing for that specific pre-trained model. Model could be change on enviroment and the width and height of image too.

Endpoints

As it had been said, this project has 2 endpoints. Uploading an image

URL: localhost:5000/api/upload
Content-type: form-data
Body:
    > image: image with extension .jpeg, .jpg or .png
    > token: secret of a user for saving method

Prediction on image

URL: localhost:5000/api/predict
Content-type: form-data
Body:
    > image: image with extension .jpeg, .jpg or .png
    > token: secret of a user for saving method

Installation

Project uses PyTorch and other packages, to install run the followings commands

pip install -r requirements.txt
pip install https://download.pytorch.org/whl/cu100/torch-1.1.0-cp37-cp37m-linux_x86_64.whl
pip install https://download.pytorch.org/whl/cu100/torchvision-0.3.0-cp37-cp37m-linux_x86_64.whl

Note that could be other versions as pip is currently not working for PyTorch.

Enviroment

Project does use an .env with the following variables

UPLOAD_DIRECTORY='path/to/directory/' default='app/persistent/'
DEV_MODE=True default
PORT=5000 default
MODEL_PATH='path/to/model' default='app/models'
WIDTH=224
HEIGHT=224

Running

Running on root folder of the project and set the FLASK APP as following Windows

set FLASK_APP=app/app.py

Unix

export FLASK_APP=app/app.py

then run

flask run

Tests

Running on root folder of the project

pytest

Docker

Project has a dockefile running a not standart PyTorch image. To build the image

docker build -t curious_api .

To run the container

docker run -p 5000:5000 curious_api

License

MIT