Learning to See in the Dark Flask App

This repo is a Flask App built upon the PyTorch Learning to See In the Dark Repository.

Setup

1. Clone the repository including the submodules
   git clone --recurisve https://github.com/SEANDOUGHTY/learning-to-see-in-the-dark-flask.git

2. Add the pretrained model into /checkpoint/checkpoint.t7

3. Create an AWS bucket for storing input and output images and assign environment variables:

   • AWS_ACCESS_KEY_ID
   • AWS_SECRET_ACCESS_KEY
   • AWS_REGION

4. Build the docker containers
   docker-compose build

5. Start the docker container
   docker-compose up

API Calls

After the docker containers are successfully running inference can be performed by using a POST request with the following format:

{
    "Bucket": $BUCKETNAME,
    "input-image": $INPUTIMAGE_KEY,
    "output-image": $OUTPUTIMAGE_KEY",
    "ratio": $RATIO
}

Ratio is a scaling factor between the original image exposure and the output image exposure.