Installation¶

First obtain the code from github.

git clone https://github.com/bw0248/SimpleSR
cd SimpleSR

Running natively¶

If you want to run natively on your own machine set up a Python virtual environment and install the requirements. You can either do this manually by invoking:

python3.6 -m venv .env
source .env/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

Or you can use the supplied make file and things will be initialized for you.

python3.6 -m venv .env
make init

Running inside docker container¶

Another option is to run inside a docker container. The docker image can be build with the supplied Dockerfile.

Note

The Dockerfile is based on a nvidia-cudnn image which is not particular lightweight, but will be needed for training on GPUs. If you don’t plan on using a GPU you can alter the Dockerfile and inherit from another base image (Ubuntu for instance). Training on a GPU is definitely recommended though.

Make sure you have docker installed and running.

To build the image and start the container (depending on your docker installation you may need to run the following commands as root):

# build image
docker build -t simple_sr .

# make sure the image was created successfully
docker images     # you should see an image called simple_sr

# obtain image id (can also be done manually, by checking third column of 'docker images')
img_id=$(docker images | grep simple_sr | awk '{print $3}')

# start and enter container (container will stop automatically once you exit it)
docker run -p 6006:6006 -it $img_id /bin/bash

# you should now be inside the container
# check that everything worked and requirements are installed
cd dev/
source .env/bin/activate
pip list

To check that tensorboard (and everything else) is working correctly you can start the minimal training example.

Note

This will train a model on a tiny dataset containing only 8 images, so results will naturally be very bad. This is just to see if everything works.

# assuming you're still inside the docker container...
# start tensorboard server with supplied Makefile (or otherwise if you like)
make tensorboard &

# start minimal training example
make training_example TRAINING_CONFIG=minimal_example.yaml

If you now navigate to http://localhost:6006 in your browser you should see tensorboard and after a short while stats from your running training session.