In the context of the Climate Informatics 2020 conference data challenge, we attempt to predict visible imagery at night using thermal infra-red observation. More details on the challenge and datasets in here. In this work, we propose to train a pix2pix-like cGANs at spectral domain translation.
From command line
Setup a configuration file specifying experiment hyperparameters, templates provided here.
Then run
$ python run_training.py --cfg=path_to_config_file --o=path_to_logs_directory --device=gpu_id
To reproduce the experiment submitted to the hackathon, one should use configurations specfied by config/base_unet.yaml
From a notebook
See demonstration notebook on how to run an experiment.
Results
Example of predicted visible images from nightly infrared inputs
Code implemented in Python 3.8
Clone and go to repository
$ git clone https://github.com/shahineb/ci-hackathon.git
$ cd ci-hackathon
Create and activate environment
$ pyenv virtualenv 3.8.2 hackathon
$ pyenv activate hackathon
$ (hackathon)
Install dependencies
$ (hackathon) pip install -r requirements.txt