A Deep Learning Workflow enhanced with Optical Flow Fields for Flood Risk Estimation

Citation request

Ranieri, C.M., Souza, T.L.D., Nishijima, M., Krishnamachari, B. and Ueyama, J., 2024. A deep learning workflow enhanced with optical flow fields for flood risk estimation. Applied Intelligence. DOI: https://doi.org/10.1007/s10489-024-05466-2

Data

The dataset used in this research is available at: https://www.kaggle.com/dsv/8350235

Docker

Using Docker images is a way of improving reproducibility of a project, as well as automating install, setup, and running of the code. Hence, we provided a Dockerfile and supporting scripts for building and running the container. By following these instructions, you will have a GPU-enabled container running, with all the dependencies for this project.

If you don't have Docker installed yet:

1. Install Docker to your machine. Follow the instructions here. We recommend performing the post-install steps, so that you don't need to run the containers as root (i.e., sudo).
2. Install the NVIDIA support for Docker. Follow the instructions here.

After you have Docker installed with NVIDIA support, proceed with the following.

To build the Docker container, run:

cd docker
bash build-docker.sh

To run a Docker container, change the variables in run-docker.sh to those in your local machine. For example, if you stored the data in the /home/data diretory, set $data_dir='/home/data'. The other variables, $checks_dir and $logs_dir, refer to the checkpoints and logs generated while training the deep learning models. Please, set them as your convenience. After setting up the paths, run:

bash run-docker.sh

To launch jupyter notebook from within the Docker container, run:

jupyter notebook --port=8888 --no-browser --ip=0.0.0.0 --allow-root