Two-Stage Super-Resolution Simulation Method for Three-Dimensional Flow Fields Around Buildings for Real-Time Prediction of Urban Micrometeorology
- The experiments have been conducted in the Singularity container.
- The same experimental environment can be made using Docker.
- Check if the command
singularity
works. If necessary, install Singularity. - Build a container:
$ singularity build -f pytorch.sif ./singularity/pytorch.def
- Change preferences in
./script/start_singularity_container.sh
if needed. - Run a container:
$ ./script/start_singularity_container.sh
- Check if the command
docker compose
works. If necessary, install Docker. - Change preferences in
docker-compose.yml
if needed. - Build a container:
$ docker compose build
- Run a container:
$ docker compose up -d
- This repository contains only the deep-learning code, not the data.
- If the data were in an appropriate directory, the following scripts would work.
- Training:
$ ./script/train_unet_lr.sh
- Two NVIDIA A100 GPU cards are assumed.
- Inference:
$ ./script/make_lr_inference.sh
- A single NVIDIA A100 GPU cards are assumed.
- Evaluation
- Start JupyterLab:
$ ./script/start_singularity_container.sh
- Run a notebook
python/notebooks/evaluate_lr_models.ipynb
- Start JupyterLab:
- Training:
$ ./script/train_unet_hr.sh
- Two NVIDIA A100 GPU cards are assumed.
- Inference:
$ ./script/make_hr_inference.sh
- A single NVIDIA A100 GPU cards are assumed.
- Evaluation
- Start JupyterLab:
$ ./script/start_singularity_container.sh
- Run a notebook
python/notebooks/evaluate_hr_models.ipynb
- Start JupyterLab: