datasets/DrivAerML/README.md

Downloading dataset

Run download_dataset.sh after setting LOCAL_DIR (path to where files should be downloaded)
Currently only stl files (drivaer_i.stl) and surface points and fields files (boundary_i.vtp) are being downloaded.

Preprocess dataset

Run preprocess_data.py after setting the paths of downloaded data (input_dir) and where processed data should be saved (output_dir) in data_preprocessing.yaml. This will create .npy files in output_dir.
Currently only surface quantities are being stored in the npy files. Change model_type in data_preprocessing.yaml to 'volume' or 'combined' for other quantities to be stored.
The dataset does not provide train/val/test splits. splits_creation.py can be used for creating and saving the splits in train_val_test_splits.json.

Dataset and Dataloader

dataset_drivaerml.py has the Dataset class. Use get_dataloaders function for the Dataloaders.
For first run use config argument 'presampled_exists': False.
Set 'num_points' argument in config file and those many points will be sampled from each sample and stored in path set in 'presampled_data_path' argument in config.

Refer to \url{https://huggingface.co/datasets/neashton/drivaerml} and \url{https://github.com/NVIDIA/physicsnemo/tree/main/examples/cfd/external_aerodynamics/domino} for further details.