Improve DrivAerML dataset card

README.md

@@ -1,82 +1,84 @@
 ---
-viewer: false
+pretty_name: DrivAerML
 license: cc-by-sa-4.0
+viewer: false
+tags:
+- computational-fluid-dynamics
+- scientific-machine-learning
+- automotive-aerodynamics
+- high-fidelity-simulation
+- surrogate-modeling
+- ai-for-science
+- open-data
 ---
 
-DrivAerML: High-Fidelity Computational Fluid Dynamics Dataset for Road-Car External Aerodynamics
--------
+# DrivAerML: High-Fidelity Computational Fluid Dynamics Dataset for Road-Car External Aerodynamics
+
+[Project website](https://caemldatasets.org) · [Paper](https://arxiv.org/abs/2408.11969) · [Hugging Face paper page](https://huggingface.co/papers/2408.11969) · [Official splits](splits/)
+
+## At a Glance
+
+| | |
+|---|---|
+| Domain | External aerodynamics of realistic road-car configurations |
+| Scale | 500 parametrically morphed designs; 484 runs are publicly available |
+| CFD method | Scale-resolving CFD using OpenFOAM v2212 with UpstreamCFD modifications |
+| Geometry | DrivAer notchback generic vehicle |
+| Published data | Surface and volume fields, geometry, meshes, slices, images, and force/moment coefficients |
+| License | CC BY-SA 4.0 |
 
-Contact: 
-----------
-Neil Ashton (contact@caemldatasets.org)
+## Summary
 
-Website:
-----------
-https://caemldatasets.org
+Machine learning has the potential to enable rapid aerodynamic predictions early in the vehicle design process, but progress is limited by the availability of open training data for realistic road cars generated with high-fidelity CFD methods.
 
-Summary:
--------
+DrivAerML addresses this gap with a high-fidelity open dataset based on 500 parametrically morphed variants of the widely used DrivAer notchback generic vehicle. Mesh generation and scale-resolving CFD were executed using consistent, validated, automated workflows representative of industrial practice. Geometry and rich aerodynamic data are published in open formats.
 
-Machine Learning (ML) has the potential to revolutionise the field of automotive aerodynamics, enabling split-second flow predictions early in the design process. 
-However, the lack of open-source training data for realistic road cars, using high-fidelity CFD methods, represents a barrier to their development.
-To address this, a high-fidelity open-source (CC-BY-SA) public dataset for automotive aerodynamics has been generated, based on 500 parametrically morphed variants of the widely-used DrivAer notchback generic vehicle. Mesh generation and scale-resolving CFD was executed using consistent and validated automatic workflows representative of the industrial state-of-the-art. Geometries and rich aerodynamic data are published in open-source formats. To our knowledge, this is the first large, public-domain dataset for complex automotive configurations generated using high-fidelity CFD. 
+## CFD Solver
 
-CFD Solver:
-----------
-All cases were run using the open-source finite-volume code OpenFOAM v2212 with custom modifications by UpstreamCFD. Please see the paper below for full details on the code and validation:
+All cases were run with the open-source finite-volume code OpenFOAM v2212 with custom modifications by UpstreamCFD. See the [paper](https://arxiv.org/abs/2408.11969) for full solver and validation details.
 
-How to cite this dataset:
-----------------
-In order to cite the use of this dataset please cite the paper below which contains full details on the dataset.
+## How to Cite This Dataset
 
-''
+Please cite the corresponding paper:
+
+```bibtex
 @article{ashton2024drivaer,
-    title = {{DrivAerML: High-Fidelity Computational Fluid Dynamics Dataset for Road-Car External Aerodynamics}},
-    year = {2024},
-journal = {arxiv.org},
-    url={https://arxiv.org/abs/2408.11969},
-    author = {Ashton, N., Mockett, C., Fuchs, M., Fliessbach, L., Hetmann, H., Knacke, T., Schonwald, N.,
-Skaperdas, V., Fotiadis, G., Walle, A., Hupertz, B., and Maddix, D}
+  title   = {{DrivAerML: High-Fidelity Computational Fluid Dynamics Dataset for Road-Car External Aerodynamics}},
+  author  = {Ashton, Neil and Mockett, Charles and Fuchs, Marian and Fliessbach, Louis and Hetmann, Hendrik and Knacke, Thilo and Schonwald, Norbert and Skaperdas, Vangelis and Fotiadis, Grigoris and Walle, Astrid and Hupertz, Burkhard and Maddix, Danielle},
+  journal = {arXiv preprint arXiv:2408.11969},
+  year    = {2024},
+  url     = {https://arxiv.org/abs/2408.11969}
 }
-''
-
-Files:
--------
+```
 
-### Native boundary polygon areas
+## Dataset Structure and Files
 
-For each published `run_N`, `run_N/boundary_cell_area_N.npy` is a one-dimensional little-endian float32 (`<f4`) array with one area in `m^2` per native polygon (and matching `CellData` tuple) of `run_N/boundary_N.vtp`, in its original cell order. It is not valid for an STL, PhysicsNeMo PDMsh, retriangulated mesh, or reordered cells. The method and per-case metadata are in [`surface_cell_areas/`](surface_cell_areas/).
+Each `run_i` folder corresponds to a different geometry and contains the following files, where `i` is the run number:
 
+* `drivaer_i.stl`: Geometry STL (approximately 135 MB).
+* `geo_ref_i.csv`: Reference values for the geometry.
+* `geo_parameters_i.csv`: Parameters defining the geometry.
+* `boundary_i.vtp`: Surface boundary data (approximately 500 MB).
+* `boundary_cell_area_i.npy`: Native polygon areas in `m^2`, in the exact `CellData` tuple order of `boundary_i.vtp`.
+* `volume_i.vtu`: Volume field data (approximately 50 GB). On Hugging Face, this is stored in two parts that must be concatenated to recreate `volume_i.vtu`.
+* `force_mom_i.csv`: Time-averaged force and moment coefficients using geometry-dependent frontal area and wheelbase references.
+* `force_mom_constref_i.csv`: Time-averaged force and moment coefficients using constant frontal area and wheelbase references.
+* `slices/`: VTP slices in the x, y, and z directions containing flow-field variables.
+* `images/`: Flow-variable images—including `Cp`, `CpT`, and `UMagNorm`—on slices and the vehicle surface, plus force-coefficient time-averaging and residual-convergence plots.
 
-Each folder (e.g run1,run2...run"i" etc) corresponds to a different geometry that contains the following files where "i" is the run number: 
-* geometry stl (~135mb): drivaer_i.stl 
-* reference values for each geometry: geo_ref_i.csv 
-* reference geometry for each geometry: geo_parameters_i.csv 
-* Boundary VTU (~500mb): boundary_i.vtp 
-* Native boundary polygon cell areas (`m^2`, matching `boundary_i.vtp` `CellData` order): `boundary_cell_area_i.npy`
-* Volume field VTU (~50GB): volume_i.vtu ( please note on HuggingFace this is split into part 1 and part2 - please cat them together to create the volume_i.vtu)
-* forces/moments time-averaged (using varying frontal area/wheelbase): force_mom_i.csv 
-* forces/moments time-averaged (using constant frontal area/wheelbase): force_mom_constref_i.csv 
-* slices: folder containing .vtp slices in x,y,z that contain flow-field variables 
-* Images: This folder contains images of various flow variables (e.g. Cp, CpT, UMagNorm) for slices of the domain at X, Y, and Z locations (M signifies minus, P signifies positive), as well as on the surface. It also includes evaluation plots of the time-averaging of the force coefficients (via the tool MeanCalc) and a residual plot illustrating the convergence.
+The repository root also contains:
 
-In addition to the files per run folder, there are also: 
-* openfoam_meshes : this folder contains the OpenFOAM meshes (in OpenFOAM format) used for these simulations. The 0 and system folders are just the default output from ANSA and were not those used in this study. Please refer to the arxiv paper for full details of the CFD setup. We hope that by providing the meshes, groups may wish to expand the dataset as they see fit.  
-* force_mom_all.csv : forces/moments time-averaged (using varying frontal area/wheelbase) for all runs 
-* force_mom_constref_all.csv : forces/moments time-averaged (using constant frontal area/wheelbase) for all runs 
-* geo_parameters_all.csv: reference geometry values for each geometry for all runs 
-* [`splits/`](splits/): deterministic benchmark manifests, documentation, source metrics, diagnostic figures, and generation code.
+* `openfoam_meshes/`: OpenFOAM meshes used for the simulations. The included `0/` and `system/` directories are default ANSA output and were not used in the study; consult the paper for the CFD setup.
+* `force_mom_all.csv`: Force and moment coefficients for all runs using geometry-dependent reference values.
+* `force_mom_constref_all.csv`: Force and moment coefficients for all runs using constant reference values.
+* `geo_parameters_all.csv`: Geometry parameters for all runs in one file.
+* [`splits/`](splits/): Deterministic benchmark manifests, documentation, source metrics, diagnostic figures, and generation code.
 
-## Recommended dataset splits
+## Recommended Dataset Splits
 
-For reproducible machine-learning evaluation, DrivAerML provides eight
-deterministic split families in [`splits/manifest.json`](splits/manifest.json).
-Case identifiers match the top-level `run_N` directories.
+For reproducible machine-learning evaluation, DrivAerML provides eight deterministic split families in [`splits/manifest.json`](splits/manifest.json). Case identifiers match the top-level `run_N` directories.
 
-The split construction is based on the 484 publicly available runs. The 16
-unavailable or held-back runs are excluded from every partition. Reduced-data
-variants intentionally use subsets of the standard training population while
-retaining fixed validation and test sets.
+The split construction is based on the 484 publicly available runs. The 16 unavailable or held-back runs are excluded from every partition. Reduced-data variants intentionally use subsets of the standard training population while retaining fixed validation and test sets.
 
 | Split | Type | Train | Validation | Test | Intended use |
 |---|---:|---:|---:|---:|---|
@@ -93,102 +95,95 @@ The data-efficiency training sets are nested:
 
 `super_scarce_train ⊂ scarce_train ⊂ medium_train ⊂ full_train`
 
-They share the same validation and test sets, allowing direct comparisons
-across training-set sizes. For the OOD splits, validation cases are sampled
-from the training-side population; the held-out extreme is reserved for final
-testing.
+They share the same validation and test sets, allowing direct comparisons across training-set sizes. For the OOD splits, validation cases are sampled from the training-side population; the held-out extreme is reserved for final testing.
 
-Use `full` for a standard baseline, the nested sequence for data-efficiency
-studies, `geometry` for surface-shape extrapolation, `high_drag` or
-`low_drag` for coefficient-regime extrapolation, and `rear_separation` for
-flow-structure generalization.
+Use `full` for a standard baseline, the nested sequence for data-efficiency studies, `geometry` for surface-shape extrapolation, `high_drag` or `low_drag` for coefficient-regime extrapolation, and `rear_separation` for flow-structure generalization.
 
-Download only the split package with:
+Complete definitions, construction methods, diagnostic figures, source metrics, and reproducibility instructions are provided in [`splits/README.md`](splits/README.md).
 
-```bash
-hf download neashton/drivaerml \
-  --repo-type dataset \
-  --include "splits/**" \
-  --local-dir ./drivaerml
-```
+## Native Boundary Polygon Areas
 
-Complete definitions, construction methods, diagnostic figures, source
-metrics, and reproducibility instructions are provided in
-[`splits/README.md`](splits/README.md).
+For each published `run_N`, `run_N/boundary_cell_area_N.npy` is a one-dimensional little-endian float32 (`<f4`) array with one area in `m^2` per native polygon—and matching `CellData` tuple—of `run_N/boundary_N.vtp`, in its original cell order. It is not valid for an STL, PhysicsNeMo PDMsh, retriangulated mesh, or reordered cells. The method and per-case metadata are in [`surface_cell_areas/`](surface_cell_areas/).
 
-How to download:
-----------------
+## How to Download
 
-The dataset is now available on HuggingFace. Below are some examples of how to download all or selected parts of the dataset. Please refer to the HuggingFace documentation for other ways to accessing the dataset and building workflows.
+DrivAerML is publicly available on Hugging Face and is approximately 31 TB in full. We recommend the Hugging Face CLI for selective, resumable downloads and a dry-run preview before transferring large files.
 
-Example 1: Download all files (~31TB)
-----------
-Please note you’ll need to have git lfs installed first, then you can run the following command:
+### Prerequisites
 
-git clone git@hf.co:datasets/neashton/drivaerml
-
-Example 2: only download select files (STL,images & force and moments):
---------
+```bash
+pip install -U huggingface_hub hf_xet
+```
 
-Create the following bash script that could be adapted to loop through only select runs or to change to download different files e.g boundary/volume.
+Authentication is optional for this public dataset, but recommended for large downloads to avoid unauthenticated rate limits:
 
+```bash
+hf auth login
 ```
-#!/bin/bash
 
-# Set the path and prefix
-HF_OWNER="neashton"
-HF_PREFIX="drivaerml"
+### Preview or Download the Full Dataset
 
-# Set the local directory to download the files
-LOCAL_DIR="./drivaer_data"
+```bash
+hf download neashton/drivaerml \
+  --repo-type dataset \
+  --local-dir ./drivaerml_data \
+  --dry-run
+```
 
-# Create the local directory if it doesn't exist
-mkdir -p "$LOCAL_DIR"
+Remove `--dry-run` to start the download.
 
-# Loop through the run folders from 1 to 500
-for i in $(seq 1 500); do
-    RUN_DIR="run_$i"
-    RUN_LOCAL_DIR="$LOCAL_DIR/$RUN_DIR"
+### Preview Selected Metadata and Benchmark Files
 
-    # Create the run directory if it doesn't exist
-    mkdir -p "$RUN_LOCAL_DIR"
+```bash
+hf download neashton/drivaerml \
+  --repo-type dataset \
+  --local-dir ./drivaerml_data \
+  --include "force_mom_all.csv" \
+  --include "force_mom_constref_all.csv" \
+  --include "geo_parameters_all.csv" \
+  --include "splits/**" \
+  --dry-run
+```
 
-    # Download the drivaer_i.stl file
-    wget "https://huggingface.co/datasets/${HF_OWNER}/${HF_PREFIX}/resolve/main/$RUN_DIR/drivaer_$i.stl" -O "$RUN_LOCAL_DIR/drivaer_$i.stl"
+Remove `--dry-run` to download those files.
 
-    # Download the force_mom_i.csv file
-    wget "https://huggingface.co/datasets/${HF_OWNER}/${HF_PREFIX}/resolve/main/$RUN_DIR/force_mom_$i.csv" -O "$RUN_LOCAL_DIR/force_mom_$i.csv"
+### Preview a Small Per-Run Subset
 
-done
+```bash
+hf download neashton/drivaerml \
+  --repo-type dataset \
+  --local-dir ./drivaerml_data \
+  --include "run_1/drivaer_1.stl" \
+  --include "run_1/force_mom_1.csv" \
+  --dry-run
 ```
 
-Credits
------
+Remove `--dry-run` to download the selected files.
 
-* CFD solver and workflow development by Charles Mockett, Marian Fuchs, Louis Fliessbach, Henrik Hetmann, Thilo Knacke & Norbert Schonwald (UpstreamCFD)
-* Geometry parameterization by Vangelis Skaperdas, Grigoris Fotiadis (BETA-CAE Systems) & Astrid Walle (Siemens Energy) 
-* Meshing development workflow by Vangelis Skaperdas & Grigoris Fotiadis (BETA-CAE Systems) 
-* DrivAer advise and consultation by Burkhard Hupertz (Ford) 
-* Guidance on dataset preparation for ML by Danielle Maddix (Amazon Web Services - now NVIDIA)
-* Simulation runs, HPC setup and dataset preparation by Neil Ashton (Amazon Web Services - now NVIDIA) 
+## Credits
 
-License
-----
-This dataset is provided under the CC BY SA 4.0 license, please see LICENSE.txt for full license text.
+* CFD solver and workflow development by Charles Mockett, Marian Fuchs, Louis Fliessbach, Hendrik Hetmann, Thilo Knacke, and Norbert Schonwald (UpstreamCFD).
+* Geometry parameterization by Vangelis Skaperdas and Grigoris Fotiadis (BETA CAE Systems), and Astrid Walle (Siemens Energy).
+* Meshing workflow development by Vangelis Skaperdas and Grigoris Fotiadis (BETA CAE Systems).
+* DrivAer advice and consultation by Burkhard Hupertz (Ford).
+* Guidance on dataset preparation for ML by Danielle Maddix (Amazon Web Services, now NVIDIA).
+* Simulation runs, HPC setup, and dataset preparation by Neil Ashton (Amazon Web Services, now NVIDIA).
 
-version history:
----------------
-* 17/08/2026 - Added per-run native boundary polygon cell-area arrays and supporting metadata in `surface_cell_areas/`.
-* 17/08/2026 - Added deterministic official train/validation/test splits, including data-efficiency and out-of-distribution evaluation protocols.
+## License
 
-* 04/03/2025 - Now available on HuggingFace!
-  
-* 11/11/2024 - the 15 of the 17 cases that were missing are being considered for use as a blind study. For the time-being these are available but password protected in the file blind_15additional_cases_passwd_required.zip. Once we setup a benchmarking sysystem we will provide details on how people can test their methods against these 15 blind cases. 
+This dataset is provided under the CC BY-SA 4.0 license. See `LICENSE.txt` for the full license text.
 
-* 08/10/2024 - The OpenFOAM meshes (in OpenFOAM format) that were generated in ANSA have been uploaded to the openfoam_meshes folder. The 0 and system folders are just the default output from ANSA and were not those used in this study. Please refer to the arxiv paper for full details of the CFD setup. We hope that by providing the meshes, groups may wish to expand the dataset as they see fit.
+## Contact
 
-* 10/09/2024 - Run_0 has been added as a blind study for the AutoCFD4 workshop. Post-workshop the results from this additional run will be uploaded to the dataset.
+Dataset questions and corrections: Neil Ashton at [contact@caemldatasets.org](mailto:contact@caemldatasets.org).
 
-* 29/07/2024 - Note: please be aware currently runs 167, 211, 218, 221, 248, 282, 291, 295, 316, 325, 329, 364, 370, 376, 403, 473 are not in the dataset. 
+## Version History
 
-* 03/05/2024 - draft version produced 
+* 17/08/2026 - Added per-run native boundary polygon cell-area arrays and supporting metadata in `surface_cell_areas/`.
+* 17/08/2026 - Added deterministic official train/validation/test splits, including data-efficiency and out-of-distribution evaluation protocols.
+* 04/03/2025 - Dataset made available on Hugging Face.
+* 11/11/2024 - Fifteen of the seventeen previously missing cases were reserved for a blind study and made available in the password-protected `blind_15additional_cases_passwd_required.zip`; benchmarking instructions will be provided separately.
+* 08/10/2024 - Uploaded the OpenFOAM meshes generated in ANSA. The included `0/` and `system/` directories are default ANSA output and were not used in the study; consult the paper for the CFD setup.
+* 10/09/2024 - Added `run_0` as a blind study for the AutoCFD4 workshop, with results to be uploaded after the workshop.
+* 29/07/2024 - Runs 167, 211, 218, 221, 248, 282, 291, 295, 316, 325, 329, 364, 370, 376, 403, and 473 were not present in the dataset at this date.
+* 03/05/2024 - Draft version produced.