license: cc-by-4.0
tags:
- ShapeNetCar
- 3D Aerodynamic Flow Field Prediction
language:
- en
- zh
ShapeNetCar
Dataset Description
The ShapeNetCar dataset comes from the paper Learning Three-dimensional Flow for Interactive Aerodynamic Design by Umetani and Bickel, published in ACM Transactions on Graphics (SIGGRAPH 2018). Based on three-dimensional car geometries from ShapeNet, the dataset uses CFD simulations to obtain velocity fields around the vehicles, surface pressure, and drag coefficients. It supports research on rapidly predicting aerodynamic physical fields and forces from three-dimensional geometry.
Supported Tasks
This standardized data repository is organized for the ShapeNetCar external flow field prediction task. It contains raw training data, preprocessed graph data, normalization statistics, linear regression utility code, paper figure and table data, and result comparison plots. It can be used for training, inference, evaluation, and visualization with the OneScience/Transolver-Car-Design model. This repository contains 889 car geometry samples grouped from param0 through param8. It retains the raw CFD arrays, meshes, and VTK files while also providing preprocessed samples for graph neural network models. Each preprocessed sample consists of node features, target physical fields, three-dimensional coordinates, a surface mask, and graph edge indices.
Dataset Format and Structure
In the preprocessed graph data, each preprocessed_data/param*/<sample_id>/ directory contains:
| File | shape | dtype | Description |
|---|---|---|---|
x.npy |
[num_nodes, 7] |
float64 |
Node input features, including position, signed distance function (SDF), normals, and other features |
y.npy |
[num_nodes, 4] |
float64 |
Target physical fields; the first three channels are velocity components and the last channel is pressure |
pos.npy |
[num_nodes, 3] |
float32 |
Three-dimensional node coordinates |
surf.npy |
[num_nodes] |
float64 |
Surface node mask |
edge_index.npy |
[2, num_edges] |
int64 |
Edge indices of the graph structure |
Normalization statistics files are located in stats/:
| File | shape | dtype | Description |
|---|---|---|---|
mean_in.npy |
[7] |
float32 |
Mean of the input features |
std_in.npy |
[7] |
float32 |
Standard deviation of the input features |
mean_out.npy |
[4] |
float32 |
Mean of the output physical quantities |
std_out.npy |
[4] |
float32 |
Standard deviation of the output physical quantities |
How to Use the Dataset
This dataset is compatible with the OneScience-Group/Transolver-Car-Design model. At runtime, scripts in the model package can organize the data from this repository under data/.
- Files and Download:
hf download --dataset OneScience-Group/ShapeNetCar --local-dir ./data
Official OneScience Information
| Platform | OneScience Main Repository | Skills Repository |
|---|---|---|
| Gitee | https://gitee.com/onescience-ai/onescience | https://gitee.com/onescience-ai/oneskills |
| GitHub | https://github.com/onescience-ai/OneScience | https://github.com/onescience-ai/oneskills |
Citation and License
- This dataset is organized or converted from the ShapeNetCar three-dimensional aerodynamic flow field data released by Nobuyuki Umetani and Bernd Bickel. When using it, please cite the original ShapeNetCar paper: Learning Three-dimensional Flow for Interactive Aerodynamic Design
- If this dataset is used to reproduce the Transolver car aerodynamic design experiments, please also cite the original Transolver paper: Transolver: A Fast Transformer Solver for PDEs on General Geometries
- This repository retains source attribution and is organized for automated OneScience ModelScope execution scenarios. Before public distribution, please verify the license requirements of the upstream project.