Mohamedelrefaie/DrivAerNet

Comprises 8,150 parametrically-defined car geometries with 10+ data modalities including volumetric CFD fields (pressure, velocity, turbulence), surface coefficients, point clouds, meshes, and photorealistic renderings—enabling training of neural surrogates for aerodynamic prediction and design exploration. Parametric designs span conventional body styles (fastback, notchback, estateback) with semantic annotations across 29 car components, covering both ICE and EV architectures. Hosted on Harvard Dataverse (39TB), the dataset pairs with CarBench leaderboard for standardized model evaluation on generalization to unseen aerodynamic configurations.

484 stars.