BLASTNet Momentum128 3D SR Dataset

Description

The BLASTNet Momentum128 3D SR Dataset is a benchmark dataset for developing and evaluating 3D super-resolution (SR) methods on turbulent flow data. It is a curated subset of the larger BLASTNet datasets, specifically designed to facilitate high-fidelity reconstruction of velocity fields from low-fidelity fields. This dataset includes 2,000 volumetric samples of 128\(^3\) grid points, each containing the three components of the velocity field (u, v, w) and the density. The sub-volumes were extracted from high-resolution direct numerical simulations (DNS) which span a range of flow regimes and statistical variations.

Each sample is accompanied by pre-computed low-resolution inputs at multiple downsampling ratios (e.g., 8×, 16×, and 32×), enabling the evaluation of SR models under different reconstruction challenges. Data is provided in binary .dat format using single-precision floating point (little-endian) ordering. The dataset is split into training, validation, and test sets, with metadata stored in accompanying CSV. These include physical statistics summary (e.g., skewness, kurtosis, variance).

Quick Info

• Kaggle Link
  
• Contributors: Wai Tong Chung, Bassem Akoush, Matthias Ihme
• N_x = 128, N_y = 128, N_z = 128, N_ɸ = 4
• Size = 75.15 GB
• DOI
  
• .bib