| layout | title | description |
|---|---|---|
default |
Benchmarks |
Benchmarking Scaling Behavior in 3D Super-resolution of Turbulent Flows |
We demonstrate the utility of BLASTNet 2.0 data for open and fair evaluation of popular ML approaches in a recent benchmark. From BLASTNet 2.0, we pre-process direct numerical simulation data to form the Momentum128 3D SR dataset for benchmarking 3D super-resolution of turbulent flows.
- Curate BLASTNet 2.0, a diverse public 3D compressible turbulent flow DNS dataset.
- Benchmark performance and cost of five 3D ML approache for SR with this publicly accessible dataset.
- Show that SR model performance can scale with the logarithm of model size and cost.
- Demonstrate the persisting benefits of a popular physics-based gradient loss term with increasing model size.
This is summarized in the figure below:
The predictions from the five model approaches are compared with tricubic interpolation:
Here's some quantitative results in the form of neural scaling laws when measured with the SSIM of the subgrid-scale stress.
Check out our >50 page NeurIPS paper for more results and documentation of this benchmark.
W. T. Chung, B. Akoush, P. Sharma, A. Tamkin, K. S. Jung, J. H. Chen, J. Guo, D. Brouzet, M. Talei, B. Savard, A.Y. Poludnenko & M. Ihme. Turbulence in Focus: Benchmarking Scaling Behavior of 3D Volumetric Super-Resolution with BLASTNet 2.0 Data. Advances in Neural Information Processing Systems (2023) 36. [Paper, .bib]