Update Dataset Card for NavierStokes-2D dataset

README.md

@@ -1,3 +1,85 @@
 ---
 license: mit
+task_categories:
+  - time-series-forecasting
+  - image-to-image
+language:
+  - en
+tags:
+  - pde
+  - navier-stokes
+  - fluid-dynamics
+  - turbulence
+  - neural-operator
+  - fno
+  - cfd
+  - surrogate-modeling
+  - scientific-ml
+  - huggingscience
+  - science
+pretty_name: PDE Arena — Navier–Stokes 2D
 ---
+
+# Dataset Card for pdearena/NavierStokes-2D
+
+2D Navier–Stokes simulation data for scientific ML (operator learning, surrogates, forecasting). Files are HDF5. License: MIT.
+
+## Dataset Details
+
+- **Provided by:** PDE Arena org on Hugging Face (see also Maxwell-3D, ShallowWater-2D).
+- **License:** MIT (per repo).
+- **Format:** `.h5` files (HDF5) containing simulation tensors.
+- **Typical use:** Train/evaluate neural operators (FNO/UNO/DeepONet), PINNs baselines, and CFD surrogates.
+
+## Dataset Structure
+
+- **Splits:** `train`, `validation`, `test`.
+- **Common arrays (by file):** velocity components, pressure (naming may vary).
+- **Shape note:** Many PDE datasets follow `[t, x, y, channels]` or `[channels, t, x, y]`. Inspect keys to confirm.
+
+## How to Load
+
+```python
+# 1) Simple: download a file and inspect keys
+import h5py, numpy as np, requests, io
+
+url = "https://huggingface.co/datasets/pdearena/NavierStokes-2D/resolve/main/NavierStokes2D_test_509778_0.50000.h5"
+f = h5py.File(io.BytesIO(requests.get(url).content), "r")
+print(list(f.keys()))           # e.g., ['u', 'v', 'p'] or similar
+u = np.array(f["u"])            # (T, H, W)
+```
+```python
+# 2) Convert to PyTorch tensors (example)
+import torch
+U = torch.from_numpy(u).float()  # (T, H, W)
+# if channels present: (T, H, W, C) -> (T, C, H, W)
+
+```
+
+## Uses
+
+### Direct Use
+- **Neural operators / surrogates:** learn mappings between initial/boundary conditions and flow fields.
+- **Forecasting:** autoregressive prediction of future velocity/pressure frames.
+- **Downstream eval:** error metrics in pixel/physical/Fourier space (e.g., **RMSE**, **nRMSE**, **fRMSE**).
+
+### Out-of-Scope
+- **Clinical or safety-critical decisions.**
+- **Treating as a labeled vision benchmark** (no class labels).
+
+## References
+- **PDE Arena (org page):** dataset family (Navier–Stokes 2D/conditioned, ShallowWater-2D, Maxwell-3D): https://huggingface.co/datasets/pdearena
+- **Hugging Face:** general datasets hub: https://huggingface.co/datasets
+- **PDEBench (related benchmark & metrics):** Burgers, Darcy, Navier–Stokes, Shallow Water, etc.: https://darus.uni-stuttgart.de/
+- **Neural operators (FNO paper):** https://arxiv.org/abs/2010.08895
+
+## Citation
+```bibtex
+@misc{pdearena_navierstokes2d,
+  title        = {PDE Arena — Navier–Stokes 2D},
+  author       = {PDE Arena},
+  howpublished = {\url{https://huggingface.co/datasets/pdearena/NavierStokes-2D}},
+  year         = {2023}
+}
+```
+