Add dataset card, link to paper and GitHub repository

README.md

@@ -1,3 +1,37 @@
 ---
 license: mit
----
+task_categories:
+- other
+tags:
+- 3d
+- point-cloud
+---
+
+# Point Cloud Corruption Datasets
+
+This repository hosts the clean training and corrupted test point cloud datasets utilized in the paper [Point-Selection Fine-Tuning Framework for Robust Point Cloud Classification](https://arxiv.org/abs/2607.19711) (PRCV 2026).
+
+Official Code: [GitHub - CVChMA/PSFT](https://github.com/CVChMA/PSFT)
+
+## Dataset Specifications
+
+These benchmarks are used to evaluate how well 3D point cloud classification models perform under various common corruptions (such as noise, dropping points, or scaling).
+
+| Dataset | Classes | Clean Training Data | Corrupted Test Data |
+| :--- | :--- | :--- | :--- |
+| `ModelNet-C` | 40 | `modelnet40_ply_hdf5_2048` | `modelnet_c` |
+| `ModelNet40-C` | 40 | `modelnet40_ply_hdf5_2048` | `modelnet40_c` |
+| `ScanObjectNN-C` | 15 | `ScanObjectNN/h5_files/main_split` | `scanobjectnn_c` |
+
+## Citation
+
+If you find these datasets or the PSFT framework useful for your research, please cite the following paper:
+
+```bibtex
+@misc{arxiv2607.19711,
+  title        = {Point-Selection Fine-Tuning Framework for Robust Point Cloud Classification},
+  author       = {Da Li, Chang Ma, and Dongfu Yin},
+  year         = {2026},
+  url          = {https://arxiv.org/abs/2607.19711}
+}
+```