Add pipeline tag, paper link, and comprehensive model card

README.md

@@ -1,3 +1,106 @@
 ---
 license: mit
+pipeline_tag: other
 ---
+
+# Point-Selection Fine-Tuning Framework for Robust Point Cloud Classification (PRCV 2026)
+
+This repository contains the official pre-trained model checkpoints and dataset references for **PSFT**, a parameter-efficient point-selection fine-tuning framework that improves the robustness of 3D point cloud pre-trained models against noise and corruptions.
+
+[[\ud83d\udcc4 Paper (arXiv:2607.19711)](https://arxiv.org/abs/2607.19711)] [[\ud83d\udcbb Code](https://github.com/CVChMA/PSFT)]
+
+---
+
+## Pretrained Models
+
+The released checkpoints are hosted in this repository. Each checkpoint contains the classifier model and point-selection model used by `test_ps.py`.
+
+ModelNet-C and ModelNet40-C share the same checkpoints. Use the `ModelNet-C` checkpoint files for both `ModelNet-C` and `ModelNet40-C` evaluation.
+
+| Dataset | Backbone | Method | Augmentation | Download |
+| :--- | :--- | :--- | :--- | :--- |
+| `ModelNet-C` / `ModelNet40-C` | `Point-BERT` | PSFT | None | [Download](./ModelNet-C/Train-PS-Point-BERT-PG-ModelNet-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt) |
+| `ModelNet-C` / `ModelNet40-C` | `Point-BERT` | PSFT + Aug. | WOLFMix | [Download](./ModelNet-C/Train-PS-Point-BERT-PG-ModelNet-C-add_FFM-Augmentation%3AWOLFMix-train_alpha%3A0.5_epoch_300.pt) |
+| `ModelNet-C` / `ModelNet40-C` | `Point-MAE` | PSFT | None | [Download](./ModelNet-C/Train-PS-Point-MAE-PG-ModelNet-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt) |
+| `ModelNet-C` / `ModelNet40-C` | `Point-MAE` | PSFT + Aug. | WOLFMix | [Download](./ModelNet-C/Train-PS-Point-MAE-PG-ModelNet-C-add_FFM-Augmentation%3AWOLFMix-train_alpha%3A0.5_epoch_300.pt) |
+| `ModelNet-C` / `ModelNet40-C` | `ULIP-2` | PSFT | None | [Download](./ModelNet-C/Train-PS-ULIP-2-PG-ModelNet-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt) |
+| `ModelNet-C` / `ModelNet40-C` | `ULIP-2` | PSFT + Aug. | WOLFMix | [Download](./ModelNet-C/Train-PS-ULIP-2-PG-ModelNet-C-add_FFM-Augmentation%3AWOLFMix-train_alpha%3A0.5_epoch_300.pt) |
+| `ModelNet-C` / `ModelNet40-C` | `Uni3d-B` | PSFT | None | [Download](./ModelNet-C/Train-PS-Uni3d-B-PG-ModelNet-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt) |
+| `ModelNet-C` / `ModelNet40-C` | `Uni3d-B` | PSFT + Aug. | WOLFMix | [Download](./ModelNet-C/Train-PS-Uni3d-B-PG-ModelNet-C-add_FFM-Augmentation%3AWOLFMix-train_alpha%3A0.5_epoch_300.pt) |
+| `ScanObjectNN-C` | `Point-BERT` | PSFT | None | [Download](./ScanObjectNN-C/Train-PS-Point-BERT-PG-ScanObjectNN-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt) |
+| `ScanObjectNN-C` | `Point-BERT` | PSFT + Aug. | WOLFMix | [Download](./ScanObjectNN-C/Train-PS-Point-BERT-PG-ScanObjectNN-C-add_FFM-Augmentation%3AWOLFMix-train_alpha%3A0.5_epoch_300.pt) |
+| `ScanObjectNN-C` | `Point-MAE` | PSFT | None | [Download](./ScanObjectNN-C/Train-PS-Point-MAE-PG-ScanObjectNN-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt) |
+| `ScanObjectNN-C` | `Point-MAE` | PSFT + Aug. | WOLFMix | [Download](./ScanObjectNN-C/Train-PS-Point-MAE-PG-ScanObjectNN-C-add_FFM-Augmentation%3AWOLFMix-train_alpha%3A0.5_epoch_300.pt) |
+| `ScanObjectNN-C` | `ULIP-2` | PSFT | None | [Download](./ScanObjectNN-C/Train-PS-ULIP-2-PG-ScanObjectNN-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt) |
+| `ScanObjectNN-C` | `ULIP-2` | PSFT + Aug. | WOLFMix | [Download](./ScanObjectNN-C/Train-PS-ULIP-2-PG-ScanObjectNN-C-add_FFM-Augmentation%3AWOLFMix-train_alpha%3A0.5_epoch_300.pt) |
+| `ScanObjectNN-C` | `Uni3d-B` | PSFT | None | [Download](./ScanObjectNN-C/Train-PS-Uni3d-B-PG-ScanObjectNN-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt) |
+| `ScanObjectNN-C` | `Uni3d-B` | PSFT + Aug. | WOLFMix | [Download](./ScanObjectNN-C/Train-PS-Uni3d-B-PG-ScanObjectNN-C-add_FFM-Augmentation%3AWOLFMix-train_alpha%3A0.5_epoch_300.pt) |
+
+---
+
+## Datasets
+
+Below are the links to clean training data and corrupted test data:
+
+| Dataset | Classes | Training data | Corrupted test data | Checkpoint |
+| :--- | :--- | :--- | :--- | :--- |
+| `ModelNet-C` | 40 | [Download](https://huggingface.co/datasets/SZUChangMa/PointCloudCorruption/tree/main/modelnet40_ply_hdf5_2048) | [Download](https://huggingface.co/datasets/SZUChangMa/PointCloudCorruption/tree/main/modelnet_c) | `ModelNet-C` ckpt |
+| `ModelNet40-C` | 40 | [Download](https://huggingface.co/datasets/SZUChangMa/PointCloudCorruption/tree/main/modelnet40_ply_hdf5_2048) | [Download](https://huggingface.co/datasets/SZUChangMa/PointCloudCorruption/tree/main/modelnet40_c) | reuse `ModelNet-C` ckpt |
+| `ScanObjectNN-C` | 15 | [Download](https://huggingface.co/datasets/SZUChangMa/PointCloudCorruption/tree/main/ScanObjectNN/h5_files/main_split) | [Download](https://huggingface.co/datasets/SZUChangMa/PointCloudCorruption/tree/main/scanobjectnn_c) | `ScanObjectNN-C` ckpt |
+
+---
+
+## Quick Start
+
+### Installation
+
+Clone the official repository and install the dependencies:
+
+```bash
+git clone https://github.com/CVChMA/PSFT.git
+cd PSFT
+
+pip install torch torchvision torchaudio
+pip install numpy h5py scikit-learn timm pyyaml huggingface_hub
+pip install "git+https://github.com/erikwijmans/Pointnet2_PyTorch.git#egg=pointnet2_ops&subdirectory=pointnet2_ops_lib"
+pip install utils/KNN_CUDA-0.2-py3-none-any.whl
+```
+
+### Download Checkpoints
+
+You can easily download all the released checkpoints to your local machine:
+
+```bash
+pip install huggingface_hub
+huggingface-cli download SZUChangMa/PSFT \
+  --local-dir ckpts/PSFT \
+  --include "ModelNet-C/*.pt" "ScanObjectNN-C/*.pt"
+```
+
+### Evaluation
+
+To evaluate a downloaded PSFT checkpoint, execute:
+
+```bash
+CUDA_VISIBLE_DEVICES=0 python test_ps.py \
+  --model_path ckpts/PSFT/ModelNet-C/Train-PS-Point-BERT-PG-ModelNet-C-add_FFM-Augmentation%3ANone-train_alpha%3A0.5_epoch_300.pt \
+  --model_name Point-BERT \
+  --train_mode PG \
+  --add_FFM \
+  ModelNet-C
+```
+
+Please check the official [GitHub repository](https://github.com/CVChMA/PSFT) for detailed guidance on preparing pre-trained backbone models and training configs.
+
+## Citation
+
+If you find this work or the pre-trained weights useful, please cite:
+
+```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}
+}
+```