README.md

---
license: mit
task_categories:
- image-to-3d
tags:
- 3d-physics
- material-properties
- gaussian-splatting
- clip-features
- 3d-assets
---

# Pixie Dataset

This dataset contains data and pre-trained models for the paper [Pixie: Fast and Generalizable Supervised Learning of 3D Physics from Pixels](https://huggingface.co/papers/2508.17437).

- Project Page: https://pixie-3d.github.io/
- Code: https://github.com/vlongle/pixie

## Contents

- `checkpoints_continuous_mse/`: Continuous material property prediction model checkpoints
- `checkpoints_discrete/`: Discrete material classification model checkpoints
- `real_scene_data/`: Real scene data for evaluation
- `real_scene_models/`: Trained models for real scenes

## Sample Usage

First, use the download script in the Pixie repository to automatically download this data and models:

```bash
python scripts/download_data.py
```

Then, you can run the main pipeline with a synthetic Objaverse object, for example:

```python
python pipeline.py obj_id=f420ea9edb914e1b9b7adebbacecc7d8 material_mode=neural
```
This command will:
1. Download the specified Objaverse asset.
2. Render it and train 3D representations (NeRF, Gaussian Splatting).
3. Generate a voxel feature grid.
4. Use the trained neural networks to predict the physics field.
5. Run the MPM physics solver using the predicted physics parameters.

For more detailed usage, including real-scene processing and training, refer to the [Github repository's usage section](https://github.com/vlongle/pixie#usage).

## Citation

If you find this work useful, please consider citing:

```bibtex
@article{le2025pixie,
  title={Pixie: Fast and Generalizable Supervised Learning of 3D Physics from Pixels},
  author={Le, Long and Lucas, Ryan and Wang, Chen and Chen, Chuhao and Jayaraman, Dinesh and Eaton, Eric and Liu, Lingjie},
  journal={arXiv preprint arXiv:2508.17437},
  year={2025}
}
```