Datasets:

Tasks:
Other
ArXiv:
License:
File size: 2,759 Bytes
4e8e99a
 
7c38965
 
fc338a3
 
 
4e8e99a
 
b6ec201
4e8e99a
7c38965
 
 
4e8e99a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7c38965
 
 
 
 
 
4e8e99a
 
 
 
 
 
 
 
 
 
 
 
 
 
7c38965
4e8e99a
 
 
7c38965
4e8e99a
 
 
 
 
 
fc338a3
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
---
license: mit
task_categories:
- other
tags:
- physics-simulation
- physics-foundation-model
---

# GeoPT

[Project Page](https://physics-scaling.github.io/GeoPT/) | [Paper](https://huggingface.co/papers/2602.20399) | [GitHub](https://github.com/Physics-Scaling/GeoPT)

This repository contains the physics simulation data for the paper **GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training**.

GeoPT is a unified model pre-trained on large-scale geometric data for general physics simulation, unlocking a scalable path for neural simulation.

<p align="center">
<img src="assets/GeoPT.png" height = "120" alt="" align=center />
</p>

## Overview

GeoPT is evaluated on the following five simulation tasks.

| Dataset   | Mesh Size | Variable                       | Training | Test | Total Size | Source                                                       |
| --------- | --------- | ------------------------------ | -------- | ---- | ---------- | ------------------------------------------------------------ |
| DrivAerML | ~160M     | Geometry                       | 100      | 20   | ~6TB       | [Link](https://huggingface.co/datasets/neashton/drivaerml)   |
| NASA-CRM  | ~450K     | Geometry, Speed, AoA           | 105      | 44   | ~3GB       | [Link](https://drive.google.com/drive/folders/1KhoZiEHlZhGI8omMwHrp2mZRKGiSAydO) |
| AirCraft  | ~330K     | Geometry, Speed, AoA, Sideslip | 100      | 50   | ~7GB       | [Transolver++](https://arxiv.org/abs/2502.02414)             |
| DTCHull   | ~240K     | Geometry, Yaw Angle            | 100      | 20   | ~2GB       | GeoPT                                                        |
| Car-Crash | ~1M       | Impact Angle                   | 100      | 30   | ~8GB       | GeoPT                                                        |

## Load Data

```python
from datasets import load_dataset

# For AirCraft, DTCHull, Car-Crash, Radiosity
load_dataset("GeoPT/Downstream_Physics_Simulation") 

# For DrivAerML
load_dataset("neashton/drivaerml") 
```

NASA-CRM can be obtained from [Google Drive](https://drive.google.com/drive/folders/1KhoZiEHlZhGI8omMwHrp2mZRKGiSAydO).

## Examples

<p align="center">
<img src="assets/examples.png" height = "140" alt="" align=center />
</p>

## Citation

If you find this repo useful, please cite our paper. 

```bibtex
@article{wu2026GeoPT,
  title={GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training},
  author={Haixu Wu, Minghao Guo, Zongyi Li, Zhiyang Dou, Mingsheng Long, Kaiming He, Wojciech Matusik},
  journal={arXiv preprint arXiv:2602.20399},
  year={2026}
}
```

## Contact

If you have any questions or want to use the code, please contact Haixu Wu (wuhaixu98@gmail.com) and Minghao Guo (guomh2014@gmail.com).