---
license: mit
pipeline_tag: image-segmentation
library_name: pytorch
tags:
- point-cloud
- point-cloud-backbone
- graph-learning
- pytorch
authors:
- Yuanwen Yue
- Damien Robert
- Jianyuan Wang
- Sunghwan Hong
- Jan Dirk Wegner
- Christian Rupprecht
- Konrad Schindler
---

This repository contains model weights for **LitePT: Lighter Yet Stronger Point Transformer**, a lightweight, high-performance 3D point cloud architecture. 

LitePT embodies the simple principle "convolutions for low-level geometry, attention for high-level relations" and strategically places only the required operations at each hierarchy level. LitePT is equipped with a novel, parameter-free 3D positional encoding, PointROPE. The resulting model achieves state-of-the-art performance while being significantly more efficient.

## Paper & Resources
- **Paper:** [LitePT: Lighter Yet Stronger Point Transformer](https://huggingface.co/papers/2512.13689)
- **Arxiv:** [https://arxiv.org/abs/2512.13689](https://arxiv.org/abs/2512.13689)
- **Project Page:** [https://litept.github.io/](https://litept.github.io/)
- **Codebase:** [https://github.com/prs-eth/LitePT](https://github.com/prs-eth/LitePT)

## Models
We release the pretrained model weights for the benchmarks we reported in our paper.

### Semantic segmentation 
| Model | Params | Benchmark  | val mIoU | Config | Checkpoint |
|:-|-:|:-:|:-:|:-:|:-:|
| LitePT-S | 12.7M | NuScenes | 82.2 | [link](https://github.com/prs-eth/LitePT/blob/main/configs/nuscenes/semseg-litept-small-v1m1.py) | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/nuscenes-semseg-litept-small-v1m1/model/model_best.pth) |
| LitePT-S | 12.7M | Waymo | 73.1 |[link](https://github.com/prs-eth/LitePT/blob/main/configs/waymo/semseg-litept-small-v1m1.py) | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/waymo-semseg-litept-small-v1m1/model/model_best.pth) |
| LitePT-S | 12.7M | ScanNet  | 76.5 |[link](https://github.com/prs-eth/LitePT/blob/main/configs/scannet/semseg-litept-small-v1m1.py) | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/scannet-semseg-litept-small-v1m1/model/model_best.pth) |
| LitePT-S | 12.7M | Structured3D | 83.6 | [link](https://github.com/prs-eth/LitePT/blob/main/configs/structured3d/semseg-litept-small-v1m1.py) | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/structured3d-semseg-litept-small-v1m1/model/model_best.pth) |
| LitePT-B | 45.1M | Structured3D | 85.1  | [link](https://github.com/prs-eth/LitePT/blob/main/configs/structured3d/semseg-litept-base-v1m1.py) | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/structured3d-semseg-litept-base-v1m1/model/model_best.pth) |
| LitePT-L | 85.9M | Structured3D | 85.4 | [link](https://github.com/prs-eth/LitePT/blob/main/configs/structured3d/semseg-litept-large-v1m1.py) | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/structured3d-semseg-litept-large-v1m1/model/model_best.pth) |

### Instance segmentation 
| Model | Params | Benchmark  | mAP<sub>25</sub> | mAP<sub>50</sub> | mAP | Config | Checkpoint |
|:-|-:|:-:|:-:|:-:|:-:|:-:|:-:|
| LitePT-S* | 16.0M | ScanNet | 78.5 | 64.9 | 41.7 | [link](https://github.com/prs-eth/LitePT/blob/main/configs/scannet/insseg-litept-small-v1m2.py) | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/scannet-insseg-litept-small-v1m2/model/model_best.pth) |
| LitePT-S* | 16.0M | ScanNet200 | 40.3 | 33.1 | 22.2 | [link](https://github.com/prs-eth/LitePT/blob/main/configs/scannet200/insseg-litept-small-v1m2.py) | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/scannet200-insseg-litept-small-v1m2/model/model_best.pth) |
### Object detection
| Model | Params | Benchmark  | mAPH | Config | Checkpoint |
|:-|-:|:-:|:-:|:-:|:-:|
| LitePT | 9.0M | Waymo  | 70.7 | link | [Download](https://huggingface.co/yuanwenyue/LitePT/blob/main/waymo-objdet-litept-small-v1m3/model/model_best.pth) |

## Citation

```
@article{yuelitept2025,
    title={{LitePT: Lighter Yet Stronger Point Transformer}},
    author={Yue, Yuanwen and Robert, Damien and Wang, Jianyuan and Hong, Sunghwan and Wegner, Jan Dirk and Rupprecht, Christian and Schindler, Konrad},
    journal={arXiv preprint arXiv:2512.13689},
    year={2025}
}
```