---
title: LitePT
emoji: 🧊
colorFrom: green
colorTo: yellow
sdk: gradio
sdk_version: 6.30.0
app_file: app.py
short_description: 3D point cloud semantic segmentation with LitePT
python_version: "3.10"
startup_duration_timeout: 30m
license: mit
models:
  - prs-eth/LitePT
---

# LitePT: Lighter Yet Stronger Point Transformer

Gradio demo for [LitePT](https://github.com/prs-eth/LitePT) (CVPR 2026) — per-point semantic segmentation of indoor RGB scans (ScanNet / Structured3D checkpoints) and outdoor LiDAR sweeps (nuScenes checkpoint). Weights: [prs-eth/LitePT](https://huggingface.co/prs-eth/LitePT). Model code under MIT license (see `LICENSE-LitePT`).

## Example assets

- `examples/nuscenes_sample.bin` — LiDAR sweep from the authors' demo dataset [prs-eth/LitePT_demo](https://huggingface.co/datasets/prs-eth/LitePT_demo) (MIT), originally from nuScenes.
- `examples/icl_nuim_livingroom.ply` — the Open3D living-room fragment reconstructed from the [ICL-NUIM](https://www.doc.ic.ac.uk/~ahanda/VaFRIC/iclnuim.html) living-room scene (Handa et al., ICRA 2014), licensed **CC BY 3.0**; rotated to z-up for this demo.