Rui Li1 · Biao Zhang1 · Zhenyu Li1 · Federico Tombari2,3 · Peter Wonka2,3
1KAUST · 2Google · 3Technical University of Munich
arXiv 2025
LaRI is a single-feed-forward method that models unseen 3D geometry using layered point maps. It enables complete, efficient, and view-aligned geometric reasoning from a single image.
- Inference code & Gradio demo
- Object- and Scene-level evaluation data & code (ETA: Mid May)
- Training data & code (ETA: Mid May)
conda create -n lari python=3.10 -y
conda activate lari
pip install -r requirements.txtWe currently provide the object-level model at our HuggingaFace Model Hub. Try out the examples or your own images via the methods below:
Launch the Gradio interface locally:
python app.pyOr try it online via HuggingFace Demo.
Run object-level modeling with:
python demo.py --image_path assets/cole_hardware.pngThe input image path is specified via
--image_path. Set--is_remove_backgroundto remove the background. Layered depth maps and the 3D model will be saved in the./resultsdirectory by default.
Please cite our paper if you use the code in this repository:
@inproceedings{li2025lari,
title={LaRI: Layered Ray Intersections for Single-view 3D Geometric Reasoning},
author={Li, Rui and Zhang, Biao and Li, Zhenyu and Tombari, Federico and Wonka, Peter},
booktitle={arXiv preprint arXiv:2504.18424},
year={2025}
}