Olbedo: An Albedo and Shading Aerial Dataset for Large-Scale Outdoor Environments

This repository contains the official implementation and inference code for Olbedo.

🔗 Resources

We provide a comprehensive suite of resources for this project:

• Project Page: https://gdaosu.github.io/olbedo/
• Interactive Demo: Hugging Face Spaces
• Pre-trained Models: Hugging Face Model Hub
• Dataset: Hugging Face Datasets
• Paper: arXiv:2602.22025

🚀 Usage

We provide Docker support to ensure a consistent environment for running inference.

1. Build the Environment

First, clone this repository and build the Docker image. This will set up all necessary dependencies.

bash build_docker.sh

2. Run Inference

To run inference on your own images, use the run_inference.sh script. You must specify the input directory containing your images and the output directory where results will be saved.

bash run_inference.sh <input_directory> <output_directory>

📂 Data & Models

If you wish to use the data or models separately, they are hosted on Hugging Face:

Resource	Link	Description
Model Weights	Download Here	Pre-trained checkpoints for the Olbedo architecture.
Dataset	Download Here	The dataset used for training and evaluation.

📝 Citation

If you find this project useful for your research, please consider citing our work:

@misc{song2026olbedoalbedoshadingaerial,
      title={Olbedo: An Albedo and Shading Aerial Dataset for Large-Scale Outdoor Environments}, 
      author={Shuang Song and Debao Huang and Deyan Deng and Haolin Xiong and Yang Tang and Yajie Zhao and Rongjun Qin},
      year={2026},
      eprint={2602.22025},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={[https://arxiv.org/abs/2602.22025](https://arxiv.org/abs/2602.22025)}, 
}

🙏 Acknowledgements

This codebase is largely built upon the following excellent projects:

• Marigold
• RGBX

We thank the authors for their open-source contributions.

🎫 License

The dataset associated with this work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0) (as defined in LICENSE-DATA).

This code of this work is licensed under the Apache License, Version 2.0 (as defined in the LICENSE).

The Marigold pretrained and fine-tuned models are licensed under RAIL++-M License (as defined in the LICENSE-MODEL).

The RGBX pretrained and fine-tuned models are licensed under ADOBE RESEARCH LICENSE(as defined in the LICENSE-ADOBE).

By downloading and using the dataset, code and models you agree to the terms in LICENSE-DATA, LICENSE, LICENSE-MODEL, and LICENSE-ADOBE respectively.