README.md

metadata

pipeline_tag: depth-estimation
license: apache-2.0

Depth Any Panoramas:
A Foundation Model for Panoramic Depth Estimation

Xin Lin · Meixi Song · Dizhe Zhang · Wenxuan Lu · Haodong Li
Bo Du · Ming-Hsuan Yang · Truong Nguyen · Lu Qi

This repository presents Depth Any Panoramas (DAP), a panoramic metric depth foundation model that generalizes across diverse scene distances. It explores a data-in-the-loop paradigm for both data construction and framework design, combining public datasets, high-quality synthetic data, and real panoramic images. The model adopts DINOv3-Large as its backbone and introduces innovations such as a plug-and-play range mask head, sharpness-centric optimization, and geometry-centric optimization to enhance robustness and ensure geometric consistency across views. Experiments demonstrate strong performance and zero-shot generalization, providing robust and stable metric predictions in diverse real-world scenes.

More details can be found in the paper and on the project page.

🔨 Installation

Clone the repo first:

git clone https://github.com/Insta360-Research-Team/DAP
cd DAP

(Optional) Create a fresh conda env:

conda create -n dap python=3.12
conda activate dap

Install necessary packages (torch > 2):

# pytorch (select correct CUDA version, we test our code on torch==2.7.1 and torchvision==0.22.1)
pip install torch==2.7.1 torchvision==0.22.1

# other dependencies
pip install -r requirements.txt

🤝 Pre-trained model

Please download the pretrained model from this Hugging Face repository: Insta360-Research/DAP-weights.

📒 Inference

python test/infer.py 

🚀 Evaluation

python test/eval.py 

🤝 Acknowledgement

We appreciate the open source of the following projects:

• PanDA
• Depth-Anything-V2

Citation

If you find our work useful, please cite our paper:

@article{lin2025dap,
          title={Depth Any Panoramas: A Foundation Model for Panoramic Depth Estimation},
          author={Lin, Xin and Song, Meixi and Zhang, Dizhe and Lu, Wenxuan and Li, Haodong and Du, Bo and Yang, Ming-Hsuan and Nguyen, Truong and Qi, Lu},
          journal={arXiv},
          year={2025}
        }