[ECCV 2026] Holo360D: A Large-Scale Real-World Dataset with Continuous Trajectories for Advancing Panoramic 3D Reconstruction and Beyond
## 🎉 NEWS
- [2026.07.04] 🔥 We have fixed incorrect RGB face masks within the Holo360D **test set**. If you downloaded **the test set before July 2, 2026**, please re-download the updated test dataset.
- [2026.07.01] 🔥 We have released the inference code for **fine-tuned Pi3** on the Holo360D, which supports both **single-view and multi-view panoramic 3D reconstruction**.
- [2026.06.27] 🔥 We have released all data of the **Holo360D** dataset on Hugging Face, featuring 56 indoor scenes and 19 outdoor scene.
- [2026.06.18] 🎉 **Holo360D** has been accepted by ECCV 2026.
- [2026.06.03] 🔥 We have released test data of the **Holo360D** dataset on Hugging Face, featuring 13 indoor scenes and 4 outdoor scene.
---
## ✨ Overview
We present Holo360D, the first large-scale real-world panoramic 3D dataset, containing 109,495 panoramas paired with LiDAR-derived ground truth, including precise meshes, point clouds, depth maps, and camera poses. More importantly, Holo360D is the first panoramic dataset to offer accurately aligned high-completeness depth
maps with continuous camera trajectories over long sequences.
Key characteristics (from the paper):
- Large-scale real-world 360 panorama 3D dataset.
- Continuous trajectory capture for multi-view settings.
- Accurately aligned high-completeness depth maps for training and testing.
- A benchmark setup for model fine-tuning and evaluation.
## 🔮 Inference
We have released the inference code and checkpoint for fine-tuned Pi3 on the Holo360D ([click here](https://github.com/Jou719/Holo360D/tree/main/Pi3_Finetuned_Holo360d))
## 📦 Dataset Structure
```
Holo360D/
├── train/
│ ├── Indoor_xxx/
│ │ ├── rgb/ # panoramic RGB images (.jpg)
│ │ ├── depth/ # depth maps (.exr)
│ │ │ ├── mesh_depth/ # depth maps (.exr)
│ │ │ ├── pointcloud_depth/ # depth maps (.exr)
│ │ │ ├── visual_mesh_depth/ # visualization (.jpg)
│ │ │ └── visual_pointcloud_depth/# visualization (.jpg)
│ │ ├── mask/ # masks (.jpg)
│ │ └── poses/ # camera poses (.txt)
│ ├── Indoor_xxx/
│ ├── Outdoor_xxx/
│ │ ├── rgb/ # panoramic RGB images (.jpg)
│ │ ├── depth/
│ │ │ ├── mesh_depth/ # depth maps (.exr)
│ │ │ ├── pointcloud_depth/ # depth maps (.exr)
│ │ │ ├── visual_mesh_depth/ # visualization (.jpg)
│ │ │ └── visual_pointcloud_depth/# visualization (.jpg)
│ │ ├── mask/ # masks (.jpg)
│ │ └── poses/ # camera poses (.txt)
│ ├── Outdoor_xxx/
│ └── ...
└── test/
├── Indoor_xxx/
│ ├── rgb/
│ ├── depth/
│ │ ├── mesh_depth/ # depth maps (.exr)
│ │ ├── pointcloud_depth/ # depth maps (.exr)
│ │ ├── visual_mesh_depth/ # visualization (.jpg)
│ │ └── visual_pointcloud_depth/# visualization (.jpg)
│ ├── mask/
│ └── poses/
├── Indoor_xxx/
├── Outdoor_xxx/
│ ├── rgb/
│ ├── depth/
│ │ ├── mesh_depth/
│ │ ├── pointcloud_depth/
│ │ ├── visual_mesh_depth/
│ │ └── visual_pointcloud_depth/
│ ├── mask/
│ └── poses/
├── Outdoor_xxx/
└── ...
```
Notes:
- Timestamp-like file names are shared across modalities to support frame-level alignment.
## 💡 Dataset Download
Detailed download links and full-package release plan are **to be released**.
- [Hugging Face](https://huggingface.co/datasets/ouou123/Holo360D/tree/main)
## 📬 Contact
If you have any other questions, you can open an issue on GitHub or contact us via email at jou719@connect.hkust-gz.edu.cn.
## Citation
If you find this dataset useful, please cite our paper.
```bibtex
@article{ou2026holo360d,
title={Holo360D: A Large-Scale Real-World Dataset with Continuous Trajectories for Advancing Panoramic 3D Reconstruction and Beyond},
author={Ou, Jing and Cao, Zidong and Ren, Yinrui and Li, Zhuoxiao and Zhu, Jinjing and Hua, Tongyan and Zhang, Shuai and Xiong, Hui and Zhao, Wufan},
journal={arXiv preprint arXiv:2604.22482},
year={2026}
}
```
