pc4653/JHU-ULTRA-360

JHU-ULTRA-360

📖 Overview

JHU-ULTRA-360 is collected at Johns Hopkins University's Homewood campus and aims to reconstruct and visualize the entire campus in 4D, with collected across the span of two years, where the video frames are calibrated with manual inspection and aligned to a consistent coordinate system. JHU-ULTRA-360 provides:

1. Immersive Ground Acquisition: Both perspective and 360 panorama images on the ground level are collected and calibrated to facilitate immersive 3D reconstruction.

2. Multi-Elevation Acquisition: Both ground and aerial images from multiple elevations are collected and calibrated to ensure full coverage of the buildings.

3. Multi-Seasonality Acquisition: Images are acquired across multiple seasons in a two-year period, capturing the gradual changes on campus over time.

4. Large-Scale Calibration: Twelve academic buildings are collected across multiple elevations, camera models, and years, and are calibrated together.

Each building folder includes:

• Raw RGB images organized by acquisition type (e.g., aerial, panorama, transition, etc.)
• COLMAP-generated SfM ground-truth (cameras.bin, images.bin, points3D.bin)

---

🗂️ Dataset Structure

 
JHU-ULTRA-360/ 
    ├── 10_AMES/ 
    │ ├── aerial/ 
    │ │ ├── ac/
    │ │ │ ├── 10-ac-000001.jpg
    │ │ │ ├── 10-ac-000002.jpg
    │ │ │ ├── ... │
    │ │ ├── ah/
    │ │ ├── ... │
    │ ├── panorama/ 
    │ │ ├── ps/
    │ │ ├── pw/
    │ ├── perspective/ 
    │ ├── transition/ 
    │ ├── sparse/ 
    │ │ ├──0/
    │ │ │ ├── cameras.bin
    │ │ │ ├── images.bin
    │ │ │ ├── points3D.bin
    ├── 24_Clark/
    ├── 28_Garland/  
    ├── .../ |
 

• Image folders contain high-resolution RGB aerial, panorama, perspective and transition images collected across different seasons and at varying times of day.
• Registration folders contain ground truth camera positions aligned at various elevations.

---

🖹 Reconstruction Split Files

This zip file 'Recon_exp_splits.zip' contains the train/test split files used in the dense reconstruction experiment. If you want to reproduce the results in the paper, use the image names provided in these files 😊

The directory of the folder looks like:

 
Recon_exp_splits/ 
    ├── multi_appearance_split/ 
    │ ├── 10_AMES/
    │ │ ├── test_split.txt
    │ │ ├── train_split.txt
    │ ├── 24_Clark/
    │ ├── ... │
    ├── multi_elevation_split/ 
    │ ├── 10_AMES/
    │ │ ├── test_split.txt
    │ │ ├── train_split.txt
    │ ├── 24_Clark/
    │ ├── ... │
 

---

Please cite:

@misc{liu2024immersivemultielevationmultiseasonaldataset, title={An Immersive Multi-Elevation Multi-Seasonal Dataset for 3D Reconstruction and Visualization}, author={Xijun Liu and Yifan Zhou and Yuxiang Guo and Rama Chellappa and Cheng Peng}, year={2024}, eprint={2412.14418}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2412.14418}, }

🔽 How to Download

You can download the dataset using the Hugging Face Client:

huggingface-cli download pc4653/JHU-ULTRA-360 \
--repo-type dataset \
--local-dir JHU-ULTRA-360 \
--local-dir-use-symlinks False
cd JHU-ULTRA-360
unzip JHU-ULTRA-360.zip