Datasets:

adadai3132
/

PanoHK360

Tasks:

Depth Estimation

Image-to-Image

Modalities:

Formats:

Languages:

Size:

Tags:

Libraries:

License:

Dataset card Data Studio Files Files and versions

xet

Community

Dataset Viewer

Auto-converted to Parquet Duplicate

Split (1)

train · 155k rows

Search is not available for this dataset

image image

End of preview. Expand in Data Studio

🏙️ PanoHK360

A Large-Scale 8K Urban Panoramic Dataset and Benchmark for Depth Estimation

City-scale 8K equirectangular RGB · LiDAR-derived metric depth · surface normals · raw point clouds · 6-DoF poses · continuous outdoor sequences

🎬 Dataset Teaser

▶ show.mp4 · Vehicle-mounted RGB–LiDAR matching and temporal geometric alignment along an urban Hong Kong trajectory.

The teaser above visualizes how continuously captured 8K panoramic frames are aligned, frame by frame, with the synchronized LiDAR sweeps that produce our metric depth annotations.

✨ Highlights

🛰️ Sensor-Grounded Metric Depth

Depth is measured, not predicted. Every depth pixel originates from a survey-grade LiDAR return projected into the 8K equirectangular frame — no synthetic rendering, no teacher-network pseudo labels.

🏙️ City-Scale Outdoor Coverage

Captured in dense urban Hong Kong with a vehicle-mounted multi-sensor rig. Continuous driving sequences yield diverse street canyons, intersections, façades, and complex occlusions.

🧭 Multi-Modal Annotations

Each frame ships with RGB panorama, metric depth, surface normals, point cloud, and 6-DoF camera pose, enabling depth, geometry, and reconstruction tasks from a single source.

🧪 Benchmark-Ready

The release is structured for plug-and-play training and evaluation of panoramic depth-estimation models, including ERP-aware architectures and 360° geometric perception methods.

📊 Dataset at a Glance

Field	Value
Dataset	PanoHK360
Scene type	Outdoor urban scenes
Location	Hong Kong
Image format	Equirectangular projection (ERP)
Resolution	8000 × 4000 px
Modalities	RGB panorama · metric depth · surface normals · LiDAR point cloud · 6-DoF pose
Depth source	Survey-grade LiDAR
Camera	Teledyne FLIR Ladybug panoramic camera
Release in this repository	`R101 20230413--filter`
License	CC BY 4.0

🛠️ Hardware & Acquisition Platform

The data are collected with a vehicle-mounted, time-synchronized rig combining a panoramic camera (RGB) and a survey-grade LiDAR scanner (geometry). Synchronization between the two sensors is what enables direct, geometry-faithful RGB ↔ depth correspondence.

📷 Teledyne FLIR Ladybug
_{360° panoramic imaging}

📡 RIEGL VUXR-1HA22
_{Survey-grade LiDAR scanning}

📁 Repository Layout

PanoHK360/
├── README.md                       # This file
├── R101 20230413--filter/          # 📂 Browsable filtered data release
├── R101 20230413--filter.zip       # 📦 Same release, compressed (2.77 GB)
├── show.mp4                        # 🎬 RGB–LiDAR alignment demo video
├── ladybug_image.png               # 📷 Panoramic camera reference image
├── RIEGL VUXR-1HA22.png            # 📡 LiDAR scanner reference image
└── .gitattributes

Note: R101 20230413--filter/ (folder) and R101 20230413--filter.zip (archive) contain the same filtered release. Browse the folder online, or download the zip for local experiments.

📦 Data Access

Option 1 — Browse online

Open the folder directly in this repository:

R101 20230413--filter/

Option 2 — Download the compressed package

# After cloning / downloading the repository:
unzip "R101 20230413--filter.zip" -d PanoHK360_release
ls PanoHK360_release

If you use Git LFS or Xet-backed storage, make sure the large files are fully pulled before running experiments.

🚀 Quick Start

# 1. Clone via your preferred Hugging Face / Git workflow
#    (see the "Use this dataset" button at the top of the page)

# 2. Extract the filtered release
unzip "R101 20230413--filter.zip" -d PanoHK360_release

# 3. Inspect the frame-level files and plug into your data loader
ls PanoHK360_release

📄 File Descriptions

File / Folder	Description
`R101 20230413--filter/`	Extracted filtered data release — frame-level RGB, depth, normals, poses, and point clouds
`R101 20230413--filter.zip`	Compressed version of the same filtered release (≈ 2.77 GB)
`show.mp4`	Demo video showing vehicle-driving RGB–LiDAR matching and temporal alignment
`ladybug_image.png`	Reference image of the Teledyne FLIR Ladybug panoramic camera
`RIEGL VUXR-1HA22.png`	Reference image of the RIEGL VUXR-1HA22 LiDAR scanner

📚 Citation

If you find PanoHK360 useful in your research, please cite the paper once the citation information is available:

@misc{panohk360,
  title        = {PanoHK360: A Large-Scale 8K Urban Panoramic Dataset and Benchmark for Depth Estimation},
  author       = {Anonymous Authors},
  year         = {2026},
  note         = {Under review}
}

📜 License

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Please provide proper attribution when using the dataset, figures, or derived annotations.

_{— Built for the panoramic 3D perception community —}

Downloads last month: 26,972