# Dur360BEV: A Real-world 360-degree Single Camera Dataset and Benchmark for Bird-Eye View Mapping in Autonomous Driving

## Abstract  
We present **Dur360BEV**, a novel spherical camera autonomous driving dataset featuring a **128-channel 3D LiDAR** and an **RTK-refined GNSS/INS system**, along with a benchmark for **Bird's Eye View (BEV) map generation** using a **single spherical camera**. This dataset addresses the challenges of BEV generation by **reducing hardware complexity**, replacing multiple perspective cameras with a **single 360-degree camera**.  

Our benchmark includes **SI2BEV (Spherical-Image-to-BEV)**, a novel module that projects spherical imagery into BEV space with a **refined sampling strategy**, as well as an **adaptive Focal Loss** formulation to handle **extreme class imbalance** in BEV segmentation. Extensive experiments demonstrate that **Dur360BEV** simplifies the sensor setup while achieving **competitive performance**.60BEV: A Real-world 360-degree Single Camera Dataset and Benchmark for Bird-Eye View Mapping in Autonomous Driving

## Sensor placement

Dur360BEV is equipped with a **spherical camera, 3D LiDAR, and a high-precision GNSS/INS system**, providing comprehensive environmental perception.

- **LiDAR**: Ouster OS1-128 LiDAR sensor with 128 channels vertical resolution
- **Spherical Camera**: Ricoh Theta S featuring a **dual 1/2.3" 12M CMOS sensor**, **1280 x 640** resolution, and **15 Hz capture frequency**
- **GNSS/INS**: OxTS RT3000v3 global navigation satellite and inertial navigation system, supporting localization from GPS, GLONASS, BeiDou, Galileo, PPP and SBAS constellations


## Dataset Version Description

- **Dur360BEV**: the train and validation split used in our paper.
- **Dur360BEV-Extended**: an extension of the original Dur360BEV dataset, providing an additional 15,400 frames of synchronized data captured with the same sensor setup.
- **Dur360BEV-Complete**: the combination of **Dur360BEV** and **Dur360BEV-Extended**. The provided scripts will automatically download both subsets and organize the dataset structure for seamless use.

## Dataset File Description

dataset/ 
|-- image
|   |-- data
|   +-- timestamps.txt
|-- labels
|   |-- data
|   |-- dataformat.txt
|   +-- timestamps.txt
|-- md5sums.txt
|-- metadata
|   +-- os1.json
|-- ouster_points
|   |-- data
|   |-- dataformat.txt
|   +-- timestamps.txt
+-- oxts
    |-- data
    |-- dataformat.txt
    +-- timestamps.txt

## Citation

If you use this dataset, please cite our paper:

W. E, C. Yuan, Y. Sun, Y. F. A. Gaus, A. Atapour-Abarghouei, and T. P. Breckon, **"Dur360BEV: A Real-world 360-degree Single Camera Dataset and Benchmark for Bird-Eye View Mapping in Autonomous Driving,"** in *Proceedings of the IEEE International Conference on Robotics and Automation (ICRA)*, 2025.