Dual-modal Roadside Traffic Dataset

# Overview

The Dual-modal Roadside Traffic Dataset (DRTD) is a  dataset designed for roadside traffic object detection using synchronized RGB images and event-based vision data.

The dataset enables research on RGB-event multimodal fusion, event-based perception, and robust traffic scene understanding.

The dataset is associated with the data descriptor paper:

**"A Spatially Aligned RGB-Event Modality Dataset for Roadside Traffic Object Detection"**, under review at *Scientific Data*.

---

## Data Acquisition

The dataset was collected using a roadside sensing system equipped with an RGB camera and an event camera.

The two sensors were spatially calibrated and temporally synchronized. RGB images capture appearance information,

while event data record asynchronous brightness changes with high temporal resolution, enabling robust perception under challenging lighting and motion conditions.

Raw event data are stored as asynchronous event streams with fields `(x, y, p, t)`, representing pixel coordinates, event polarity, and timestamp, respectively.

---

## Data Organization

The dataset is organized into aligned and raw data components.

The aligned data include spatially and temporally synchronized RGB images, event-based representations, and unified object detection annotations.

The raw data include unprocessed RGB images and raw event streams, allowing users to implement custom preprocessing or alignment methods.

The aligned data are split into training, validation, and test subsets.

RGB images, event representations, and annotation files share a unified naming convention to ensure one-to-one correspondence across modalities.

Due to the large number of files, the dataset is distributed as multiple compressed archives.

Each archive preserves the original directory structure described in the accompanying README file.

---

## Annotations

Object annotations are provided in the YOLO format. Each annotation file corresponds to one RGB image and its aligned event representation.

Bounding box coordinates are normalized with respect to image dimensions.

The dataset includes four object categories:

- Car

- Motorcycle

- Bus

- Truck

Detailed label definitions and category mappings are provided in the `README.md`.

---

## Potential Applications

The DRTD dataset supports a wide range of research tasks, including:

- RGB-only and event-only traffic object detection

- Multimodal RGB-event fusion for object detection

- Event-based representation learning

- Robust traffic perception under challenging illumination and motion conditions

---

## Availability and Usage

The dataset is hosted on Zenodo and provided for research and educational purposes.

Example code for data loading, event processing, multimodal fusion, and benchmark training is available in a public GitHub repository:

**GitHub:** https://github.com/myzhuang/DRTD-demo

Users are encouraged to follow the provided train/validation/test split and evaluation protocols for fair comparison and reproducibility.

---

## MD5sum Code for Main zip file

The MD5sum code of the main file is provided to verify the integrity of the file.

aligned.zip md5:ca639fb5f72be5a7f7b48ec76f908b35

calibration.zip md5: 9b78e423a09e0bff0570f4929ee7522f

raw_RGB.zip md5: c3537ddf1ab414f9cd020b728a51abc8

raw_event.zip md5:74e0d9d0de25c5f8b49f2fe6180d8122

---

## License

This dataset is released under the **Creative Commons Attribution 4.0 International (CC BY 4.0)** license.  

See `LICENSE.txt` for details.