Open datasets are essential for advancing AI, but truly diverse real-world traffic data remains difficult to find.
To help address this gap, we built a Bangladesh Rickshaw Traffic Dataset designed for Physical AI, Computer Vision, Robotics, and Autonomous Driving research.
The dataset captures complex urban traffic scenes where rickshaws, motorcycles, buses, trucks, pedestrians, and private vehicles interact naturally in challenging environments.
Most publicly available traffic datasets are collected in structured road environments where traffic behavior is relatively predictable.
Bangladesh presents a very different scenario.
Urban roads are shared by rickshaws, motorcycles, buses, trucks, private vehicles, street vendors, bicycles, and pedestrians. These complex interactions make the dataset particularly valuable for training AI systems that must operate in diverse real-world environments rather than controlled conditions.
This diversity helps improve model robustness and generalization across different traffic scenarios.
Some key characteristics of the dataset include:
We believe metadata is just as important as the video itself.
Each clip is accompanied by structured metadata that helps researchers filter, search, and organize data efficiently.
Examples include:
This reduces preprocessing time and makes the dataset easier to integrate into machine learning workflows.
Many researchers and companies hesitate to evaluate datasets before purchasing them.
To make evaluation easier, we released a free sample through multiple platforms so users can inspect the data quality before requesting larger commercial datasets.
Our goal is transparency and long-term collaboration with the AI community.
This is only the beginning.
Upcoming releases will include:
We plan to continuously improve both dataset quality and the supporting data pipeline.
🌐 Website
💻 GitHub
🤗 Hugging Face
📊 Kaggle