mehmetkeremturkcan/traffic-lights-of-new-york

Traffic Lights of New York (TLoNY)

Traffic Lights of New York (TLoNY)

An object detection dataset for fine-grained classification of traffic lights in New York City. The dataset includes both vehicular and pedestrian traffic signals with detailed color state annotations.

Overview

TLoNY provides accurately labeled traffic light data suitable for training object detection models, sourced from publicly available images from Unsplash. The dataset captures various lighting conditions, weather scenarios, and urban environments of New York City.

Download

Dataset

• Huggingface

Models

• Huggingface

Dataset Format

• Format: YOLOv8 (ultralyics)
• Resolution: Various
• Images: RGB, JPG format
• Annotations: txt files in YOLO format

Classes

ID	Class	Description
0	red	Vehicular red light
1	green	Vehicular green light
2	yellow	Vehicular yellow light
3	red+yellow	Red and yellow combination
4	unknown	Light visible but color unclear
5	pedred	Pedestrian signal - red/stop
6	pedgreen	Pedestrian signal - green/walk
7	pedunknown	Pedestrian signal - color unclear

Directory Structure

tlony/
├── train/
│   ├── images/
│   └── labels/
├── val/
│   ├── images/
│   └── labels/
└── data.yaml

Acknowledgements

The creation of this dataset was supported by the ERC for Smart Street Scapes, funded through NSF award EEC-2133516.

Cite

Cite

@misc {mehmet_kerem_turkcan_2025,
    author       = { {Mehmet Kerem Turkcan} },
    title        = { traffic-lights-of-new-york (Revision 2999230) },
    year         = 2025,
    url          = { https://huggingface.co/datasets/mehmetkeremturkcan/traffic-lights-of-new-york },
    doi          = { 10.57967/hf/4331 },
    publisher    = { Hugging Face }
}