metadata

license: cc-by-nc-nd-4.0
task_categories:
  - image-classification
  - object-detection
language:
  - fr
  - en
tags:
  - quebec
  - road
  - sign
pretty_name: Quebec Traffic Signs
size_categories:
  - 1K<n<10K

Quebec Traffic Signs Dataset

Dataset Description

The Quebec Traffic Signs Dataset is a specialized image dataset designed for the recognition and interpretation of road and traffic signs specific to the province of Quebec, Canada. This dataset aims to provide a comprehensive collection of signs, including regulatory, warning, and informational signs, with a particular focus on the unique bilingual (French/English) and complex parking/construction signage prevalent in Quebec, especially in urban areas like Montreal.

This dataset is being developed to support the creation of intelligent mobile applications that can assist drivers in understanding local traffic regulations, particularly for parking restrictions and construction zone alerts, which are often a source of confusion.

Dataset Structure

The dataset is organized as follows:

quebec-traffic-signs/
├── images/
│   ├── P-010-fr.png
│   ├── P-010-en.png
│   └── ... (all image files)
├── metadata.jsonl
└── README.md

images/: This directory contains all the image files of the traffic signs. These include both pristine digital assets (SVG/PNG renders) and will be augmented with real-world photos captured under various challenging conditions (e.g., blur, low quality, occlusions, snow, vandalism).
metadata.jsonl: This JSON Lines file serves as the primary metadata index for the dataset. Each line is a JSON object corresponding to a unique traffic sign image and contains the following fields (aligned with the D1 database schema):
- image_id: A unique identifier for the image.
- file_name: The relative path to the image file within the images/ directory (e.g., images/P-010-fr.png).
- source: Indicates the origin of the image (e.g., digital_asset, real_world_photo, synthetic_diffusion).
- reference_id: A unique identifier for the sign, often corresponding to official Quebec sign codes (e.g., P-010-fr, P-120-10 (left)).
- explanation_fr: A detailed explanation of the sign's meaning in French.
- explanation_en: A detailed explanation of the sign's meaning in English.
- real_world_conditions: A JSON array describing any real-world conditions simulated or observed (e.g., ["snow_occlusion", "blur"]).
- is_synthetic: A boolean indicating if the image is synthetically generated.
- original_url: The original source URL from which the digital asset was downloaded (primarily Wikimedia Commons).

Data Collection and Annotation

The initial phase of data collection focuses on digital assets sourced from Wikimedia Commons, which provides high-quality SVG renders of official Quebec road signs. These digital assets are processed to extract reference_id and original_url, and then downloaded locally.

Annotations, including detailed explanations (French and English) and structured metadata (e.g., source, real_world_conditions, is_synthetic), are compiled into the metadata.jsonl file.

Future iterations will incorporate:

Real-world photos: Images captured in urban environments (e.g., Montreal) under diverse conditions (varying lighting, weather, angles, partial obstructions, wear and tear, graffiti) to reflect real-world challenges.
Synthetic data generation: Leveraging multimodal diffusion models (e.g., Stable Diffusion with ControlNet) to augment the dataset with synthetically generated images that simulate challenging real-world conditions, based on the pristine digital assets. This will significantly increase the dataset's robustness for edge cases.

Usage

This dataset is designed to be easily loaded and used with the Hugging Face datasets library.

from datasets import load_dataset

# To load the dataset (once it's publicly available on the Hub)
dataset = load_dataset("RDLTechworks/quebec-traffic-signs")

# Example of accessing data
# Accessing the first example in the 'train' split
first_example = dataset["train"][0]

print(f"Image ID: {first_example['image_id']}")
print(f"Reference ID: {first_example['reference_id']}")
print(f"French Explanation: {first_example['explanation_fr']}")
print(f"English Explanation: {first_example['explanation_en']}")
print(f"Source: {first_example['source']}")
print(f"Real-world Conditions: {first_example['real_world_conditions']}")
print(f"Is Synthetic: {first_example['is_synthetic']}")
print(f"Original URL: {first_example['original_url']}")

# To display the image (requires Pillow installed: pip install Pillow)
# from PIL import Image
# image = first_example['image'] # This will be a PIL Image object
# image.show()

Citation Information

If you are using this dataset, please cite:

@misc{mosaïque_2025,
     author       = { mosaïque and Darleison Rodrigues and Rayane Nascimento },
    title        = { quebec-traffic-signs (Revision 07519ff) },
    year         = 2025,
    url          = { https://huggingface.co/datasets/RDLTechworks/quebec-traffic-signs },
    doi          = { 10.57967/hf/6044 },
    publisher    = { Hugging Face }
}

License

This dataset is intended for research and non-commercial use. The images sourced from Wikimedia Commons are typically under Creative Commons licenses (e.g., CC BY-SA), and users should verify the specific license for each image if commercial use is intended. The metadata and annotations created for this dataset will be released under a permissive license (e.g., CC BY-SA 4.0).

Contact

For questions or contributions, please contact [[email protected]].