More than 1.2 million car license plates posted. The dataset solves the problems of car detection, visibility classification and car license plate text recognition.

Photos of cars with license plates

The use of machine learning in the automotive industry
The process of identifying objects in photos for training a system to recognize and interpret them
Data annotation
Allocation of the bounding box object for solving computer vision problems
The ability to recognize and analyze images and video
Computer Vision
1 209 438
recognized license plates
896 338
unique car license plates
countries from all over the world
Dataset’s statistics
Percentage of photos (Labeled Images) by country:
Percentage of photos (Labeled Images) by country:
20% - Baltic States
17% - MENA Countries
12% - Central Asia
10% - Other Europe
6% - Caucasus Countries
5% - Asian Countries
Dataset's Features:
Photos of cars from two sources: YouTube and PlatesMania
Marking the Bounding Box for each captured car
Visibility assessment, coordinate marking and OCR marking for license plates

Metadata in the dataset
CSV files for each folder contain the following information about photos:
file_name - name of the original image - the country where the photo of the car was taken
bbox- Bounding Box marking for car license plate
license_plate.visibility — type of license plate visibility - unique id of the number markup
license_plate.mask - normalized coordinates of the corner points of the plate with a number in the format
license_plate. rows_count - single-line or two-line car number
license_plate.number - recognized license plate text
license_plate.serial - only for UAE - sign series
license_plate.region - UAE only - license plate sub-region
license_plate.color - only for Saudi Arabia - the color of the cell icon with the international number code
Tell us about your project!
See other datasets