Overview

License plate recognition plays a vital role in modern intelligent transportation systems, powering applications such as automated tolling, parking management, traffic monitoring, and law enforcement. Automatically detecting and reading license plates from vehicle images enables faster identification, enhances public safety, and supports efficient traffic operations.

This challenge invites participants to build end-to-end computer vision models capable of detecting license plates and recognizing their alphanumeric content under real-world conditions. The task requires handling diverse environments, including variations in lighting, orientation, distance, and plate styles.

The goal is to develop accurate, robust, and scalable solutions for smart-city infrastructure, autonomous systems, and large-scale traffic analytics. Successful models must both locate license plates correctly and extract their text content with high precision, combining object detection and optical character recognition (OCR).

Skills You'll Practice

• Python
• Deep learning
• Object detection
• Optical Character Recognition (OCR)

Evaluation

Goal

Your objective is to:

1. Detect license plates in the test images.
2. Recognize the plate characters accurately.

Metrics

The final evaluation considers both detection and recognition performance:

• Detection: Mean Average Precision (mAP) at IoU = 0.5

• Recognition: Character Error Rate (CER)

• Final Score: A weighted combination of both metrics:

  Final Score = 0.5 × mAP + 0.5 × (1 − CER)
  

Submission

Format

Submit a CSV file with two columns:

• id: The unique identifier (image filename) for each image in the test set.
• predict_label: A string describing the predicted result(s) for that image.

Each prediction follows the format:

class confidence_score x_center y_center box_width box_height license_number

Field	Description
class	Object category (1 = license plate).
confidence_score	Confidence of the prediction (range 0-1).
x_center, y_center	Normalized coordinates of the bounding-box center.
box_width, box_height	Normalized width and height of the bounding box.
license_number	The predicted license plate number.

If an image contains multiple plates, concatenate the predictions in the same cell, separated by a single space.

Note: x_center and box_width are normalized by image_width; y_center and box_height are normalized by image_height.

Example:

id	predict_label
0001.jpg	1 0.927 0.191 0.808 0.208 0.103 AI48004
0002.jpg	1 0.927 0.191 0.808 0.208 0.103 BM74043 1 0.555 0.123 0.138 0.338 0.443 HL14623

See the submission guide for full instructions.

Tracks

The challenge accepts two submission tracks:

• Public: predict on the public test set and upload the resulting CSV.
• Private:: upload your model, which is scored on a held-out private test set.

Challenge Timeline

Phase	Duration	Start Date	End Date	Description
Registration	2 weeks	2026-06-15	2026-06-28	Participant registration period.
Training Phase	2 weeks	2026-06-29	2026-07-12	50% of the dataset released for training and public submission.
Testing Phase	2 weeks	2026-07-13	2026-07-26	Remaining 50% of the dataset released for training and public submission.
Final Submission	1 week	2026-07-27	2026-08-02	Private submission phase for the final model.
Judging Phase	2 weeks	2026-08-03	2026-08-16	AIOZ AI evaluates final private submissions, verifies model reproducibility, and finalizes leaderboard rankings.
Result Announcement	1 day	2026-08-17	2026-08-17	Final results and prize winners announced.

*Timeline in ET

Prizes

Total Challenge Rewards: 4,500 AIOZ tokens

• Top 1: 2,500 AIOZ tokens
• Top 2: 1,500 AIOZ tokens
• Top 3: 500 AIOZ tokens

Changelog

Public Dataset

• ⏳ Phase 1 (Training Phase): 2026-06-29
• ⏳ Phase 2 (Testing Phase): 2026-07-13

Final Submission

• ⏳ Upload your private solution: 2026-07-27