Rules & Registration

Participation Overview

Eligibility

Open to researchers, students, and practitioners worldwide
Teams or individual participants allowed
Academic and industry participation welcomed
No participation fees required

Key Requirements

Registration through official challenge platform
Dataset license agreement
Submission of 2-page extended abstract
Reproducible code submission (Docker) for top selected participants

Critical Timeline Information

Review the challenge timeline carefully. Submissions must include both the super-resolution outputs AND the 2-page extended abstract by the final deadline.

Super-Resolution Approach

Video-Based Super-Resolution: No Restrictions

Each sequence contains 10 consecutive frames of the same license plate. Participants are free to use any approach—single-frame, multi-frame, or hybrid.

Single-Image SR (SISR)

Permitted ✓

Select any one frame from the 10-frame sequence
Apply standard single-image super-resolution
No requirement to use all frames
Frame selection strategy is part of your method

ŝ = 𝒢(I_{LR})

Multi-Image SR (MISR)

Permitted ✓

Use any subset or all 10 frames from the sequence
Leverage temporal information across frames
Frame fusion, alignment, or selection allowed
Any temporal aggregation strategy permitted

ŝ = 𝒢(I_{LR}^1, ..., I_{LR}^{10})

Method Freedom

There is no restriction on the number of frames used per sequence. Participants may:

Use exactly 1 frame (any frame of their choice)
Use all 10 frames
Use any subset between 2-9 frames
Dynamically select frames per sequence
Combine multiple frames into a single enhanced image

Evaluation:
Only the final predicted license plate text is scored. The method is free.

This rule applies to ALL phases: development, public validation, and blind test set evaluation.

Challenge Rules & Guidelines

Evaluation & Scoring

Primary Metric: Character Error Rate (CER) with custom scoring rules
Format Requirement: Predictions must match French license plate format: old license plates 123-AAA-12 or 1234-AA-12 (3 or 4 digits, 2 or 3 letters, 2 digits), and new license plate AA-123-AA (2 letters, 3 digits, 2 letters).
Scoring System:
- Correct character at correct position: +2 points
- No prediction for character: 0 points
- Wrong character (hallucination): -1 point
Maximum Score (new): 14 points (7 characters × 2 points)
Maximum Score (old): 16 points (8 characters × 2 points)
Minimum Score: -7 (new) or -8 (old) points

Submission Requirements

Final Submission: Must include both:
- Super-resolution outputs for all test sequences
- 2-page extended abstract in IEEE conference format
Output Format: CSV file with columns: sequence_id and predicted_lp
Prediction Format: Use underscore _ for missing characters. Must be exactly 7 or 8 characters long.
Abstract Format: PDF following IEEE ICIP 2026 template guidelines
Model Submission: Top-performing teams must submit reproducible Docker containers on invitation.

Prohibited Practices

Manual annotation or manipulation of test data
Using undisclosed external data
Multiple accounts or team splitting
Attempting to reverse-engineer the blind test set
Any form of cheating or unfair advantage

Permitted Practices

Use of any publicly available data for training
Pre-trained models and transfer learning
Data augmentation and synthetic data generation
Ensemble methods and model combination
External tools for preprocessing

Prediction Examples

Below is an actual low-resolution license plate from the development set. The examples show how different predictions are scored against the ground truth.

✓ Ground Truth: NH898KV 7 characters (new format)

Prediction	Visual Representation	Score	Explanation
`NH898KV`	N H 8 9 8 K V	14	Perfect prediction
`NH898KX`	N H 8 9 8 K X	12	Last character wrong (-1)
`NH898K_`	N H 8 9 8 K _	12	Last character missing (0)
`NH89___`	N H 8 9 _ _ _	8	First 4 correct, last 3 missing
`NH_____`	N H _ _ _ _ _	4	First 2 correct, rest missing
`_____KV`	_ _ _ _ _ K V	4	Last 2 correct, rest missing
`NHX98KV`	N H X 9 8 K V	11	Position 3 wrong (-1), rest correct
`NH_98KV`	N H _ 9 8 K V	12	Position 3 missing (0), rest correct
`NX_98KV`	N X _ 9 8 K V	9	Position 2 wrong (-1), position 3 missing (0)
`_______`	_ _ _ _ _ _ _	0	All missing
`1234567`	1 2 3 4 5 6 7	-7	All wrong (hallucinations)

Pro tip: It's better to use _ (0 points) than to guess a wrong character (-1 point)!

Submission format: Your prediction must be exactly 7 or 8 characters long. Use underscore _ for positions where your model has no prediction. Do not use spaces or leave blanks.

Recommended Public Datasets

Participants are encouraged to use additional data. Below are some publicly available license plate datasets suitable for pre‑training or data augmentation. Please respect each dataset's license terms and citation requirements.

UFPR-SR-Plates – GitHub
100k images with paired low/high-resolution sequences for super‑resolution. Real‑world surveillance conditions.
Realistic License Plate Restoration and Recognition (RLPR) – Mendeley Data
31‑frame low‑quality sequences paired with high‑quality targets for multi‑frame restoration.
Chinese City Parking Dataset (CCPD) – GitHub
Over 250k images with subsets for blur, tilt, weather, distance. Widely used for detection/recognition.

Ukrainian License Plate Dataset (Synthetic) – Zenodo
10k synthetic LP images generated with diffusion models.
European License Plates (ELP 1.0) – GitHub
Covers 20 European countries with diverse plate formats.

Some datasets may require registration or license acceptance.

2-Page Extended Abstract Requirements

The extended abstract is mandatory for all final submissions and must include:

Complete description of the proposed super-resolution method
Architecture details and training methodology
List of all external data used for training
Experimental results on the validation set
References to related work
Compliance with IEEE ICIP 2026 formatting guidelines

Registration Process

1

Platform Access

Receive credentials for CodaBench challenge platform and dataset download instructions.

2

Sign License Agreement

Accept the IMPROVED dataset terms of use and license agreement.

3

Complete Registration Form

Fill out the official registration form with team member details and contact information.

4

Begin Development

Access the development dataset and start building your super-resolution solution.

Ready to Participate?

Complete your registration to access the development dataset and join the XLPSR Challenge leaderboard.

Register Now View Dataset Details

Registration approval typically within 24-48 hours.
Questions? Review the contact page for support.

Registration Now Open