melika0m/RIMIAChallenge

my contribution in the rimia competition to improve the automation vehicle license plate in mauritania License Plate Recognition with Ensemble of Deep Learning Models This repository contains an end-to-end pipeline for recognizing license plate numbers using an ensemble of deep learning models (ResNet50, InceptionV3, and Xception). The model is trained with K-Fold cross-validation and utilizes data augmentation and a custom weighted loss function to improve accuracy. Table of Contents Overview Installation Data Preparation Training Inference Submission References Overview This project implements an ensemble of models to recognize license plates using a series of deep learning models with K-Fold cross-validation. The core components include: Models: ResNet50, InceptionV3, and Xception (from TensorFlow/Keras) K-Fold Cross Validation: Ensures robust training across multiple folds. Ensemble Learning: Averages predictions from the models to enhance overall accuracy. YOLOv5 (optional): Can be used for license plate detection. Installation Requirements Python 3.12.2 or later TensorFlow OpenCV NumPy Pandas Scikit-learn Steps Clone the repository: bash git clone cd RimIA Set up a virtual environment: python -m venv rim-ai-env .\rim-ai-env\Scripts\activate # On Windows source rim-ai-env/bin/activate # On Linux or MacOS Install dependencies: pip install numpy pandas tensorflow opencv-python matplotlib scikit-learn pillow Verify installation: python --version pip list # To verify that all dependencies are installed Data Preparation Before training the model, ensure that your dataset is organized as follows: bash Copier le code dataset/ │ ├── images/ │ ├── train/ │ └── test/ │ ├── │ train_labels.csv │ ├── submission_template.csv Train images: Place the training images inside images/train/. Test images: Place the test images inside images/test/. Labels: Ensure that the training labels are stored in labels/train_labels.csv with two columns: img_id: Unique identifier for each image. plate_number: The corresponding license plate number. Submission template: Ensure the …