Car plate recognition project.

This project aims to detect car plate and number in cars. The dataset used can be found in roboflow: here.

The projects consists of:

• dataset download.
• model download.
• data exploration and visualization.
• model training.

Get started with the project

Get your roboflow key

Open the dataset link. then click Dataset in the dashboard. Choose V4 of the data. then click download with YOLOv9 format.

you will see something like this:

copy the key and paste it example.env file. then rename it to .env

dataset download

you can download the dataset py running this command:

./scripts/download_ds.sh

Model download

you can download the pretrained models py running this command:

./scripts/download_model.sh <MODEL_NAME>

available model names:

• YOLOv9t
• YOLOv9s
• YOLOv9m
• YOLOv9c
• YOLOv9s

data exploration and visualization

Explore the data by seeing examples from train, test, and validation splits by running this command:

python ./scripts/data_visualization.py

the resulting visualiztion will be in images folder under the name data_examples.png

Also view the count of each letter in the car plates dataset by running this command:

python ./scripts/data_explore.py

the resulting visualiztion will be in images folder under the name labels_count.png

Train the model

After downloading the dataset and the model you can start training the model.

You can specify the model name in training_args.yaml file to match the model that you have downloaded.

You can change any training argument or add training arguments in training_args.yaml.

Then to start training run this command:

python ./scripts/train.py

After training, the train logs and checkpoints will be in train_results directory.

based on the number of times you have run the training script, you will have directories train, train1, train2, ...

each directory contatins the logs, checkpoints and the best checkpoint during the training.

you can continue training from any checkpoint by specifying the model and train_dir in training_args.yaml.

then run this command:

python ./scripts/train --cont True