Car plates project. Segmentation model (part 1/3)

This is the project for car plate OCR recognition, which include:

1. Neural network segmentation model for car plate area with number selection (part 1/3)
2. Neural network OCR model for plate character recognition (part 2/3)
3. API service for these two models (part 3/3)
4. Additional exemple how to use API service in Telegram bot

Dataset

Dataset include 1754 images of car with plates in jpg from 4 countries (include Russia) in COCO format.

I use several open data from kaggle to compile single dataset:

• Automatic-Number-Plate-Recognition
• Car-and-License-Plate-Detection
• Car-License-Plate-Detection
• Car-plate-object-detection

On each image was selected box with car plate and for some data was selected box with car (plate and car classes).

Notes:

• For some data only clearly visible car plates were selected.
• Example of data you can see in notebook.
• Model inference and export you can see in notebook

To download data:

make download_dataset

Environment setup

1. Create and activate python venv

   python3 -m venv venv
   . venv/bin/activate

2. Install libraries

    make install

3. Run linters

   make lint

4. Tune config.yaml

5. Train

   make train

Additional information

• Inference example in notebook
• Best experiment in ClearML
• History of experiments