https://github.com/tristan296/deepplate
A custom-trained YOLOv11-based license plate detection with PaddleOCR for accurate ANPR.
https://github.com/tristan296/deepplate
cv2 numpy paddleocr sqlite3 ultralytics yolov11
Last synced: 6 months ago
JSON representation
A custom-trained YOLOv11-based license plate detection with PaddleOCR for accurate ANPR.
- Host: GitHub
- URL: https://github.com/tristan296/deepplate
- Owner: Tristan296
- Created: 2025-04-02T08:59:50.000Z (6 months ago)
- Default Branch: main
- Last Pushed: 2025-04-02T09:52:14.000Z (6 months ago)
- Last Synced: 2025-04-02T10:35:14.856Z (6 months ago)
- Topics: cv2, numpy, paddleocr, sqlite3, ultralytics, yolov11
- Language: Python
- Homepage:
- Size: 17.8 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Image Detection
> A project that utilizes machine learning to detect and classify objects in images.
> Live demo [_here_](https://www.example.com).
## Table of Contents
* [General Info](#general-information)
* [Technologies Used](#technologies-used)
* [Features](#features)
* [Screenshots](#screenshots)
* [Setup](#setup)
* [Usage](#usage)
* [Project Status](#project-status)
* [Model Training Summary](#model-training-summary)
* [Room for Improvement](#room-for-improvement)
* [Acknowledgements](#acknowledgements)
* [Contact](#contact)
## General Information
- This project aims to simplify the process of detecting and classifying objects in images using advanced machine learning techniques.
- It solves the problem of manual image classification by automating the process.
- The purpose of this project is to provide an easy-to-use tool for developers and researchers working with image data.
## Technologies Used
- Python - version 3.9
- TensorFlow - version 2.10
- OpenCV - version 4.5
## Features
List the ready features here:
- Object detection in real-time
- Classification of detected objects
- Support for multiple image formats
## Screenshots
## Setup
- Install the required dependencies listed in `requirements.txt`.
- Clone the repository and navigate to the project directory.
- Run the following command to install dependencies:
```
pip install -r requirements.txt
```
## Usage
- To detect objects in an image, run the following command:
```
python detect.py --image path/to/image.jpg
```
- For real-time detection using a webcam:
```
python detect.py --webcam
```
## Project Status
Project is: _in progress_. Further improvements and optimizations are being worked on.
## Room for Improvement
Room for improvement:
- Enhance the accuracy of object detection.
- Add support for additional machine learning models.
To do:
- Implement a user-friendly GUI.
- Add functionality for batch image processing.
# Model Training Summary
## General Information
The model was trained using the YOLO framework with the following configuration:
- **Task**: Object Detection
- **Model**: YOLO (yolo11n.pt)
- **Dataset**: COCO8
- **Epochs**: 100
- **Batch Size**: 16
- **Image Size**: 640x640
- **Device**: MPS (Metal Performance Shaders)
## Training Results
The training process generated the following key results:
- **Final Model**: Saved as `last.pt` in the `weights/` directory.
- **Metrics**:
- **Precision**: Improved steadily, reaching a maximum of ~0.00367.
- **Recall**: Peaked at 0.66667 during early epochs.
- **mAP@50**: Reached a maximum of 0.28376.
- **mAP@50-95**: Reached a maximum of 0.13616.
## Loss Trends
- **Box Loss**: Decreased significantly over epochs, indicating better localization.
- **Classification Loss**: Reduced steadily, showing improved classification accuracy.
- **DFL Loss**: Gradually decreased, reflecting better distribution-focused learning.
## Visualizations
- **Confusion Matrices**: `confusion_matrix.png` and `confusion_matrix_normalized.png` provide insights into class-wise performance.
- **Curves**:
- `F1_curve.png`: Shows the F1-score progression.
- `P_curve.png`: Precision curve.
- `R_curve.png`: Recall curve.
- `PR_curve.png`: Precision-Recall curve.
## Training Batches
- Sample training images are available (`train_batch0.jpg`, `train_batch1.jpg`, etc.), showing the model's predictions during training.
## Validation Results
- Validation images (`val_batch0_labels.jpg`, `val_batch0_pred.jpg`) demonstrate the model's performance on unseen data.
## Observations
- The model shows steady improvement in metrics over epochs.
- The mAP values suggest room for improvement in detection accuracy.
- The training process was well-documented with visualizations and logs.
## Next Steps
- Fine-tune the model to improve mAP and precision.
- Experiment with different hyperparameters or augmentations.
- Test the model on real-world data to evaluate its generalization.
## Acknowledgements
### YOLO by Ultralytics
```bibtex
@misc{yolo2023,
author = {Jocher, Glenn and Chaurasia, Ayush and Qiu, Jing},
title = {YOLO by Ultralytics},
year = {2023},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/ultralytics/yolov5}}
}
```
### PaddleOCR
```bibtex
@article{paddleocr2021,
author = {PaddleOCR Contributors},
title = {PaddleOCR: An Open-Source Optical Character Recognition Tool Based on PaddlePaddle},
year = {2021},
journal = {GitHub repository},
howpublished = {\url{https://github.com/PaddlePaddle/PaddleOCR}}
}
```
## Contact
Created by [@tristan](https://github.com/tristan296) - feel free to contact me!