open-image-models 0.1.0

Pre-trained image models using ONNX for fast, out-of-the-box inference.

Open Image Models

• Open Image Models
  • Introduction
  • Features
  • Available Models
    • Object Detection
      • Plate Detection
    • Installation
    • Contributing

---

Introduction

We offer ready-to-use models for a range of computer vision tasks like detection, classification, and more. With ONNX support, you get fast and accurate results right out of the box.

Easily integrate these models into your apps for real-time processing—ideal for edge devices, cloud setups, or production environments. In one line of code, you can have powerful model inference running!

Features

• 🚀 Pre-trained Models: Models are ready for immediate use, no additional training required.
• 🌟 ONNX Format: Cross-platform support for fast inference on both CPU and GPU environments.
• ⚡ High Performance: Optimized for both speed and accuracy, ensuring efficient real-time applications.
• 📏 Variety of Image Sizes: Models available with different input sizes, allowing flexibility based on the task's performance and speed requirements.
• 💻 Simple API: Achieve license plate detection with just one line of code, enabling rapid integration and deployment.

Available Models

Object Detection

Plate Detection

Model	Image Size	Precision (P)	Recall (R)	mAP50	mAP50-95	Speed (ms)
yolov9-t	640	0.955	0.91	0.959	0.75	XXX
yolov9-t	512	0.948	0.901	0.95	0.718	XXX
yolov9-t	384	0.943	0.863	0.921	0.688	XXX
yolov9-t	256	0.937	0.797	0.858	0.606	XXX

^[1] Inference on Mac M1 chip using CPUExecutionProvider. Utilizing CoreMLExecutionProvider accelerates speed by 5x.

Usage

import cv2
from rich import print

from open_image_models import LicensePlateDetector

# Initialize the License Plate Detector with the pre-trained YOLOv9 model
lp_detector = LicensePlateDetector(detection_model="yolo-v9-t-384-license-plate-end2end")

# Load an image
image_path = "path/to/license_plate_image.jpg"
image = cv2.imread(image_path)

# Perform license plate detection
detections = lp_detector.predict(image)
print(detections)

# Benchmark the model performance
lp_detector.show_benchmark(num_runs=1000)

# Display predictions on the image
annotated_image = lp_detector.display_predictions(image)

# Show the annotated image
cv2.imshow("Annotated Image", annotated_image)
cv2.waitKey(0)
cv2.destroyAllWindows()

Installation

To install open-image-models via pip, use the following command:

pip install open-image-models

Contributing

Contributions to the repo are greatly appreciated. Whether it's bug fixes, feature enhancements, or new models, your contributions are warmly welcomed.

To start contributing or to begin development, you can follow these steps:

1. Clone repo

   git clone https://github.com/ankandrew/open-image-models.git
   

2. Install all dependencies using Poetry:

   poetry install --all-extras
   

3. To ensure your changes pass linting and tests before submitting a PR:

   make checks