Python Image Detection (Kystverket AI summer interns 2023)

Template for using Yolo PyTorch models for image detection.

Table of Contents

• About
• Getting Started
• Using a Pre-Trained Model
• Making, Training, and Using a Custom Data Set
  • Making a Custom Dataset
  • Training a Model Using a Custom Dataset
  • Exporting and Using a Model
• Documentation
  • Documentation YOLOv8
  • Documentation Detector.py

About

This project aims to simplify the process of image detection in YOLOv8, enabling more people to label, train, and use an image detection model.

Getting Started

1. Clone or download this directory.
2. Navigate to this directory in the command prompt.
3. Run pip install -r requirements.txt (this installs all the Python packages required).
4. Try the demo.ipynb notebook to get an understanding of image detection.

Using a Pre-Trained Model

If you have a pre-trained model (.pt file), you can pass in the relative path to the model when initializing a YOLO or a Detector object. An example of this is found in the demo.ipynb file.

Making, Training, and Using a Custom Data Set

Making a Custom Dataset

Label your dataset by "drawing" boxes around the objects you want to detect and classifying them into categories ("Excavator", "Lighthouse", etc.). Separate the images into training, validation, and test sets. We recommend using Roboflow. Our own Gravemaskin dataset is available here: Roboflow - Gravemaskin Dataset.

Training a Model Using a Custom Dataset

Train the model using a powerful GPU. In Google Colab, you can take advantage of their built-in GPU. A full tutorial is available here.

Exporting and Using a Model

After training, you obtain a PyTorch file (.pt). You can run your model as with any pre-trained models. See the tutorial under "Exporting the model".

Documentation

Documentation YOLOv8

• Refer to the Ultralytics YOLO documentation.
• Object Detection 101
• Project Report

Documentation Detector.py

Detector.py simplifies the YOLO methods in Ultralytics, making them easier to use but with some limitations. For detailed analysis, use the YOLO methods directly (as shown in the end of demo.ipynb).

• Detector(model_path): Initializes a detection object with a trained model in .pt PyTorch format.
• find_objects_image(image, conf, show_all, save_image, save_filename): Detects objects in an image, with customizable parameters.