YOLOv8-Based Cell Detection 🦠

This project focuses on the development of a cell detection model using YOLOv8, a state-of-the-art object detection architecture. The model is designed to accurately identify and localize cells within microscopy images.

Key Features:

• CVAT for Annotation: The project utilizes CVAT (Computer Vision Annotation Tool), a powerful open-source platform, for generating high-quality annotations of cell locations within the images.
• YOLOv8 Framework: Leverages the speed and accuracy of the YOLOv8 model for real-time or high-throughput cell detection.
• Customizable: The model can be easily adapted to different cell types and microscopy settings.
• Python Implementation: Training, validation, and inference are implemented in Python using the Ultralytics YOLOv8 library.

Dataset:

• Image Source: The dataset comprises microscopy images of cells.
• Annotation Format: Images are annotated in CVAT, then exported to YOLO format for model training.
• Dataset Structure:
  • cell_data/images/train/: Contains training images.
  • cell_data/images/test/: Contains test images (used for evaluation only).
  • cell_data/labels/train/: Contains YOLO format annotations for training images.
  • config.yaml: Defines dataset paths, class names, and augmentation settings.

Usage:

1. Installation:
   • Ensure you have Python installed.
   • Install the required libraries:

     pip install ultralytics opencv-python

2. Training: Run model_training.py to train the YOLOv8 model using the config.yaml file and the provided training data. The trained model will be saved in the runs/ directory.
3. Inference: Run model_testing.py to load the trained model and perform inference on images in the cell_data/images/test directory.
   • The script will display the images with predicted bounding boxes and confidence scores.
   • Press 'q' to move to the next image.
   • Feel free to add your own images to images/test to evaluate model performance.

Customisation:

• Adjust the config.yaml file to modify dataset paths, class names, and augmentation parameters.
• Explore different YOLOv8 model architectures (e.g., yolov8s, yolov8m) for varying levels of speed and accuracy.

Future Work:

• Incorporate additional cell types or features for detection.
• Experiment with different model architectures and hyperparameters to improve performance.
• Integrate the model into a larger analysis pipeline for automated cell counting and tracking.

Contributing:

Contributions are welcome! Feel free to open issues, submit pull requests, or provide feedback to help improve this project.