Sperm Cell Detection with YOLOv8

Computer vision project for automated sperm cell detection and counting using YOLOv8.

Project Overview

This project uses YOLOv8 to detect and count sperm cells in microscopic images. It can process individual images, batches of images, and videos.

What I Built

• Trained YOLOv8 models (nano and medium) on sperm cell dataset
• Image detection with counting and density calculation
• Video processing for frame-by-frame analysis
• Batch processing for multiple images

Dataset

• Training images: 240
• Validation images: 24
• Test images: 11
• Source: Roboflow Sperm Morphology Dataset

Results

YOLOv8n (Nano Model)

• mAP@50: 71.7%
• Precision: 67.3%
• Recall: 68.0%

YOLOv8m (Medium Model)

• Better accuracy with more parameters
• Training visualizations available in Sperm_counting/yolov8m_improved/

Project Structure

sperm-classification/
├── Full-Sperm-Cell-Detection-6/    # Dataset
│   ├── train/
│   ├── valid/
│   ├── test/
│   └── data.yaml
├── Sperm_counting/                  # Training results
│   ├── yolov8n_training/
│   ├── yolov8m_improved/
│   └── yolov8m_improved2/
├── sprem_classification.ipynb       # Main notebook
└── output_video.mp4                 # Sample output

How to Use

1. Install Requirements

pip install ultralytics opencv-python numpy pandas matplotlib roboflow jupyter

2. Train Model

from ultralytics import YOLO

model = YOLO('yolov8m.pt')
model.train(
    data='./Full-Sperm-Cell-Detection-6/data.yaml',
    epochs=50,
    imgsz=640,
    batch=16
)

3. Test on Images

model = YOLO('./Sperm_counting/yolov8m_improved/weights/best.pt')
results = model.predict('test_image.jpg', save=True)
print(f"Detected: {len(results[0].boxes)} sperm cells")

4. Process Video

results = model.predict('video.mp4', save=True)

Features

• ✅ Sperm cell detection
• ✅ Automatic counting
• ✅ Density calculation
• ✅ Batch processing
• ✅ Video analysis
• ✅ Training visualizations

Technologies Used

• YOLOv8 (Ultralytics)
• PyTorch
• OpenCV
• NumPy, Pandas
• Matplotlib
• Jupyter Notebook

Future Improvements

• Morphology classification
• Motility tracking
• Web interface
• Mobile app

License

MIT License

Author

Elsaraf

• GitHub: @Elsaraf1