🏊 Swimming Pool Detection using YOLOv8

📌 Project Overview

This project aims to detect swimming pools in aerial images using YOLOv8 segmentation.
The model is trained to identify pools of any shape and output their exact boundaries.

🔹 Input

• A single aerial image (.jpg, .png, etc.)

🔹 Output

• coordinates.txt → Detected pool boundary coordinates.
• output_image.jpg → Image with red outlines around pools.

---

📂 Project Structure

├── pool_detect.py        # Command-line script for pool detection
├── requirements.txt      # Required Python packages
├── server.py             # (Optional) API server for pool detection
├── test_images/          # Sample test images
│   ├── test1.jpg
│   ├── test2.jpg
│   ├── test3.jpg
│   ├── test4.jpg
│   └── test5.jpg
├── weights/
│   └── best.pt           # Trained YOLOv8 model weights
├── output_image.jpg      # Result of running pool_detect.py on test1.jpg
├── coordinates.txt       # Polygon coordinates of detected pools in test1.jpg

---

📦 Installation & Setup

1️⃣ Create a Virtual Environment

python -m venv pool_venv
source pool_venv/bin/activate  # On Linux/Mac
pool_venv\Scripts\activate     # On Windows

2️⃣ Install Dependencies

pip install -r requirements.txt

• or manually install the requiremnets *

pip install ultralytics opencv-python numpy

---

🏋 Model Training

The model was trained using YOLOv8n-seg (segmentation model) from the Ultralytics library.

🔹 Dataset Preparation

• Labeled dataset using Roboflow with polygon annotations
• Split the dataset into:
  • Train → 70%
  • Validation → 20%
  • Test → 10%
• Exported the dataset in YOLOv8 segmentation format.

🔹 Training the Model

from ultralytics import YOLO

# Load YOLOv8 segmentation model
model = YOLO("yolov8n-seg.pt")

# Train the model
model.train(data="data.yaml", epochs=50, imgsz=640, device="cpu", batch=8, workers=4)

• The best model weights are saved in:

  runs/segment/train/weights/best.pt
  

• Final model stored in weights/best.pt.

---

🚀 Running the Pool Detection Script

To detect pools in an image, use the following command:

python pool_detect.py test_images/test1.jpg

🔹 Script Explanation (pool_detect.py)

1. Loads the trained model (weights/best.pt).
2. Runs inference on the input image.
3. Filters detections with confidence ≥ 60%.
4. Extracts pool boundary coordinates and saves them to coordinates.txt.
5. Draws a red outline around detected pools (no rectangles).
6. Saves the output as output_image.jpg.

---

📊 Results

Example before & after pool detection:

• The script successfully outlines pools without adding rectangles.
• coordinates.txt contains the polygon coordinates of detected pools.

---

🌐 Testing with Gradio

You can also test the model using a Gradio interface by running the following script:

python gradio_app.py

This will launch a web interface where you can upload images and see the pool detection results in real-time.

🔹 Gradio Interface Example

• The Gradio app provides an easy-to-use interface for testing the model without writing any code.
• Simply upload an image, and the app will display the detected pools with red outlines.

---

🌐 Using the Flask Server

You can also use the server.py script to run a Flask server that takes an image and returns an image with the pool outlined.

🔹 Running the Flask Server

python server.py

🔹 Using the Flask Server

Once the server is running, you can send a POST request with an image to the server and receive the processed image in response.

Example using curl:

curl -X POST -F "file=@test_images/test1.jpg" http://127.0.0.1:5000/detect -o output_image.jpg

• The server will process the image and return the result with detected pools outlined in red.

---

🌐 Try it on Hugging Face

You can also try the model online without setting it up or installing anything by visiting the Hugging Face Space: Try Pool Detection on Hugging Face

• This space provides an interactive interface to upload images and see the pool detection results instantly.
• It's a convenient way to test the model before setting it up locally.

---

📖 References

• Ultralytics YOLOv8: https://github.com/ultralytics/ultralytics
• Dataset Source: Swimming Pool Detection - Algarve's Landscape
• Roboflow (for labeling): https://roboflow.com/

---