Birds Detection using YOLOv8 🦅

A custom implementation of YOLOv8 for detecting birds in images and videos. This project provides a complete pipeline from dataset preparation to model training and inference.

📋 Table of Contents

• Features
• Prerequisites
• Installation
• Usage
• Dataset Preparation
• Model Training
• Inference
• Best Practices
• GPU Requirements

✨ Features

• Custom YOLOv8 model training for bird detection
• Support for both image and video inference
• Real-time object tracking in videos
• Configurable confidence thresholds
• Automatic annotation visualization
• Support for both local files and URLs
• Save capabilities for annotated images and videos

🔧 Prerequisites

• Python 3.8+
• CUDA-compatible GPU (recommended)
• Google Colab (optional, for free GPU access)

📦 Installation

1. Clone this repository:

git clone https://github.com/Pushtogithub23/birds-detection-yolov8.git
cd birds-detection-yolov8

2. Install required packages:

pip install -r requirements.txt

🚀 Usage

Image Detection

# For local image
display_prediction("path/to/your/image.jpg", save_fig=True, filename="detected.jpg")

# For image URL
display_prediction("https://example.com/image.jpg", save_fig=True, filename="detected.jpg")

I have attached below few detections in images:

Video Detection

predict_in_videos(
    "path/to/your/video.mp4",
    save_video=True,
    filename="detected_video.mp4"
)

I have attached a video detection in GIF format below.

📊 Dataset Preparation

1. Create a Roboflow account and obtain your API key
2. Update the API key in the notebook:

rf = Roboflow(api_key="YOUR_API_KEY")

3. The dataset structure should be:

BIRDS-DETECTION-1/
├── train/
├── valid/
├── test/
└── data.yaml

🏋️ Model Training

The model is trained using YOLOv8-large (yolov8l.pt) as the base model:

model = YOLO('yolov8l.pt')
model.train(
    data="BIRDS-DETECTION-1/data.yaml",
    epochs=100,
    imgsz=640
)

Training parameters can be adjusted based on your needs:

• epochs: Number of training epochs
• imgsz: Input image size
• batch: Batch size (adjust based on GPU memory)

🎯 Inference

Detection Function Features

• Automatic thickness calculation based on image resolution
• Dynamic text scaling
• Confidence threshold filtering (default: 0.5)
• Color-coded annotations using 'magma' colormap
• Support for both image and video processing
• Object tracking in videos using ByteTrack

Video Processing Features

• Real-time object tracking
• FPS-synchronized processing
• Progress visualization
• Press 'p' to stop processing

💡 Best Practices

Dataset Preparation

• Ensure the dataset is well-balanced
• Include images with varying lighting conditions
• Use data augmentation for better generalization

Training Tips

• Monitor training metrics in runs/detect/train directory
• Adjust confidence threshold based on application needs
• Consider model ensembling for better results

Production Deployment

• Move API keys to environment variables
• Implement proper error handling
• Consider model quantization for faster inference

💻 GPU Requirements

This project was developed and tested on Google Colab with T4 GPU acceleration. When running on different hardware:

• Adjust batch size according to available GPU memory
• Modify image size if needed
• Consider using model quantization for faster inference on less powerful hardware