Real-Time-Object-Detection-Using-Yolo

End-to-end repository for training and testing a custom object detection model using the YOLO architecture. Demonstrates data preparation, hyperparameter tuning, and performance evaluation on custom datasets, starting with a COCO subset.

🎯 Project Overview

This repository provides an end-to-end pipeline for training and experimenting with state-of-the-art YOLO (You Only Look Once) models for object detection.The core purpose of this project is to establish a verified training environment, starting with a lightweight model and dataset, and then scaling efficiently to larger, custom applications.

Key Features

• YOLOv11 Architecture: Utilizes the lightweight YOLOv11-Nano (yolo11n) model for fast training and low resource consumption.
• Pipeline Validation: Uses the built-in COCO128 dataset (128 images) as a "sanity check" to ensure the entire training environment and configurations are correct.
• Custom Dataset Readiness: Structure is ready to integrate any custom dataset annotated in the standard YOLO format.
• Training and Metrics: Includes default configuration for training, validation, and outputting performance metrics (mAP, loss curves).

🛠️ Prerequisites

• Before running any training commands, ensure you have the required software installed.

1. Python: Python 3.8+
2. PyTorch: The underlying deep learning framework.
3. Ultralytics Package: The official YOLO library.

Installation It is highly recommended to use a virtual environment.

1. Install Ultralytics (YOLO framework):

pip install ultralytics

🚀 Getting Started (Initial Run)

The primary goal of the initial run is to confirm your environment (including GPU/CUDA if available) is set up correctly. We use the small, built-in COCO128 dataset for this purpose.

1. Training CommandRun the following command from your terminal. The coco128.yaml and yolo11n.pt files will be downloaded automatically on the first execution.

yolo detect train \
    data=coco128.yaml \
    model=yolo11n.pt \
    epochs=100 \
    imgsz=640

2. Output and Results All training results are saved to a directory named runs/detect/trainX/ (where X increments with each run).

• Weights: The final trained model files (best.pt, last.pt) are stored here.
• Logs: The results.csv file contains epoch-by-epoch metrics (loss, mAP).
• Plots: Plots showing loss curves, mAP progression, and confusion matrices.

📊 Next Steps: Scaling to a Custom Dataset

Once you have validated the pipeline using COCO128, follow these steps to use your own data:

1. Data Preparation (YOLO Format)Ensure your images and annotations are structured as follows

my_custom_data/
├── images/
│   ├── train/
│   └── val/
└── labels/
    ├── train/
    └── val/

• Every image in images/train must have a corresponding .txt annotation file in labels/train.

2. Create a Custom YAML File Create a file named custom_data.yaml in the repository root to define your dataset configuration:

 # custom_data.yaml
 # Paths relative to the YOLO working directory 
 path: /path/to/my_custom_data 
 train: images/train 
 val: images/val 
 # Number of classes 
 nc: 2 
 # Map class IDs (0, 1) to names 
 names: ['apple', 'banana']

3. Start Training Update the CLI command to reference your new YAML file:

 yolo detect train \ 
        data=custom_data.yaml \ 
        model=yolo11n.pt \ 
        epochs=50 \ 
        imgsz=640 

You can now adjust parameters like epochs and the model size(e.g., switch to yolo11s.pt for higher accuracy) based on your custom task and hardware constraints.