YOLOv11 Object Detection Pipeline

This repository contains a Jupyter Notebook for training and evaluating a YOLOv11 object detection model on a custom dataset.

Table of Contents

• YOLOv11 Object Detection Pipeline
  • Table of Contents
  • Overview
  • Dataset Processing
  • Model Training
  • Inference and Evaluation
  • Implementation Strategy
  • Results
  • Usage

Overview

This project aims to train a YOLOv11 model using a custom dataset of images and XML annotations. The pipeline includes data preprocessing, augmentation, training, and inference.

Dataset Processing

• Convert XML annotations to YOLO format.
• Apply data augmentation to improve model robustness.
• Split the dataset into training and validation sets.
• Generate a dataset.yaml file for YOLO training.

Model Training

• Define hyperparameters and configure training settings.
• Train the model using YOLOv8 with hyperparameter tuning.
• Save checkpoints and analyze performance.

Inference and Evaluation

• Perform inference using Test-Time Augmentation (TTA).
• Evaluate model predictions and generate a submission file.

Implementation Strategy

The following diagram outlines the strategy used to implement the pipeline:

Results

• Model achieved 80% mAP50-90 on validation set.
• Predictions were refined using TTA for improved accuracy.

Usage

• Create a new notebook inside the Pulmonary Nodule Detection competition
• Copy the notebook from this repository
• Paste it inside the notebook on kaggle