Active Learning & Auto Labeling Workshop

Overview

Welcome to the Active Learning & Auto Labeling Workshop! This hands-on tutorial will guide you through implementing an active learning pipeline for object detection using YOLO, Label Studio, MLflow, and DagsHub.

By the end of this workshop, you will be able to:

• Set up an environment for local model inference.
• Train a base object detection model using YOLO.
• Auto-label new data with the trained model.
• Identify low-confidence predictions for manual review.
• Retrain your model with improved data for better performance.

Steps Covered

1. Setup & Configuration

   • Install necessary dependencies.
   • Configure DagsHub and Colab for Git and dataset management.

2. Dataset Creation & Annotation

   • Import and structure your dataset.
   • Upload dataset versions to DagsHub.
   • Define metadata and splits (train, validation, test).

3. Training a Base Model

   • Train an object detection model using YOLO.
   • Track experiments and results with MLflow.
   • Log datasets and models to ensure reproducibility.

4. Active Learning & Auto-Labeling

   • Use the trained model to predict labels for new data.
   • Filter low-confidence predictions for manual review.
   • Store updated labels in DagsHub.

5. Model Retraining & Evaluation

   • Incorporate newly labeled data into the training set.
   • Retrain and fine-tune the model.
   • Compare performance improvements.

Setup Instructions

Open the notebook and click the "Open in Colab" badge. Follow the step-by-step instructions provided in the notebook.

Resources

• DagsHub Documentation
• YOLO Documentation
• MLflow Documentation

This project provides a practical approach to improving model accuracy through active learning. Happy coding! 🚀