Cloud Classifier

About This Project

This project implements an end-to-end image classifier capable of recognizing and distinguishing different types of clouds. It leverages pre-trained deep learning models (via transfer learning) with TensorFlow and offers an interactive web interface built with Gradio to test predictions on new images.

A key feature is a Selenium-based script that automatically captures images from a live webcam feed, classifies them, and saves the results, allowing the dataset to be continuously expanded.

---

Table of Contents

• Features
• Project Structure
• Installation
• Usage
• Dataset
• Technologies Used

Features

• Model Training: Use the CloudsClassifier class to build, train, and fine-tune models based on well-known architectures like ResNet50V2, MobileNetV2, and VGG16.
• Web Interface: A Gradio application (scripts/app.py) allows you to upload an image and get an instant prediction of the cloud type.
• Dataset Expansion: A script (src/cloudproject/expand_dataset.py) uses Selenium to capture images from a live webcam, classify them, and save the results, enabling the dataset to grow over time.
• Model Management: Easily save and load trained models in .keras format.
• Visualization: Plot learning curves (loss and accuracy) with Matplotlib.

Project Structure

.
├── data/
│   ├── dataset/          # Training and validation datasets
│   │   ├── train/
│   │   └── valid/
│   └── scraped/          # Images and predictions from scraping
├── models/               # Saved trained models (e.g., ResNet50V2.keras)
├── scripts/
│   ├── app.py            # Gradio application launch script
│   └── setup_env.sh      # Environment setup script (Linux, MacOS)
│   └── setup_env.bat     # Environment setup script (Windows)
├── src/
│   └── cloudproject/
│       ├── __init__.py
│       ├── classifier.py
│       └── scraping.py
├── .gitignore
├── README.md     
├── requirements.txt
└── setup.py

Installation

1. Clone the repository:

   git clone https://github.com/corentinpradier/cloud-project.git

2. Run the setup script: This script will create a virtual environment, install Python dependencies, and configure the project.

   Linux/MacOS :

   ./scripts/setup_env.sh

   Windows :

   ./scripts/setup_env.bat

   Note: The script uses brew to install protobuf on macOS. If you use another operating system, you will need to install it manually with the appropriate package manager.

3. Activate the virtual environment: At the start of each new terminal session, don't forget to activate the environment:

   source venv/bin/activate

Usage

Launch the prediction application

To start the Gradio web interface, run:

python scripts/app.py

Then open your browser at http://127.0.0.1:7860.

Train a new model

Training is done using the CloudsClassifier class from src/cloudproject/classifier.py. You can create a Python script or use a Jupyter notebook to orchestrate the training.

Here is an example code to train a ResNet50V2 model:

from cloudproject import CloudsClassifier

# Initialize the classifier with data paths
classifier = CloudsClassifier(
    train_dir="data/dataset/train",
    valid_dir="data/dataset/valid",
    img_height=200,
    img_width=200,
    batch_size=32,
)

# Build the model
classifier.build_model(base_model_name="ResNet50V2", learning_rate=0.001)

# Start training and fine-tuning
classifier.train(
    epochs=20,
    fine_tune_epochs=10,
    model_name="MySuperModel" # The model will be saved in models/MySuperModel.keras
)

# Display learning curves
classifier.plot_history()

Expand the dataset via scraping

To run the script that captures an image from the webcam, classifies it, and saves the result:

python scripts/get_cloud_image.py

Note: This script requires a chromedriver compatible with your version of Google Chrome and accessible in your PATH.

Dataset

The project expects the data to be structured with a train folder and a valid folder, each containing images and a _classes.csv file. This CSV should map filenames to classes using one-hot encoding.

Example predictions.csv:

filename,Class1,Class2,Class3
image_01.jpg,1,0,0
image_02.jpg,0,1,0

Technologies Used

• Language: Python 3.11
• Deep Learning: TensorFlow / Keras
• Web Interface: Gradio
• Data Manipulation: Pandas, NumPy
• Web Scraping: Selenium
• Visualization: Matplotlib