Number Recognition Using ANN

This application is a number recognition tool that utilizes deep learning models to classify hand-drawn digits. Built with Streamlit, it provides an interactive way to test a Convolutional Neural Network (CNN) model trained on the MNIST dataset.

Features

• Users can draw digits (0-9) on a canvas.
• The application processes the drawing and makes a prediction using a trained CNN model.
• Compares predictions from two different models.
• Displays the resized image and confidence scores for each prediction.

Technologies Used

• Python: Primary programming language.
• Streamlit: Web-based interactive UI framework.
• OpenCV: Image processing.
• NumPy & Pandas: Data manipulation.
• Keras & TensorFlow: Deep learning framework for model inference.
• PIL (Pillow): Image handling.
• streamlit-drawable-canvas: Interactive drawing canvas.

Installation

To run this application locally, follow these steps:

1. Clone the repository:

   git clone https://github.com/WillOfHeaven/MinorGP.git
   cd MinorGP

2. Install the required dependencies:

   pip install -r requirements.txt

3. Run the Streamlit application:

   streamlit run app.py

Usage

1. Open the web interface.
2. Draw a digit (0-9) on the canvas.
3. The application will process and predict the number.
4. Compare results between different models.

Model Details

• New Model: new_mnist(4x4)_epoch70.h5
• Old Model: mnist.h5
• Both models were trained on the MNIST dataset but have different architectures and training parameters.

Retraining the Model

To improve the model performance, you can retrain it using additional data or fine-tune its parameters.

Steps for Retraining:

1. Prepare Data

   • Collect additional handwritten digits.
   • Preprocess images (resize to 28x28, normalize, grayscale conversion).

2. Load Existing Model

   from keras.models import load_model
   model = load_model("new_mnist(4x4)_epoch70.h5")

3. Fine-Tune or Train from Scratch

   • Freeze earlier layers if using transfer learning.
   • Use an augmented dataset to avoid overfitting.

   from keras.preprocessing.image import ImageDataGenerator
   
   datagen = ImageDataGenerator(
       rotation_range=10,
       zoom_range=0.1,
       width_shift_range=0.1,
       height_shift_range=0.1
   )
   
   model.fit(datagen.flow(X_train, y_train, batch_size=32), epochs=10, validation_data=(X_test, y_test))

4. Save the Updated Model

   model.save("updated_mnist_model.h5")

5. Deploy the Updated Model

   • Replace new_mnist(4x4)_epoch70.h5 with updated_mnist_model.h5 in app.py.
   • Restart the application.

Contributors

• Gaurav Rawat ( Alternative Solution explorer )
• Karthik Sharma Dhulipati ( ML Research )
• Mohak Kumar Srivastava ( UI )
• Naman Jain ( Research )
• Sambuddha Chatterjee ( ML Pipeline )

Feedback

We welcome your feedback and contributions! If you encounter any issues or have suggestions for improvements, feel free to submit an issue or pull request on GitHub.