Skip to content

Latest commit

 

History

History
89 lines (64 loc) · 2.43 KB

README.md

File metadata and controls

89 lines (64 loc) · 2.43 KB

Image Classification with Convolutional Neural Networks

This project demonstrates how to build and train a Convolutional Neural Network (CNN) for image classification using TensorFlow and Keras. The model is trained on a dataset of images, each labeled with one of 48 classes.

Project Structure

  • image_classification.py: Main Python script to build, train, and evaluate the CNN model.
  • famous48/: Directory containing the dataset files (x24x24.txt, y24x24.txt, z24x24.txt).

Requirements

  • Python 3.x
  • TensorFlow
  • Keras
  • NumPy
  • scikit-learn
  • Matplotlib

Installation

  1. Clone the repository:

    git clone https://github.com/wiktornazaruk/face_recognition.git
cd face_recognition

  2. Install the required Python packages:

    pip install tensorflow keras numpy scikit-learn matplotlib

Dataset

The dataset consists of three text files (x24x24.txt, y24x24.txt, z24x24.txt) located in the famous48 directory. Each file contains image data in a flattened format along with their corresponding labels.

Usage

  1. Run the image_classification.py script:

    python image_classification.py

  2. The script will:

    • Read and preprocess the dataset.
    • Split the dataset into training and testing sets.
    • Build and train a CNN model.
    • Evaluate the model on the test set.
    • Plot the training and validation accuracy.

Model Architecture

The CNN model is built using the Keras Sequential API and consists of the following layers:

  • Input layer: Shape (24, 24, 1)
  • Conv2D layer: 96 filters, kernel size 3x3, ReLU activation
  • MaxPooling2D layer: Pool size 3x3, strides 2
  • Dropout layer: 30%
  • Conv2D layer: 256 filters, kernel size 5x5, ReLU activation
  • MaxPooling2D layer: Pool size 3x3, strides 2
  • Dropout layer: 40%
  • Conv2D layer: 384 filters, kernel size 3x3, ReLU activation
  • Dropout layer: 50%
  • Conv2D layer: 384 filters, kernel size 3x3, ReLU activation
  • MaxPooling2D layer: Pool size 3x3, strides 2
  • Flatten layer
  • Dropout layer: 60%
  • Dense layer: 384 units, ReLU activation
  • Dense layer: 48 units (number of classes), softmax activation

Results

After training, the script prints the following metrics:

  • Train set accuracy
  • Train loss
  • Validation set accuracy
  • Validation set loss
  • Test set accuracy
  • Test set loss

It also plots the training and validation accuracy over epochs.

Example Output

results

chart