Rock-Paper-Scissors Image Classifier

This repository contains a TensorFlow-based deep learning project aimed at classifying images of hand gestures into three categories: Rock, Paper, or Scissors. Using Convolutional Neural Networks (CNNs) within the Keras API, the model is adept at recognizing and categorizing gestures from visual data.

Model Overview

• Framework: TensorFlow 2.15.0
• API: Keras
• Model Architecture: Convolutional Neural Network (CNN)

Architecture Details:

• Feature Extraction Layers: Conv2D layers with ReLU activation.
• Downsampling Layers: MaxPooling2D layers.
• Classification Layers: Dense layers with ReLU activation.
• Regularization: Dropout layers and L2 regularization to mitigate overfitting.
• Output Layer: Dense layer with Softmax activation for multi-class classification.

Dataset Information

• Source: Dicoding Academy
• Contents: Hand gesture images representing Rock, Paper, and Scissors.
• Split:
  • Training set: 1,312 images
  • Validation set: 876 images

Model Performance

• Optimization: Adam optimizer with a custom learning rate schedule.
• Regularization Techniques: L2 regularization and Dropout.
• Best Validation Loss: 0.26198
• Best Validation Accuracy: 96.58%

Features

• Data Augmentation: Implemented to introduce variability in the training dataset, enhancing model robustness.
• Learning Rate Scheduling: Applied to optimize convergence rates.
• Early Stopping and Model Checkpointing: Utilized to halt training at the optimal point and save the best model iteration, respectively.
• Comprehensive Evaluation: Both training and validation metrics are extensively monitored to gauge model performance.