🏆 Intel Image Classification with CNNs

📜 Overview

This project focuses on experimenting with Convolutional Neural Networks (CNNs) to classify images from the Intel Image Classification dataset. The dataset consists of 17,000 color images (150x150 pixels) across 6 categories:

• 🏢 Buildings
• 🌲 Forest
• ❄️ Glacier
• ⛰️ Mountain
• 🌊 Sea
• 🏙️ Street

📌 Goals:

1. Train a CNN from scratch to classify images.
2. Experiment with different model architectures and hyperparameters.

🏗️ Model Development & Hyperparameter Tuning

✅ Data Augmentation is permitted.

Model Hyperparameters to Experiment With

• Convolutional Layers:
  • 🏗️ Number of layers
  • 🎛️ Filter size (e.g., 3x3, 5x5, 7x7)
  • 🔍 Number of filters
• Fully Connected Layers:
  • 💡 Number and size of layers
• Regularization Techniques:
  • 🔄 Dropout layers & percentages
  • 📏 L1/L2 Regularization
  • 📊 Batch Normalization

Training Hyperparameters

• 🎯 Learning rate
• 🎭 Regularization method
• 🎲 Dropout percentage
• 📦 Mini-batch size

📌 Goal: Experiment with different configurations and analyze their impact on model performance.

📊 3️⃣ Model Evaluation

• Loss Function: categorical_crossentropy
• Evaluation Metric: log_loss (log likelihood)

Other Performance Considerations

✅ Model Complexity (Fewer parameters = faster inference)
✅ Computational Efficiency (Training time & memory usage)
✅ Learning Stability (Smooth convergence vs. overfitting)

📌 Summary

✅ Built a CNN from scratch for Intel Image Classification.

✅ Experimented with different model architectures & hyperparameters.