Intel Image Classification

This project aims to classify images from the Intel Image Classification dataset using two renowned deep learning models: AlexNet and ResNet18. The notebook alex-res-net-classification.ipynb encompasses the entire workflow, including data preprocessing, model training, evaluation, and inference.

Project Overview

1. Dataset Acquisition:

   • The dataset is sourced from Kaggle using the kagglehub library.
   • The dataset path is displayed for reference.

2. Libraries and Dependencies:

   • Key libraries such as torch, torchvision, matplotlib, and os are imported.

3. Data Preprocessing:

   • A series of transformations are defined using torchvision.transforms to preprocess the images.
   • Images undergo resizing, random flipping, rotation, tensor conversion, and normalization.

4. Dataset Loading:

   • Training and testing directories are specified.
   • The dataset is loaded using ImageFolder and split into training and validation sets (70% training, 30% validation).

5. Data Loaders:

   • DataLoader objects are created for training, validation, and testing datasets to enable batch processing.

6. Image Display Function:

   • A function imshow is defined to display an image from the dataset along with its corresponding label.

7. Sample Images Display:

   • A batch of images is retrieved from the training dataset, and six sample images are displayed.

8. Model Preparation:

   • Pretrained models (AlexNet and ResNet18) are loaded from torchvision.models.
   • The final classification layer is adjusted to match the 6 output classes for Intel Image Classification.
   • Models are transferred to the available device (CPU or GPU).

9. Training Function:

   • A comprehensive training function is defined, incorporating:
     • Training and validation phases.
     • Loss and accuracy tracking.
     • Learning rate scheduler.
     • Model saving based on validation accuracy.
     • Training history recording.

10. Hyperparameters and Training:

    • Hyperparameters such as loss function, optimizer, learning rate scheduler, and number of epochs are defined.
    • Both AlexNet and ResNet18 models are trained, and their training history is recorded.

11. Model Evaluation:

    • Functions to evaluate the trained models on the test set are defined.
    • Metrics such as accuracy, F1-score, and confusion matrix are computed and displayed.

12. Training History Visualization:

    • Functions to plot the training and validation loss and accuracy curves are defined.
    • Training history for each model is visualized.

13. Model Inference:

    • The trained ResNet18 model is loaded and prepared for inference.
    • New images are classified using the trained model, and predictions are saved to a CSV file.

Conclusion

This project showcases the complete workflow of image classification using deep learning models. It covers data preprocessing, model training, evaluation, and inference, providing a comprehensive guide for similar tasks.