This project demonstrates how to use PyTorch and torchvision to classify pet images from the Oxford-IIIT Pet Dataset using a deep learning model (ReXNet) with transfer learning. It includes data visualization, class balancing insight, training monitoring, and model explainability with Grad-CAM.

We use the Oxford-IIIT Pet Dataset, which contains 37 categories (breeds of cats and dogs), with roughly 200 images for each class.

• ✅ Training Set
• 🔍 Validation Set
• 🧪 Test Set

1. Data Loading & Preprocessing

   • Loading the dataset using torchvision.datasets.OxfordIIITPet.
   • Applying necessary transformations (resizing, normalization, etc.).

2. Data Visualization

• ✅ Random Samples from Dataset – to understand image formats and contents.

  

• ⚖️ Class Distribution Charts – visualize imbalance across train, validation, and test sets.

• 📉 Training and Validation Accuracy Curves – to monitor learning progress.

  ---

  • 🔍 Grad-CAM Heatmaps - to interpret what regions of the image the model is focusing on.

    ---

    • 🔀 Confusion Matrix – to evaluate class-wise performance.

We use ReXNet v1.5, a lightweight convolutional neural network optimized for efficiency and performance:

• Backbone: rexnet_150 (loaded from timm)
• Pretrained: Yes (imagenet)
• Final Layer: Modified to match the number of pet classes (num_classes=37)
• Loss: CrossEntropyLoss
• Metrics: Accuracy and F1Score (torchmetrics.F1Score with multiclass)

The training pipeline is built using a custom TrainValidation class:

• Optimizer: Adam with learning rate = 3e-4
• Scheduler: ReduceLROnPlateau (monitors validation loss)
• Early Stopping: Stops training if F1-score doesn’t improve for 3 epochs
• Epochs: 25 (with early stopping)
• Best Model Saving: Based on highest F1-Score on validation set
• Device: GPU (cuda) or CPU fallback
• Dev Mode: For debugging small runs with dev_mode=True

1. Clone the Repository

   git clone https://github.com/matrasulov/Oxford-IIIT-Pet-Dataset---Image-Classification-using-PyTorch.git
   cd Oxford-IIIT-Pet-Dataset---Image-Classification-using-PyTorch