🐾 Dog Breed Classifier using CNN (PyTorch)

This repository contains a convolutional neural network (CNN) implemented in PyTorch to classify images of different dog breeds. The model is trained on a custom image dataset using a handcrafted CNN architecture, including data preprocessing, training loop, evaluation, and result visualization.

---

📁 Dataset

• Path: archive_4dog
• The dataset should follow the ImageFolder structure:

archive_4dog/
  ├── breed1/
  │    ├── img1.jpg
  │    ├── img2.jpg
  ├── breed2/
  │    ├── img3.jpg
  │    ├── ...

• Only .jpg and .png files are supported.
• Corrupted images are automatically detected and removed before training.

---

🧠 Model Overview

The network includes:

• 5 convolutional layers (with increasing filters: 32 → 512)
• ReLU activations and MaxPooling
• Fully connected classifier with dropout
• Custom image size: 150×150

---

🚀 How to Run

1. Install Requirements

pip install torch torchvision matplotlib Pillow

2. Confirm Dataset Structure

Make sure the images are sorted into subfolders by class name (breed).

3. Run the Training Script

python train_cnn.py

This will:

• Train the model
• Save model as pet_classifier_final.pth
• Save training graphs as training_results.png

---

⚙️ Hyperparameters

Parameter	Value
Batch Size	32
Image Size	150×150
Epochs	30
Optimizer	Adam
Learning Rate	0.0001
Loss Function	CrossEntropy

---

📊 Output

After training completes, the script generates:

• pet_classifier_final.pth: Final trained model weights
• training_results.png: Training/validation loss and accuracy curves

Example output:

Epoch [30/30]
Train Loss: 0.1734 | Val Loss: 0.2156 | Accuracy: 93.87%

---

📈 Training Visualization

---

🛠 Features

• Custom convolutional neural network (from scratch)
• Real-time training statistics
• Automatic removal of corrupted images
• Random rotation, flip, resize, crop augmentation

---