Dog Breed Classification using Deep Learning

Overview

This project focuses on classifying dog breeds from images using deep learning. The model is fine-tuned on the MobileNetV2 architecture and trained on the Kaggle Dog Breed Identification dataset. The implementation was done in Google Colab using TensorFlow and Keras.

Dataset

• Source: Kaggle Dog Breed Identification Dataset
• Number of Classes: 120 dog breeds
• Data Type: Images with corresponding breed labels

Model

• Pretrained Model: MobileNetV2 (TensorFlow 2)
• Fine-Tuning: Trained on the Kaggle dataset with additional layers added for classification
• Input Size: 224x224 pixels
• Optimization: Adam optimizer with categorical cross-entropy loss
• Evaluation: Accuracy and loss metrics

Implementation

Technologies Used

• Python
• TensorFlow & Keras
• Google Colab
• NumPy, Pandas, Matplotlib,Tensor Flow

Steps

1. Data Preprocessing

   • Loaded images from the Kaggle dataset
   • Resized images to 224x224 pixels
   • Normalized pixel values
   • One-hot encoded the labels

2. Model Training

   • Used MobileNetV2 as the base model
   • Added fully connected layers for classification
   • Fine-tuned the model on the dataset
   • Used data augmentation to improve generalization

3. Evaluation

   • Split dataset into training and validation sets
   • Measured accuracy and loss
   • Tested on unseen images

4. Deployment (Future Work)

   • Integrating the model into a Kotlin Jetpack Compose App
   • Using CameraX for real-time image capture

How to Use

1. Clone the repository:

   git clone https://github.com/yourusername/dog-breed-classification.git
   cd dog-breed-classification

2. Install dependencies:

   pip install -r requirements.txt

3. Run the training script:

   python train.py

4. Test the model:

   python predict.py --image path/to/image.jpg

Results

• Achieved high accuracy in breed classification
• Improved performance using fine-tuning and data augmentation
• Future goal: Deploy in a mobile app for real-time prediction

Future Improvements

• Optimize model inference speed for mobile deployment
• Improve accuracy with additional data augmentation techniques
• Implement a lightweight version for edge devices

Acknowledgments

• Kaggle for providing the dataset
• Google for the MobileNetV2 model
• TensorFlow & Keras for deep learning support
• Google colab for programming