CIFAR-10 Image Classification using ResNet50

Introduction

This project focuses on classifying images from the CIFAR-10 dataset using a deep learning model based on the ResNet50 architecture. The CIFAR-10 dataset consists of 50,000 32x32 color images in 10 different classes, with 5,000 images per class. Leveraging the power of the pre-trained ResNet50 model, the project aims to achieve high accuracy through transfer learning.

Project Overview

The steps involved in the project are as follows:

1. Dataset Preparation

   • Download the CIFAR-10 dataset using Kaggle API.
   • Extract the dataset and perform necessary preprocessing.
   • Apply One-Hot Encoding to the labels.

2. Data Preprocessing

   • Load image data and convert them into numpy arrays.
   • Normalize pixel values by scaling them to the range [0,1].
   • Split the dataset into training and testing sets.

3. Model Development

   • Utilize the ResNet50 pre-trained model with imagenet weights.
   • Add additional layers such as upsampling, dense, dropout, and batch normalization.
   • Compile the model using RMSprop optimizer and categorical crossentropy loss function.

4. Model Training and Evaluation

   • Train the model with the training dataset and validate using a validation split.
   • Evaluate model performance using accuracy and loss metrics.

5. Results Visualization

   • Plot training vs validation loss.
   • Plot training vs validation accuracy.

Requirements

To run this project, you need the following dependencies:

pip install numpy pandas matplotlib tensorflow keras opencv-python PIL scikit-learn kaggle

Execution Steps

1. Ensure you have the kaggle.json file configured to download the dataset.
2. Run the Python script to download, preprocess, train, and evaluate the model.
3. Observe the evaluation results and plots to analyze model performance.

Model Performance

The model achieved a good test accuracy, showing the effectiveness of transfer learning using ResNet50. Further improvements can be made by tuning hyperparameters and using data augmentation techniques.

Conclusion

This project demonstrates the application of deep learning for image classification using pre-trained models. ResNet50, with its powerful feature extraction capabilities, enhances the classification accuracy, making it suitable for practical applications.

Author

Arpan Pramanik

Acknowledgments

• Kaggle for the CIFAR-10 dataset.
• TensorFlow and Keras libraries for deep learning implementation.

License

This project is licensed under the Apache License.