CIFAR-10 Image Classification using ResNet

This project aims to classify images from the CIFAR-10 dataset using a Residual Network (ResNet). The notebook implements a deep learning model that leverages the ResNet50 architecture, pretrained on ImageNet, to achieve high accuracy in image classification tasks.

Dataset

The CIFAR-10 dataset consists of 60,000 32x32 color images in 10 classes, with 6,000 images per class. The classes are:

• Airplane
• Automobile
• Bird
• Cat
• Deer
• Dog
• Frog
• Horse
• Ship
• Truck

Model Architecture

The model uses the ResNet50 architecture with pre-trained ImageNet weights. It includes additional custom layers to adapt the network for CIFAR-10 classification.

Getting Started

Prerequisites

Ensure you have the following libraries installed:

• Python 3.x
• TensorFlow
• Keras
• NumPy
• Matplotlib
• Seaborn
• scikit-learn

Installation

1. Clone the repository:

   git clone https://github.com/debjit-mandal/cifar10-resnet-classification

2. Navigate to the project directory:

   cd cifar10-resnet-classification

3. Install the required packages:

   pip install -r requirements.txt

Running the Notebook

1. Start Jupyter Notebook:

   jupyter notebook

2. Open the CIFAR10_Image_Classification_ResNet.ipynb notebook and run the cells to execute the project.

Project Structure

CIFAR10_Image_Classification_ResNet/
├── CIFAR10_Image_Classification_ResNet.ipynb
├── README.md
├── requirements.txt
├── LICENSE

Results

The model achieved high accuracy in classifying images into 10 categories. Below is the classification report and confusion matrix:

Classification Report

              precision    recall  f1-score   support

           0       0.87      0.89      0.88      1000
           1       0.95      0.94      0.94      1000
           2       0.84      0.82      0.83      1000
           3       0.75      0.73      0.74      1000
           4       0.87      0.86      0.87      1000
           5       0.86      0.85      0.85      1000
           6       0.92      0.92      0.92      1000
           7       0.91      0.91      0.91      1000
           8       0.92      0.94      0.93      1000
           9       0.94      0.94      0.94      1000

    accuracy                           0.88     10000
   macro avg       0.88      0.88      0.88     10000
weighted avg       0.88      0.88      0.88     10000

Conclusion

Key Findings

• Achieved high accuracy in classifying images into 10 categories.
• Effective data augmentation improved model performance.
• Insights into model performance using the classification report and confusion matrix.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgments

• CIFAR-10 dataset
• TensorFlow
• Keras

---

Feel free to suggest any kind of improvements.