https://github.com/lasyakonduru/cifar-10-image-classification-with-cnns-baseline-and-enhanced-models

This project focuses on multi-class image classification using CNNs with the CIFAR-10 dataset. It compares a baseline and an enhanced model to classify 10 categories, including trucks, for real-world applications like preventing deer-vehicle collisions. Includes architecture, training, and evaluation insights.
https://github.com/lasyakonduru/cifar-10-image-classification-with-cnns-baseline-and-enhanced-models

cifar10 computer-vision convolutional-neural-networks deep-learning image-classification keras machine-learning multi-class-classification tensorflow

Last synced: 9 days ago
JSON representation

This project focuses on multi-class image classification using CNNs with the CIFAR-10 dataset. It compares a baseline and an enhanced model to classify 10 categories, including trucks, for real-world applications like preventing deer-vehicle collisions. Includes architecture, training, and evaluation insights.

• Host: GitHub
• URL: https://github.com/lasyakonduru/cifar-10-image-classification-with-cnns-baseline-and-enhanced-models
• Owner: lasyakonduru
• Created: 2024-11-15T04:52:44.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2024-12-06T21:50:38.000Z (about 2 months ago)
• Last Synced: 2025-01-21T09:11:52.885Z (9 days ago)
• Topics: cifar10, computer-vision, convolutional-neural-networks, deep-learning, image-classification, keras, machine-learning, multi-class-classification, tensorflow
• Language: HTML
• Homepage:
• Size: 1.12 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# CIFAR-10 Image Classification with Convolutional Neural Networks (CNNs)

## 📖 Project Overview

This project builds and evaluates Convolutional Neural Network (CNN) models for multi-class image classification using the CIFAR-10 dataset. The dataset contains 60,000 low-resolution images (32x32x3) across 10 categories: airplanes, cars, birds, cats, deer, dogs, frogs, horses, ships, and trucks.

Our focus is on creating models that achieve high accuracy and recall, particularly for the "deer" category, to develop AI systems for preventing deer-vehicle collisions. The enhanced model is designed for integration into future AI-powered emergency systems in vehicles.

---

## 🚀 Features

- Preprocessed CIFAR-10 dataset: normalization and one-hot encoding.

- Baseline and enhanced CNN architectures.

- Performance comparison using metrics such as accuracy, recall, and confusion matrices.

- Prediction examples with probabilities and visualizations.

- Evaluation and recommendations for further model improvements.

---

## 🗂 Dataset

- **Source**: [CIFAR-10 Dataset](https://www.cs.toronto.edu/~kriz/cifar.html)

- **Structure**:

  - 50,000 training images

  - 10,000 testing images

  - Categories: Airplanes, Cars, Birds, Cats, Deer, Dogs, Frogs, Horses, Ships, Trucks

---

## ⚙️ Project Workflow

### 1. **Data Preparation**

   - Loaded CIFAR-10 dataset.

   - Normalized image pixel values to [0, 1] for faster training.

   - Applied one-hot encoding to categorical target labels.

### 2. **Baseline Model**

   - **Architecture**:

     - 2 convolutional layers (64 filters each, ReLU activation, max pooling).

     - 1 dense feedforward layer with 120 neurons.

     - Output layer with 10 classes (softmax activation).

   - **Training**: Adam optimizer, learning rate = 0.001, batch size = 32, epochs = 10.

   - **Accuracy**: ~65% on the test set.

### 3. **Enhanced Model**

   - **Architecture**:

     - 3 convolutional layers (128, 64, 32 filters respectively, ReLU activation).

     - MaxPooling and dropout for regularization.

     - 2 dense feedforward layers with 128 and 64 neurons, respectively.

   - **Training**: Similar to baseline with additional dropout for regularization.

   - **Accuracy**: ~67% on the test set.

   - Improved recall for "truck" class.

### 4. **Evaluation**

   - Confusion matrix for class-wise accuracy.

   - Classification report with precision, recall, and F1-score.

   - Visualization of predictions with confidence scores.

---

## 📊 Results

- **Baseline Model**:

  - Test Accuracy: ~65%

  - Validation Loss: Slight overfitting observed.

- **Enhanced Model**:

  - Test Accuracy: ~67%

  - Recall and precision for "truck" significantly improved.

  - Better generalization compared to baseline.

---

## 📈 Visualizations

- Accuracy and loss curves for both models.

- Confusion matrix to evaluate predictions per class.

- Example predictions with true labels and confidence scores.

---

## 📚 Key Takeaways

1. **Baseline Model**: A simple CNN achieved moderate performance (~65%).

2. **Enhanced Model**: Improvements in architecture and regularization led to better accuracy (~67%) and recall.

3. **Future Work**:

   - Increase training data through augmentation.

   - Fine-tune hyperparameters (learning rate, dropout, etc.).

   - Explore transfer learning with pre-trained models like ResNet or MobileNet.

---

## 🛠️ Tools and Libraries

- Python 3.7+

- TensorFlow / Keras

- NumPy, Matplotlib, Seaborn

- Scikit-learn

- Plotly (optional)

---

## 👩‍💻 Contributing

Contributions are welcome! Please fork the repository and create a pull request with your proposed changes.

---

## 📜 License

This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.

---

## 🙌 Acknowledgments

- CIFAR-10 Dataset by the Canadian Institute for Advanced Research.

- Inspiration from Purdue University’s study on deer-vehicle collisions.