This project explores transfer learning for image classification on the CIFAR10 dataset. Transfer learning builds on the knowledge acquired from pre-trained models on large datasets (e.g., ImageNet) and fine-tunes them for a specific task.
The CIFAR10 dataset contains 60,000 32x32 color images belonging to 10 classes. This project aims to:
- Load and preprocess the CIFAR10 dataset.
- Utilize a pre-trained CNN models
- Train the model on the CIFAR10 dataset.
- VGG16
- ResNet34
- Convext Tiny
- Freeze initial layers of the pre-trained model.
- Fine-tune the final layers for the CIFAR10 dataset.
- Evaluate the model's performance
- Python 3.x
- Pytorch
- Scikit-learn
- torchvision
- Matplotlib
- Numpy
- timm
- Transfer Learning: Leverages the knowledge of a pre-trained model to improve performance on a different but related task.
- Fine-tuning: Adapting the final layers of a pre-trained model for a new dataset.
- Dynamic Learning Rate Optimization: Adjusting the learning rate during training to improve convergence speed and stability.
The following table shows the results of the model training and validation on the CIFAR10 dataset.
| Model | Train Accuracy | Validation Accuracy | Test Accuracy |
|---|---|---|---|
| VGG16 | 96.3% | 92.94% | 92.5% |
| ResNet34 | 97.7% | 94.7% | 94.7% |
| Convext Tiny | 96.5% | 95.7% | 95.7% |