aijadugar/cifar10-efficientnet-b0-transfer-learning

Transfer Learning with EfficientNet-B0 on CIFAR-10

This project demonstrates Transfer Learning using a pretrained EfficientNet-B0 model on the CIFAR-10 dataset with PyTorch.

Two transfer learning approaches are implemented:

• Feature Extraction
• Fine-Tuning

Dataset

• Dataset: CIFAR-10
• Training Images: 45,000
• Validation Images: 5,000
• Test Images: 10,000
• Classes: 10

Training Pipeline

Feature Extraction

• Load pretrained EfficientNet-B0
• Replace the final classifier
• Freeze the backbone
• Train only the classifier

Fine-Tuning

• Load the best feature extraction model
• Unfreeze features[8] and the classifier
• Fine-tune using different learning rates

Results

Feature Extraction

Loss	Accuracy

Fine-Tuning

Loss	Accuracy

Technologies

• Python
• PyTorch
• TorchVision
• NumPy
• Matplotlib
• tqdm

Author

Ankit Bari