Road Segmentation using Deep Learning with U-Net and EfficientNet

This project develops a road segmentation model using U-Net with a pre-trained EfficientNet B0 en-

coder. It involves data augmentation, custom dataset creation, and model training with PyTorch. The

model is trained with techniques like resizing, flipping, and normalization, optimized using DiceLoss and

BCEWithLogitsLoss. After training for 16 epochs, the best model is saved and used for inference on new

images.