This project tackles road segmentation from aerial images using a conditional GAN framework that combines an Attention U-Net generator and a PatchGAN discriminator. Trained on the Massachusetts Roads dataset, the model achieves a high accuracy of 98.2%, recall of 82.3%, precision of 78.66%, an IoU of 67.19%, and an F1 score of 80.44%. These results highlight the system’s strong performance in identifying road networks while also indicating areas for future improvement, particularly in reducing false positives and enhancing overall segmentation quality.