Traffic infrastructure segmentation is widely used for digitization of urban areas for smart city applications. In this paper, we use transfer learning for carriageway edge detection for Indian urban roads where lane edge markings are often invisible. Also, the edge of the pavement often overlaps with unpaved shoulders posing challenges for image-based edge detection. The images obtained from various routes in Chennai, India are annotated and their masks were trained using DeepLabV3with ResNet101 and MobileNetV3 as backbones to predict the classes for each pixel. Two performance metrics, AUROC and F1 score, were used to evaluate the methods. Results show that DeepLabV3 with ResNet101 outperformed DeepLabV3 with MobileNetV3.