Based on the idea of transfer learning, we transferred the weights of the backbone network trained on a large semantic segmentation dataset to initialize our feature extraction network, and enhance ...