Towards Post-disaster Damage Assessment using Deep Transfer Learning and GAN-based Data Augmentation