Zero-Pair Image to Image Translation using Domain Conditional Normalization

In this paper, we propose an approach based on domain conditional\nnormalization (DCN) for zero-pair image-to-image translation, i.e., translating\nbetween two domains which have no paired training data available but each have\npaired training data with a third domain. We employ a single generator which\nhas an encoder-decoder structure and analyze different implementations of\ndomain conditional normalization to obtain the desired target domain output.\nThe validation benchmark uses RGB-depth pairs and RGB-semantic pairs for\ntraining and compares performance for the depth-semantic translation task. The\nproposed approaches improve in qualitative and quantitative terms over the\ncompared methods, while using much fewer parameters. Code available at\nhttps://github.com/samarthshukla/dcn\n