We present a novel unsupervised framework for instance-level image-to-image
translation. Although recent advances have been made by incorporating
additional object annotations, existing methods often fail to handle images
with multiple disparate objects. The main cause is that, during