Total Generate: Cycle in Cycle Generative Adversarial Networks for Generating Human Faces, Hands, Bodies, and Natural Scenes

We propose a novel and unified Cycle In Generative Adversarial Network (C2GAN) for generating human faces, hands, bodies, natural scenes. Our proposed C2GAN is cross-modal model exploring joint exploitation of the input image data guidance in an interactive manner. contains two different generators, i.e., image-generation generator guidance-generation generator. Both generators are mutually connected trained end-to-end fashion explicitly form three cycled subnets, one generation cycle cycles. Each aims at reconstructing domain simultaneously produces useful output involved another cycle. this way, cycles constrain each other implicitly providing complementary information from both modalities bringing extra supervision gradient across cycles, facilitating more robust optimization whole model. Extensive experimental results on four guided image-to-image translation subtasks, person generation, facial expression hand gesture-to-gesture translation, cross-view demonstrate that effective realistic images compared with state-of-the-art models.