2019/04/28 by Xinyang Li, Li, Xinyang, Jie Hu +11 · 2 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · #Advanced Image Processing Techniques #Cancer-related molecular mechanisms research #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Multimodal Machine Learning Applications
paper · pdf · doi:10.48550/arxiv.1904.12428
openalex publication_date 2019/04/28 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
Unpaired Image-to-Image Translation (UIT) focuses on translating images among\ndifferent domains by using unpaired data, which has received increasing\nresearch focus due to its practical usage. However, existing UIT schemes defect\nin the need of supervised training, as well as the lack of encoding domain\ninformation. In this paper, we propose an Attribute Guided UIT model termed\nAGUIT to tackle these two challenges. AGUIT considers multi-modal and\nmulti-domain tasks of UIT jointly with a novel semi-supervised setting, which\nalso merits in representation disentanglement and fine control of outputs.\nEspecially, AGUIT benefits from two-fold: (1) It adopts a novel semi-supervised\nlearning process by translating attributes of labeled data to unlabeled data,\nand then reconstructing the unlabeled data by a cycle consistency operation.\n(2) It decomposes image representation into domain-invariant content code and\ndomain-specific style code. The redesigned style code embeds image style into\ntwo variables drawn from standard Gaussian distribution and the distribution of\ndomain label, which facilitates the fine control of translation due to the\ncontinuity of both variables. Finally, we introduce a new challenge, i.e.,\ndisentangled transfer, for UIT models, which adopts the disentangled\nrepresentation to translate data less related with the training set. Extensive\nexperiments demonstrate the capacity of AGUIT over existing state-of-the-art\nmodels.\n