|   | 
Details
   web
Record
Author (up) Raul Gomez; Yahui Liu; Marco de Nadai; Dimosthenis Karatzas; Bruno Lepri; Nicu Sebe
Title Retrieval Guided Unsupervised Multi-domain Image to Image Translation Type Conference Article
Year 2020 Publication 28th ACM International Conference on Multimedia Abbreviated Journal
Volume Issue Pages
Keywords
Abstract Image to image translation aims to learn a mapping that transforms an image from one visual domain to another. Recent works assume that images descriptors can be disentangled into a domain-invariant content representation and a domain-specific style representation. Thus, translation models seek to preserve the content of source images while changing the style to a target visual domain. However, synthesizing new images is extremely challenging especially in multi-domain translations, as the network has to compose content and style to generate reliable and diverse images in multiple domains. In this paper we propose the use of an image retrieval system to assist the image-to-image translation task. First, we train an image-to-image translation model to map images to multiple domains. Then, we train an image retrieval model using real and generated images to find images similar to a query one in content but in a different domain. Finally, we exploit the image retrieval system to fine-tune the image-to-image translation model and generate higher quality images. Our experiments show the effectiveness of the proposed solution and highlight the contribution of the retrieval network, which can benefit from additional unlabeled data and help image-to-image translation models in the presence of scarce data.
Address
Corporate Author Thesis
Publisher Place of Publication Editor
Language Summary Language Original Title
Series Editor Series Title Abbreviated Series Title
Series Volume Series Issue Edition
ISSN ISBN Medium
Area Expedition Conference ACM
Notes DAG; 600.121 Approved no
Call Number Admin @ si @ GLN2020 Serial 3497
Permanent link to this record