CycleVTON: Improving Diffusion-Based Virtual Try-On with Cycle-Consistent Training
Keywords: virtual try-on, diffusion, cycle-consistency
Abstract: We present CycleVTON, a cycle-consistent diffusion-based virtual try-on framework. Unlike existing methods that rely on a single try-on network, our model consists of two conjugated networks. In addition to the regular try-on network, we design a clothing extraction network that extracts the clothing worn by the person and standardizes it into a front-facing format. These two networks are symmetrical, enabling alignment between the generated dressed human and real images of dressed human, as well as between the extracted clothing and its front-facing ground truth. This cycle-consistent optimization strategy allows for enhanced retention of clothing textures and structures, ensuring a more realistic and accurate clothing generation in virtual try-on scenarios. Moreover, the conjugated network structure not only supports traditional virtual try-on but also allows flexible clothing extraction and clothing exchange between different individuals. The experiments on VITON-HD demonstrate the effectiveness of our approach.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2473
Loading