IntraStyler: Exemplar-based Style Synthesis for Cross-modality Domain Adaptation

Han Liu, Yubo Fan, Hao Li, Dewei Hu, Daniel Moyer, Zhoubing Xu, Benoit M. Dawant, Ipek Oguz · 2026

Read IntraStyler: Exemplar-based Style Synthesis for Cross-modality Domain Adaptation by Han Liu, Yubo Fan, Hao Li, Dewei Hu, Daniel Moyer, Zhoubing Xu, Benoit M. Dawant, Ipek Oguz (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-00212.txt.

Description

Han Liu leads seven coauthors in making fake MRI scans varied enough to train tumor segmenters well. The target task marks vestibular schwannoma and the cochlea in T2 scans. Labels exist for a different scan type, not for T2. Domain adaptation bridges that gap with synthetic images. But scanners and protocols vary inside one domain too.

Older methods ignore that inner spread. Their synthetic images come out too uniform. Downstream models then generalize poorly to real scans. IntraStyler discovers fine styles on its own. A 3D style encoder pulls style apart from anatomy. Each source image can then lend its own style. The output set grows far more diverse. The method builds on a past challenge winning solution. Segmentation after training becomes more reliable. Code is public for clinical imaging teams.

Catalog record

CREATORHan Liu, Yubo Fan, Hao Li, Dewei Hu, Daniel Moyer, Zhoubing Xu, Benoit M. Dawant, Ipek Oguz
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-00212

Browse the full library · About the Tower of Babel