CSWin-UNet: Transformer UNet with cross-shaped windows for medical image segmentation

Open Webpage

Xiao Liu, Peng Gao, Tao Yu, Fei Wang, Ru-Yue Yuan

Published: 2025, Last Modified: 16 Jan 2026Inf. Fusion 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•A novel U-shaped encoder–decoder network architecture CSWin-UNet is designed for medical image segmentation.•A Transformer-based mechanism was incorporated to implement horizontal and vertical stripes self-attention learning.•A content-aware reassembly layer is applied for feature map upsampling.•Extensive experiments are conducted to indicate the effectiveness of CSWin-UNet.