基于 Vision Transformer 的 UNet 用于领域自适应语义分割
Vision Transformer-Conditioned UNet for Domain-Adaptive Semantic Segmentation
AI总结：
本文提出 ViTC-UNet，通过可学习令牌和双向注意力解码器将预训练 ViT 表示条件化于 UNet，以提升生物医学语义分割的精度与适应性。
摘要：
语义分割在生物医学研究中至关重要，但 Vision Transformers（ViTs）在该领域仍存在性能差距，尤其在稀疏、精细结构和低信噪比目标上。我们部分归因于可提示 ViT 模型中常用的轻量级像素解码器，可能缺乏高精度生物医学掩码所需的局部归纳偏置。我们通过引入 ViTC-UNet，通过可学习令牌和双向注意力解码器将预训练 ViT 表示条件化于 UNet，结合 ViT 的全局视觉先验与 UNet 的局部归纳偏置和高分辨率解码能力，同时避免端到端 ViT 微调，即使在跨领域设置中。ViTC-UNet 在 MRI 和 CT 模态的语义分割任务中均优于基线结果，证明了结构条件化的 UNet 解码可有效适应大规模视觉先验到高复杂度的生物医学分割。
英文摘要：
Semantic segmentation is essential for analysing anatomical features in biomedical research, yet a performance gap remains for Vision Transformers (ViTs) in the field, particularly for sparse, fine-structured, and low signal-to-noise targets. We attribute this challenge in part to the lightweight pixel decoders commonly used in promptable ViT models, who may lack the local inductive bias needed for high-precision biomedical masks. We bridge this gap by introducing ViTC-UNet, which conditions a UNet on frozen pre-trained ViT representations through learnable tokens and a two-way attention decoder. This combines ViT global visual priors with the local inductive bias and high-resolution decoding capacity of UNets, while avoiding end-to-end ViT fine-tuning even in cross-domain settings. ViTC-UNet outperforms baseline results in semantic segmentation tasks across MRI and CT modalities, demonstrating that structure-conditioned UNet decoding can efficiently adapt large-scale visual priors to high-complexity biomedical segmentation.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
