CRISP：面向遥感语义分割的校准感知视觉状态空间对偶模型
CRISP: Calibration-Aware Visual State Space Duality for Remote Sensing Semantic Segmentation
浏览论文内容
中文总结 AI 辅助
针对VSSD用于遥感语义分割时的边界平滑问题，提出含DCO与OMP头的CRISP框架，在约30M参数规模下提升了mF与mIoU，性能与SOTA相当。
中文摘要 AI 辅助
状态空间模型，尤其是视觉状态空间对偶模型（Visual State Space Duality，VSSD），已成为密集视觉任务中Transformer的高效线性时间替代方案。然而，我们发现VSSD会将空间上下文压缩为全局聚合，从而抑制高频响应，导致遥感语义分割中出现过度边界平滑问题。为解决该问题，我们提出CRISP，这是一个包含两个组件的校准框架。其核心为对偶校准算子（Duality Calibration Operator，DCO），通过在VSSD主干网络中注入残差和进行频率校准来恢复局部对比度与边界响应，且不改变其线性复杂度。为保留恢复的细节，正交多原型（Orthogonal Multi-Prototype，OMP）头为每个类别分配多个受正交约束的原型，以建模大的类内方差。在Potsdam、Vaihingen和LoveDA数据集上进行的大量实验表明，CRISP在参数规模约为30M的情况下，在平均F1值（mean F1，mF）和平均交并比（mean IoU，mIoU）上均取得了一致提升，同时与现有最先进方法保持竞争力。代码可在指定URL获取。
英文摘要
State space models, especially Visual State Space Duality (VSSD), have emerged as efficient linear-time alternatives to Transformers for dense visual tasks. However, we observe that VSSD compresses spatial context into a global aggregation that suppresses high-frequency responses, causing excessive boundary smoothing in remote sensing semantic segmentation. To address this, we propose CRISP, a calibration framework with two components. Its core, the Duality Calibration Operator (DCO), restores local contrast and boundary responses through residual injection and frequency calibration within the VSSD backbone, without altering its linear complexity. To retain the recovered detail, an Orthogonal Multi-Prototype (OMP) head assigns multiple orthogonally constrained prototypes per class to model large intra-class variance. Extensive experiments on Potsdam, Vaihingen, and LoveDA show that, with approximately 30M parameters, CRISP achieves consistent gains in mean F1 (mF) and mIoU while remaining competitive with state-of-the-art methods. Code is available at https://github.com/crazylifeha/CRISP.
发表机构
- Tianmushan Laboratory, Beihang University（北京航空航天大学天目山实验室）
- Hangzhou International Innovation Institute, Beihang University（北京航空航天大学杭州国际创新研究院）
- School of Astronautics, Beihang University（北京航空航天大学宇航学院）
机构由 AI 辅助整理，请以论文原文为准。
