arXiv 2605.126402026-05-19cs.CV

MambaPanoptic：基于视觉Mamba的结构状态空间框架用于全景分割

MambaPanoptic: A Vision Mamba-based Structured State Space Framework for Panoptic Segmentation

Technical University of Munich（慕尼黑技术大学）；Munich Center for Machine Learning (MCML)（慕尼黑机器学习中心）；Polytechnic University of Milan（米兰理工大学）；University of Stuttgart（斯图加特大学）；Wuhan University（武汉大学）；Karlsruhe University of Applied Sciences（卡尔斯鲁厄应用科学大学）

Qing Cheng, Damiano Bertolini, Wei Zhang, Dong Wang, Niclas Zeller, Daniel Cremers

备注：: Accepted to ISPRS Congress 2026, camera-ready version

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本研究提出MambaPanoptic，一种基于视觉Mamba的结构状态空间框架，旨在解决全景分割中长程上下文建模、多尺度特征表示和高效密集预测的挑战，通过引入MambaFPN和改进的PanopticFCN风格核生成器实现统一的实例和物质预测。

摘要：

全景分割要求同时识别可计数的实例和无形态的物质区域，对长程上下文建模、多尺度特征表示和高效密集预测提出了联合需求。现有的卷积和Transformer方法难以同时满足这三个要求：卷积架构在建模长程依赖方面能力有限，而基于Transformer的方法在高分辨率下会带来二次计算成本。在本文中，我们提出MambaPanoptic，一种完全基于Mamba的全景分割框架，通过两个主要贡献来解决这些限制。首先，我们引入MambaFPN，一种自上而下的特征金字塔，利用Mamba块生成具有线性计算复杂度的全局一致、多尺度特征表示。其次，我们采用PanopticFCN风格的核生成器，产生统一的实例和物质核用于无提案的全景预测，并通过在多个网络阶段应用QuadMamba基于的特征细化模块进行增强。在Cityscapes和COCO全景分割基准测试中，实验表明MambaPanoptic在同等模型大小下一致优于PanopticDeepLab和PanopticFCN，并在Cityscapes上以更少的参数匹配或超越Mask2Former在PQ和AP上的表现。

英文摘要：

Panoptic segmentation requires the simultaneous recognition of countable thing instances and amorphous stuff regions, placing joint demands on long-range context modelling, multi-scale feature representation, and efficient dense prediction. Existing convolutional and transformer-based methods struggle to satisfy all three requirements concurrently: convolutional architectures are limited in their capacity to model long-range dependencies, while transformer-based methods incur quadratic computational cost that is prohibitive at high resolutions. In this paper, we propose MambaPanoptic, a fully Mamba-based panoptic segmentation framework that addresses these limitations through two principal contributions. First, we introduce MambaFPN, a top-down feature pyramid that leverages Mamba blocks to generate globally coherent, multi-scale feature representations with linear computational complexity. Second, we adopt a PanopticFCN-style kernel generator that produces unified thing and stuff kernels for proposal-free panoptic prediction, enhanced by a QuadMamba-based feature refinement module applied at multiple network stages. Experiments on the Cityscapes and COCO panoptic segmentation benchmarks demonstrate that MambaPanoptic consistently outperforms PanopticDeepLab and PanopticFCN under comparable model sizes, and matches or surpasses Mask2Former on Cityscapes in PQ and AP while requiring fewer parameters.

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。