Anaximander:在任何计算后端上交互式运行地理空间深度学习模型
Anaximander: Interactively Running Geospatial Deep Learning Models on Any Compute Backend
Satej S. Soman · Akram Zaytar · Girmaw A. Tadesse · Gilles Q. Hacheme · Muhammad S. Danish · Inbal Becker-Reshef · et al.
中文摘要
在地理信息系统(GIS)内将深度学习模型应用于卫星影像,对遥感从业者而言仍是高摩擦的工作流。模型来自互不兼容的格式,目标运行环境各异,从本地工作站到无服务器云服务皆在其列。因此,每次评估都需要先编写部署、分块和地理配准代码,才能让预测结果到达分析人员的地图上。这种摩擦不利于系统性比较,而在模型选择直接影响业务结果(如田块勾画、农作物监测和灾害响应)的领域,系统化对比尤为关键。Anaximander 是一个开源系统,在统一的交互界面下屏蔽模型来源与计算位置的选择差异。系统后端是一台推理服务器,可从多种常用来源加载模型,并在任何可访问的计算后端上提供推理服务;服务器承担会话管理与模型缓存,并按瓦片(tile)流式回传推理结果。与后端配套的是一个 QGIS 插件,负责分块、结果重组、地理配准以及实时逐瓦片状态展示。另一条交互路径可将图层图例作为提示注入视觉语言模型。在农业田块勾画任务中,系统以零代码方式对三种异构模型进行并排对比演示:经云端 API 调用的 gpt-image-1、运行在远程 GPU 上的 Segment Anything Model 3(SAM3),以及运行在本地 CPU 上的 DelineateAnything。推理后端与协议均已开放,仓库地址为 https://github.com/microsoft/nxmndr。
关键要点
- 01问题:在 GIS 中部署深度学习模型存在高摩擦,不同来源、不同计算后端的模型难以统一比较,影响田块勾画、作物监测、灾害响应等业务决策
- 02方法:Anaximander 由推理服务器 + QGIS 插件组成,统一模型来源与计算后端,提供会话管理、模型缓存、按瓦片流式回传、地理配准与实时状态可视化,并支持将图层图例作为 prompt 注入视觉语言模型
- 03演示:在农业田块勾画任务中,零代码对比 gpt-image-1(云端 API)、SAM3(远程 GPU)、DelineateAnything(本地 CPU)三种异构模型
- 04局限:摘要未提供三类模型在田块勾画任务上的精度、速度等定量对比结果
- 05开源:推理后端与协议开源,地址 https://github.com/microsoft/nxmndr
解读
尚无解读。
原始英文摘要
arXiv:2610.09085v1 Announce Type: new Abstract: Applying deep learning models to satellite imagery from within geographic information systems (GIS) remains high-friction for remote sensing practitioners. Models arrive in incompatible formats and target different compute environments, from local workstations to serverless cloud services. As a result, every evaluation demands custom deployment, tiling, and georeferencing code before a single prediction reaches the analyst's map. This friction discourages systematic comparison in a domain where model choice directly affects operational outcomes such as field delineation, crop monitoring, and disaster response. We present Anaximander, an open-source system that unifies model source and compute location choice behind one interactive interface. The system's backend is an inference server that loads models from multiple commonly-used sources and serves them on any accessible compute backend. The server provides session management and model caching, and streams results back per tile. The backend is paired with a QGIS plugin that drives tiling, result reassembly, georeferencing, and real-time per-tile status visualization. An additional user-interface path injects layer legends as prompts into vision-language models. We demonstrate the system in a code-free side-by-side comparison of three heterogeneous models on an agricultural field delineation task: gpt-image-1 via a cloud API, Segment Anything Model 3 (SAM3) on a remote GPU, and DelineateAnything on a local CPU. The inference backend and protocol are open-source and available at https://github.com/microsoft/nxmndr.