Anaximander: Interactively Running Geospatial Deep Learning Models on Any Compute Backend
Quick Answer
Anaximander is an open-source system that streamlines the deployment of geospatial deep learning models across various compute environments, enabling remote sensing practitioners to easily compare models like gpt-image-1, SAM3, and DelineateAnything without custom coding.
Quick Take
The system integrates with QGIS for real-time visualization and management of model inference, significantly reducing friction in satellite imagery analysis.
Key Points
- Anaximander supports multiple model sources and compute backends through a unified interface.
- The system includes a QGIS plugin for tiling, georeferencing, and real-time status updates.
- It enables code-free comparisons of models like gpt-image-1, SAM3, and DelineateAnything.
- The open-source backend facilitates efficient inference session management and model caching.
- Accepted as a poster at the TerraBytes II workshop, ECCV 2026.
DeepSignal Analysis
What happened
Anaximander is an open-source system designed to facilitate the deployment of geospatial deep learning models in various computing environments. It allows remote sensing practitioners to compare models like gpt-image-1, SAM3, and DelineateAnything without needing custom coding, integrating with QGIS for enhanced visualization.
Key evidence
- Anaximander addresses the high-friction deployment of deep learning models for satellite imagery, which often arrive in incompatible formats and require custom deployment.
- The system features an inference server that supports multiple model sources and can operate on any accessible compute backend, streamlining the process for users.
- Anaximander includes a QGIS plugin that manages tiling, georeferencing, and real-time visualization of model inference results, enhancing user experience.
Why it matters
The ability to easily deploy and compare different deep learning models can significantly impact operational outcomes in fields like agriculture and disaster response. By reducing the friction associated with model deployment, Anaximander enables more systematic evaluations, potentially leading to better decision-making based on model performance. This is particularly relevant as the demand for effective remote sensing solutions continues to grow.
Paper Resources
📖 Reader Mode
~2 min readAbstract: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 this https URL.
| Comments: | Accepted as a poster at the TerraBytes II workshop, ECCV 2026 |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2610.09085 [cs.CV] |
| (or arXiv:2610.09085v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2610.09085 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Satej Soman [view email]
[v1]
Tue, 6 Oct 2026 20:31:38 UTC (7,744 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.