Research map / Vision-language and multimodal / Multimodal LLMs
OpenEarthAgent: A Unified Framework for Tool-Augmented Geospatial Agents
Akashah Shabbir, Muhammad Umer Sheikh, Muhammad Akhtar Munir, Hiyam Debary, Mustansar Fiaz, Muhammad Zaigham Zaheer, Paolo Fraccaro, Fahad Shahbaz Khan, Muhammad Haris Khan, Xiao Xiang Zhu, Salman Khan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| SFT on validated tool-use trajectories | Qwen3-4B-Instruct-2507 · OpenEarthAgent benchmark end-to-end | answer accuracy: 13.72 -> 45.26 | Improved |
| SFT on validated tool-use trajectories | Qwen3-4B-Instruct-2507 · OpenEarthAgent benchmark step-by-step | argument value accuracy: 33.55 -> 62.10 | Improved |
| SFT on trajectories | Qwen3-4B-Instruct-2507 · OpenEarthAgent benchmark step-by-step | summarization accuracy: 83.28 -> 83.64 | No clear effect |
| SFT on trajectories | Qwen3-4B-Instruct-2507 · ThinkGeo | Ans: 12.53 -> 16.35 | Improved |
| SFT on trajectories | Qwen3-4B-Instruct-2507 · Earth-Agent benchmark step-by-step | tool selection accuracy: 60.61 -> 70.33 | Improved |
| SFT on trajectories vs frontier gpt-4o | OpenEarthAgent · Earth-Agent benchmark step-by-step | tool selection accuracy: 71.65 -> 70.33 | Hurt |
| combine easy and hard trajectories in training (ablation) | OpenEarthAgent · OpenEarthAgent benchmark end-to-end | answer accuracy: 43.40 (easy) / 35.79 (hard) -> 45.26 | Improved |
| add in-context examples (k=15) to untuned model (ablation) | Qwen3-4B-Instruct-2507 · OpenEarthAgent benchmark end-to-end | answer accuracy: 13.72 -> 33.06 | Improved |
| increase in-context examples from 10 to 15 (ablation) | Qwen3-4B-Instruct-2507 · OpenEarthAgent benchmark end-to-end | image generation accuracy: 51.72 -> 46.90 | Hurt |
| provide full tool schema with JSON examples (Sch.5) vs minimal (ablation) | OpenEarthAgent · OpenEarthAgent benchmark end-to-end | answer accuracy: 8.44 -> 45.26 | Improved |
| provide executable-constraint schema (Sch.4) vs signature schema (Sch.3) (ablation) | OpenEarthAgent · OpenEarthAgent benchmark end-to-end | tool order AnyOrder: 26.17 -> 26.55 | No clear effect |
| use optimized tool order with order rule vs shuffled (ablation) | OpenEarthAgent · OpenEarthAgent benchmark end-to-end | answer accuracy: 43.35 -> 45.26 | Improved |
Closest papers by meaning
- RemoteReasoner: Towards Unifying Geospatial Reasoning Workflow (AAAI 2026 · similarity 0.92)
- GeoMMBench and GeoMMAgent: Toward Expert-Level Multimodal Intelligence in Geoscience and Remote Sensing (CVPR 2026 · similarity 0.91)
- Towards Faithful Reasoning in Remote Sensing: A Perceptually-Grounded GeoSpatial Chain-of-Thought for Vision-Language Models (ICLR 2026 · similarity 0.91)
- Asking like Socrates: Socrates helps VLMs understand remote sensing images (CVPR 2026 · similarity 0.89)
- Earth-Agent: Unlocking the Full Landscape of Earth Observation with Agents (ICLR 2026 · similarity 0.89)
- TerraScope: Pixel-Grounded Visual Reasoning for Earth Observation (CVPR 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
