mlsanityData to 7 Oct 2026

WeaveAgent: A Two-Stage Tool-Routing Agent for Ultra-High-Resolution Remote Sensing Imagery

arXiv 2609.31234 ↗First posted 25 Sep 2026cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

Problem. Ultra-high-resolution (UHR) remote sensing with vague user intents has two bottlenecks: visual tokens are expensive, and tool calling must be format-reliable (pretrained models emit zero tool calls zero-shot). Method. WeaveAgent, a two-stage tool-routing agent, decouples routing from visual perception. Stage A is routing-first: emission is trained, not elicited. Stage B executes conditionally: intrinsic queries enter visual answering (full-scene thumbnail; a WeaveEarth-style evidence board as an optional fixed-budget, approx. 5k-token compression interface); extrinsic queries execute tool call on original full-resolution imagery, answering from tool observations in a second, observation-masked round. Training: alignment SFT, then GRPO under reward R_WA2. Results. Alignment SFT lifts extrinsic routing from 0% to 80.75% (323/400); GRPO suppresses 9 intrinsic mis-emissions while tool selection is unchanged. The trained 2B system does not beat the zero-shot 8B baseline overall (0.263 vs. 0.250), a diagnostic contribution. Oracle attribution separates two repair ingredients: loading the observation into context lifts extrinsic answer accuracy from 0.025 to 0.425 under marker-free cross-mode returns, and the two-turn SFT stage adds a further +9.3 points to 0.518 at a small routing cost. A +/- image ablation shows emission suppression is visually grounded, and a query-register matrix shows LLM-rewritten queries cost trained checkpoints 2-11 points. Scope. All training and evaluation use the 5,000 / 3,273 / 1,000-record VagueUHR corpus (600 intrinsic + 400 tool-requiring; the base seeds synthesis and is not used for optimization). Single-pass evidence construction runs at 7.31 s per image on an RTX 4090. Code, data, and evaluation protocols will be released.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 29 of its 48 references that are in the map. First posted on arXiv 25 Sep 2026.

Builds on

29 of 48 references.

References · 29

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  2. 2023Segment Anythingcited by 8,158
  3. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  4. 2025Qwen2.5-VL Technical Reportcited by 5,418
  5. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,072
  6. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,678
  7. 2025Qwen3-VL Technical Reportcited by 2,534
  8. 2024RouteLLM: Learning to Route LLMs with Preference Datacited by 633
  9. 2025ToolRL: Reward is All Tool Learning Needscited by 342
  10. 2023GeoChat: Grounded Large Vision-Language Model for Remote Sensingcited by 320
  11. 2024MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans?cited by 284
  12. 2023RSGPT: A Remote Sensing Vision Language Model and Benchmarkcited by 204
17 more
  1. 2024EarthGPT: A Universal Multi-modal Large Language Model for Multi-sensor Image Comprehension in Remote Sensing Domaincited by 181
  2. 2023RS5M and GeoRSCLIP: A Large Scale Vision-Language Dataset and A Large Vision-Language Model for Remote Sensingcited by 150
  3. 2024ZoomEye: Enhancing Multimodal LLMs with Human-Like Zooming Capabilities through Tree-Based Image Explorationcited by 99
  4. 2024VHM: Versatile and Honest Vision Language Model for Remote Sensing Image Analysiscited by 84
  5. 2024RS-Agent: Automating Remote Sensing Tasks through Intelligent Agentcited by 46
  6. 2025GeoLLaVA-8K: Scaling Remote-Sensing Multimodal Large Language Models to 8K Resolutioncited by 41
  7. 2025When Large Vision-Language Model Meets Large Remote Sensing Imagery: Coarse-to-Fine Text-Guided Token Pruningcited by 35
  8. 2024LHRS-Bot-Nova: Improved Multimodal Large Language Model for Remote Sensing Vision-Language Interpretationcited by 25
  9. 2024Subobject-level Image Tokenizationcited by 20
  10. 2025ZoomEarth: Active Perception for Ultra-High-Resolution Geospatial Vision-Language Taskscited by 19
  11. 2025Designing Domain-Specific Agents via Hierarchical Task Abstraction Mechanismcited by 16
  12. 2026OpenEarthAgent: A Unified Framework for Tool-Augmented Geospatial Agentscited by 13
  13. 2025EarthGPT-X: A Spatial MLLM for Multi-level Multi-Source Remote Sensing Imagery Understanding with Visual Promptingcited by 11
  14. 2025Teaching Vision-Language Models to Ask: Resolving Ambiguity in Visual Questionscited by 11
  15. 2025Look Where It Matters: Training-Free Ultra-HR Remote Sensing VQA via Adaptive Zoom Searchcited by 11
  16. 2024A Gaze-grounded Visual Question Answering Dataset for Clarifying Ambiguous Japanese Questionscited by 7
  17. 2026RemoteAgent: Bridging Vague Human Intents and Earth Observation with RL-based Agentic MLLMscited by 7

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026WeaveEarth: Structured Evidence Construction and Reasoning for Training-Free UHR Remote Sensing Understanding
  2. 2026GeoEyes: On-Demand Visual Focusing for Evidence-Grounded Understanding of Ultra-High-Resolution Remote Sensing Imagery
  3. 2026The Earth in One Gaze: Training-Free Active Focus for UHR Remote Sensing Understanding
  4. 2026UHR-BAT: Budget-Aware Token Compression Vision-Language model for Ultra-High-Resolution Remote Sensing
  5. 2026EORestore-Agent: Fidelity-Guided Agentic Restoration of Remote Sensing Images with Composite Degradations
  6. 2026GeoVista: Visually Grounded Active Perception for Vision-Language Understanding of Ultra-High-Resolution Remote Sensing Images
  7. 2026HiRS-Agent: A Hierarchical Multi-Agent System for Reliable Long-Horizon Remote Sensing Task Solving
  8. 2026SA-GEM: Scale-Adaptive and Geospatial Evidence-Modulated Token Pruning for Efficient Remote Sensing Large Vision-Language Models