Research map / Vision-language and multimodal / Multimodal LLMs
Foundation Model Selection for Remote Sensing via a Constraint-Aware Agent
Binger Chen, Tacettin Bök, Behnood Rasti, Volker Markl, Begüm Demir
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 |
|---|---|---|---|
| add task-aware orchestration with multi-turn clarification over naive sequential tool use (ablation) | Remsa-Naive · RS-FMD selection benchmark (GPT-4.1) | Avg Top-1 score: 72.67 -> 75.76 | Improved |
| add task-aware orchestration over naive tool use (ablation) | Remsa-Naive · RS-FMD selection benchmark (GPT-4.1) | MRR: 0.29 -> 0.34 | Improved |
| add task-aware orchestration over naive tool use (ablation) | Remsa-Naive · RS-FMD selection benchmark (GPT-4.1) | HQ Hit Rate (%): 37.33% -> 40.00% | Improved |
| add LLM ranking and constraint reasoning to retrieval-only (ablation) | DB-Retrieval · RS-FMD selection benchmark (GPT-4.1) | Top-1 Hit Rate (%): 12.00% -> 21.33% | Improved |
| add LLM ranking and constraint reasoning to retrieval-only (ablation) | DB-Retrieval · RS-FMD selection benchmark (GPT-4.1) | HQ Hit Rate (%): 17.33% -> 40.00% | Improved |
| add structured schema grounding over unstructured RAG (ablation) | Unstr.-RAG · RS-FMD selection benchmark (GPT-4.1) | Top-1 Hit Rate (%): 13.33% -> 21.33% | Improved |
| use Remsa on DeepSeek3.2 backbone versus naive orchestration (ablation) | Remsa-Naive · RS-FMD selection benchmark (DeepSeek3.2) | Avg Top-1 score: 72.03 -> 75.35 | Improved |
| use Remsa on LLaMA-3.3-70B backbone versus naive orchestration (ablation) | Remsa-Naive · RS-FMD selection benchmark (LLaMA-3.3-70B) | Avg Top-1 score: 69.02 -> 73.39 | Improved |
| swap backbone from GPT-4.1 to LLaMA-3.3-70B | Remsa · RS-FMD selection benchmark (GPT-4.1) | Avg Top-1 score: 75.76 -> 73.39 | Hurt |
| swap backbone from GPT-4.1 to DeepSeek3.2 | Remsa · RS-FMD selection benchmark (GPT-4.1) | Avg Top-1 score: 75.76 -> 75.35 | No clear effect |
| add multi-stage orchestration (latency cost) | Remsa-Naive · RS-FMD selection benchmark (GPT-4.1) | latency per query (s): 22.7s -> 31.7s | Hurt |
| include Application Compatibility criterion in scoring (ablation) | rubric · RS-FMD criteria sensitivity | Avg Set score: 73.32 -> 75.03 | Improved |
Closest papers by meaning
- Parameter Efficient Self-Supervised Geospatial Domain Adaptation (CVPR 2024 · similarity 0.87)
- CHOICE: Benchmarking the Remote Sensing Capabilities of Large Vision-Language Models (NeurIPS 2025 · similarity 0.85)
- GeoCoT: Towards Reliable Remote Sensing Reasoning with Manifold Perspective (CVPR 2026 · similarity 0.85)
- SkySense: A Multi-Modal Remote Sensing Foundation Model Towards Universal Interpretation for Earth Observation Imagery (CVPR 2024 · similarity 0.85)
- REOBench: Benchmarking Robustness of Earth Observation Foundation Models (NeurIPS 2025 · similarity 0.85)
- RobSense: A Robust Multi-modal Foundation Model for Remote Sensing with Static, Temporal, and Incomplete Data Adaptability (CVPR 2025 · similarity 0.85)
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.
