Research map / Vision-language and multimodal / Multimodal LLMs
GeoBrowse: A Geolocation Benchmark for Agentic Tool Use with Expert-Annotated Reasoning Traces
Xinyu Geng, Yanjing Xiao, Yuyang Zhang, Hanwen Wang, Xinyan Liu, Rui Min, Tianqing Fang, Yi Ren Fung
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 |
|---|---|---|---|
| Wrap backbone in GATE tool workflow | GPT-4o · GeoBrowse Level 1 | accuracy: 23.1 -> 31.8 | Improved |
| Wrap backbone in GATE tool workflow | GPT-4o · GeoBrowse Level 2 | accuracy: 11.9 -> 21.8 | Improved |
| Wrap backbone in GATE tool workflow | Gemini-3-Pro · GeoBrowse Level 1 | accuracy: 35.7 -> 48.2 | Improved |
| Wrap backbone in GATE tool workflow | Gemini-3-Pro · GeoBrowse Level 2 | accuracy: 22.8 -> 34.7 | Improved |
| Use full GATE (image + knowledge tools) instead of search-only OmniSearch or image-only PyVision | GPT-4o · GeoBrowse Level 2 | accuracy: OmniSearch 18.8, PyVision 13.9 -> GATE 21.8 | Improved |
| Enable all tools instead of Image Processor alone (ablation) | GATE (GPT-5) · GeoBrowse Level 1 | accuracy: 38.6 -> 39.7 | Improved |
| Enable all tools instead of Web Text Search + Visit alone (ablation) | GATE (GPT-5) · GeoBrowse Level 2 | accuracy: 27.0 -> 30.7 | Improved |
| Add Code Interpreter as the only tool (ablation) | GPT-5 · GeoBrowse Level 1 | accuracy: 26.6 -> 23.1 | Hurt |
| Agentic planning instead of fixed tool policy (ablation) | GPT-4o · GeoBrowse Level 2 | accuracy: 15.6 -> 21.8 | Improved |
| Agentic planning instead of fixed tool policy (ablation) | GPT-4o · GeoBrowse Level 1 | accuracy: 28.4 -> 31.8 | Improved |
| Give weak model tools (fixed or agentic) (ablation) | Qwen-2.5-VL-7B · GeoBrowse Level 1 | accuracy: 7.5 -> 6.1 (fixed); 7.5 -> 6.5 (agentic) | Hurt |
| Give tools to Qwen-3-VL-32B on Level 2 with fixed policy (ablation) | Qwen-3-VL-32B · GeoBrowse Level 2 | accuracy: 6.9 -> 6.9 (fixed) | No clear effect |
Closest papers by meaning
- REVERSE: Reinforcing Evidence Verification and Search for Agentic Image Geolocation (NeurIPS 2026 · similarity 0.89)
- GeoAgent: Learning to Geolocate Everywhere with Reinforced Geographic Characteristics (CVPR 2026 · similarity 0.89)
- Perceive, Interact, Reason: Building Tool-Augmented Visual Agents for Spatial Reasoning (NeurIPS 2026 · similarity 0.88)
- MapEval: A Map-Based Evaluation of Geo-Spatial Reasoning in Foundation Models (ICML 2025 · similarity 0.87)
- Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.87)
- SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning (NeurIPS 2026 · similarity 0.86)
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.
