Research map / Vision-language and multimodal / Multimodal LLMs
GEO-Detective: Unveiling Location Privacy Risks in Images with LLM Agents
Xinyu Zhang, Yixin Wu, Boyang Zhang, Chenhao Lin, Chao Shen, Michael Backes, Yang Zhang
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 experience augmented prompting (EAP) (ablation) | o3 · MP16-Pro country accuracy (Moderate) | accuracy: 60.1% -> 61.6% | Improved |
| add experience augmented prompting (EAP) (ablation) | o3 · MP16-Pro country accuracy (Very Difficult) | accuracy: 25.8% -> 15.5% | Hurt |
| add experience augmented prompting (EAP) (ablation) | GPT-4o · MP16-Pro country accuracy (Easy) | accuracy: 86.6% -> 67.7% | Hurt |
| add reverse search (RS) (ablation) | o3 · MP16-Pro country accuracy (Difficult) | accuracy: 35.5% -> 37.8% | Improved |
| add reverse search (RS) (ablation) | o3 · MP16-Pro city accuracy (Difficult) | accuracy: 9.7% -> 12.3% | Improved |
| add reverse search (RS) (ablation) | o3 · MP16-Pro state accuracy (Moderate) | accuracy: 44.1% -> 38.1% | Hurt |
| add reverse search (RS) (ablation) | GPT-4o · MP16-Pro country accuracy (Very Difficult) | accuracy: 7.2% -> 24.7% | Improved |
| add segmentation to reverse search (ablation) | o3 · MP16-Pro country accuracy (Difficult) | accuracy: 32.1% -> 40.4% | Mixed |
| add segmentation to reverse search (ablation) | GPT-4o · MP16-Pro country accuracy (Moderate) | accuracy: 50.9% -> 59.1% | Improved |
| use full GEO-Detective agent instead of baseline | o3 · MP16-Pro country accuracy (Difficult) | accuracy: 35.5% -> 40.1% | Improved |
| use full GEO-Detective agent instead of baseline | o3 · MP16-Pro unknown rate (Difficult) | unknown rate: 45.8% -> 22.6% | Improved |
| use full GEO-Detective agent instead of baseline | o3 · MP16-Pro country accuracy (Moderate) | accuracy: 60.1% -> 57.7% | Hurt |
Closest papers by meaning
- Do Vision-Language Models Respect Contextual Integrity in Location Disclosure? (ICLR 2026 · similarity 0.91)
- REVERSE: Reinforcing Evidence Verification and Search for Agentic Image Geolocation (NeurIPS 2026 · similarity 0.88)
- Recognition through Reasoning: Reinforcing Image Geo-localization with Large Vision-Language Models (NeurIPS 2025 · similarity 0.87)
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.87)
- Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.87)
- GeoReasoner: Geo-localization with Reasoning in Street Views using a Large Vision-Language Model (ICML 2024 · 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.
