Research map / Adversarial robustness and security / Attacks on text and graphs
GeoShield: Safeguarding Geolocation Privacy from Vision-Language Models via Adversarial Perturbations
Xinwei Liu, Xiaojun Jia, Yuan Xun, Simeng Qin, Xiaochun Cao
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 GeoShield perturbations (eps 8/255) to street-view images | GPT-4o · Google Street View acc@1km | accuracy (%): 7.3 -> 1.1 | Improved |
| use GeoShield instead of M-Attack | GPT-4o · Google Street View acc@2500km | accuracy (%): 71.1 -> 33.8 | Improved |
| use GeoShield instead of M-Attack | Gemini-2.5 · Google Street View acc@2500km | accuracy (%): 86.3 -> 80.9 | Improved |
| add feature disentanglement module (GNFD) (ablation) | GeoShield on GPT-4o · avg distance | km: 4481 -> 7564 | Improved |
| add exposure element identification (Geo-EE) (ablation) | GeoShield on o1 · avg distance | km: 5840 -> 6780 | Improved |
| add scale-adaptive enhancement (PSAE) (ablation) | GeoShield on GPT-4o · avg distance | km: 4932 -> 7564 | Improved |
| add GNFD, Geo-EE or PSAE (ablation) | GeoShield · semantic consistency BERT-S | BERT-S: 0.8937-0.8954 -> 0.8986 | No clear effect |
| use GeoShield instead of M-Attack for semantic preservation | GPT-4o · semantic similarity eps 8/255 | BLEU: 0.09 -> 0.11 | Improved |
| use GeoShield instead of AdvDiffVLM | GPT-4o · semantic similarity eps 8/255 | BLEU: 0.14 -> 0.11 | Hurt |
| use targeted baselines (AdvDiffVLM, AnyAttack) | GPT-4o · Google Street View acc@1km | accuracy (%): 7.3 -> 5.5 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports GeoShield as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Defending from GeoLocalization through Adversarial Road Trips (ECCV 2026 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Do Vision-Language Models Respect Contextual Integrity in Location Disclosure? (ICLR 2026 · similarity 0.90)
- Disrupting Hierarchical Reasoning: Adversarial Protection for Geographic Privacy in Multimodal Reasoning Models (ICLR 2026 · similarity 0.88)
- GRE Suite: Geo-localization Inference via Fine-Tuned Vision-Language Models and Enhanced Reasoning Chains (NeurIPS 2025 · similarity 0.85)
- Recognition through Reasoning: Reinforcing Image Geo-localization with Large Vision-Language Models (NeurIPS 2025 · similarity 0.84)
- Gaussian Splatting Feature Fields for (Privacy-Preserving) Visual Localization (CVPR 2025 · similarity 0.84)
- Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.84)
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.
