Research map / Adversarial robustness and security / Black-box and transfer attacks
Defending from GeoLocalization through Adversarial Road Trips
Niccolò Niccoli, Federico Becattini, Lorenzo Seidenari
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 |
|---|---|---|---|
| replace PGD with RoadTrip beam search over intermediate geographic targets | GeoCLIP attack · Im2GPS3k eps 2/255 | ACC_GT@2500km: 8.34% -> 1.03% | Improved |
| replace PGD with RoadTrip | GeoCLIP attack · Im2GPS3k eps 2/255 | ACC_Target@1km: 71.81% -> 93.43% | Improved |
| replace PGD with RoadTrip | GeoCLIP attack · YFCC4k eps 2/255 | ACC_GT@2500km: 4.23% -> 1.19% | Improved |
| replace PGD with RoadTrip at larger budget | GeoCLIP attack · Im2GPS3k eps 4/255 | ACC_GT@2500km: 1.27% -> 0.73% | Improved |
| use RTA instead of five-start parallel PGD (multi-start) (ablation) | PGD · Im2GPS3k eps 2/255 | ACC_GT@2500km: 5.41% -> 1.03% | Improved |
| use RTA instead of random restart PGD (ablation) | PGD · Im2GPS3k eps 2/255 | ACC_Target@1km: 64.27% -> 93.43% | Improved |
| use adaptive sampled intermediate targets instead of fixed interpolated targets (ablation) | PGD · Im2GPS3k eps 2/255 | ACC_Target@1km: 32.13% -> 93.43% | Improved |
| random restart added to PGD (ablation) | PGD · Im2GPS3k eps 2/255 | ACC_Target@1km: 71.81% -> 64.27% | Hurt |
| interpolated intermediate targets added to PGD (ablation) | PGD · Im2GPS3k eps 2/255 | ACC_GT@2500km: 8.34% -> 13.77% | Hurt |
| use RTA instead of GeoShield (transfer to Img2Loc) | attack · Im2GPS3k Img2Loc eps 2/255 | ACC_GT@2500km: 63.43% -> 27.86% | Improved |
| use RTA instead of GeoShield (transfer to G3) | attack · Im2GPS3k G3 eps 2/255 | ACC_GT@2500km: 78.14% -> 19.12% | Improved |
| use RTA instead of GeoShield on perceptual quality | attack · Im2GPS3k eps 16/255 | LPIPS: 0.305 -> 0.174 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GeoShield — GeoShield: Safeguarding Geolocation Privacy from Vision-Language Models via Adversarial Perturbations (AAAI 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
- Beware of Road Markings: A New Adversarial Patch Attack to Monocular Depth Estimation (NeurIPS 2024 · similarity 0.83)
- Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.82)
- Enduring, Efficient and Robust Trajectory Prediction Attack in Autonomous Driving via Optimization-Driven Multi-Frame Perturbation Framework (CVPR 2025 · similarity 0.82)
- Full-Distance Evasion of Pedestrian Detectors in the Physical World (NeurIPS 2024 · similarity 0.82)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.82)
- Everywhere Attack: Attacking Locally and Globally to Boost Targeted Transferability (AAAI 2025 · similarity 0.82)
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.
