Research map / 3D vision / Depth and stereo
Global Underwater Geolocation from Time-Lapse Polarization Imagery
Sara Aghajanzadeh, Xiaoyang Bai, Zhongmin Zhu, David Forsyth, Viktor Gruev
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 day-of-year token (ablation) | Polar Transformer · Cross-site, Champaign | RMSE (degrees): 6.5 -> 5.2 | Improved |
| use absolute instead of relative attention in Point Transformer (ablation) | Polar Transformer · Cross-site, Champaign | median geodesic error (km): 845 km -> 567 km | Improved |
| add trajectory-dropout regularization (ablation) | Polar Transformer · Cross-site, Champaign | median geodesic error (km): 567 km -> 466 km | Improved |
| double sequence length from 64 to 128 (ablation) | Polar Transformer · Cross-site, Champaign | RMSE (degrees): RMSE 3° with marginal geolocation gains | No clear effect |
| add ViT encoder summary token (ablation) | Polar Transformer · Cross-site, Champaign | RMSE and outliers | Improved |
| use Point Transformer instead of MLP for geolocation head (ablation) | Polar Transformer · Cross-site, Champaign | median geodesic error (km): above 620 km -> 466 km | Improved |
| train on simulated sequences excluding target-site frames instead of real target-site data (ablation) | Polar Transformer · Same-site geolocation | geodesic error | No clear effect |
| include target-site frames in simulator synthesis (variant b vs c) (ablation) | Polar Transformer · Same-site geolocation | geodesic error | No clear effect |
| train on synthetic data only instead of real target-site data (ablation) | Polar Transformer · Same-site solar elevation | median RMSE | Improved |
| use physics-guided synthetic training sequences with Polar Transformer instead of prior baselines | Polar Transformer · Cross-site, site-average | median geodesic error (km): 513 km vs 2,394 km (SecTran-MIM) and 3,971 km (RI-ResNet-RDM) | Improved |
Closest papers by meaning
- Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation (CVPR 2025 · similarity 0.83)
- Localization Is All You Evaluate: Data Leakage in Online Mapping Datasets and How to Fix It (CVPR 2024 · similarity 0.81)
- Weather-Robust Cross-View Geo-Localization via Prototype-Based Semantic Part Discovery (NeurIPS 2026 · similarity 0.81)
- EarthLoc: Astronaut Photography Localization by Indexing Earth from Space (CVPR 2024 · similarity 0.81)
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.80)
- SatCLIP: Global, General-Purpose Location Embeddings with Satellite Imagery (AAAI 2025 · similarity 0.80)
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.
