Research map / 3D vision / Depth and stereo
PAUL: Uncertainty-Guided Partition and Augmentation for Robust Cross-View Geo-Localization under Noisy Correspondence
Zheng Li, Xueyi Zhang, Yanming Guo, Yuxiang Xie, Zhaoyun Ding, Siqi Cai, Haizhou Li, Mingrui Lao
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 L_match (InfoNCE on clean and augmented pairs) without L_EDL (ablation) | PAUL · GTA-UAV cross-area, 30% noise | R@1: 46.27 -> 55.72 | Improved |
| add L_EDL on noisy pairs without L_match (ablation) | PAUL · GTA-UAV cross-area, 30% noise | R@1: 46.27 -> 54.36 | Improved |
| add L_EDL on top of L_match (ablation) | PAUL · GTA-UAV cross-area, 30% noise | R@1: 55.72 -> 58.70 | Improved |
| add L_match on top of L_EDL (ablation) | PAUL · GTA-UAV cross-area, 30% noise | R@1: 54.36 -> 58.70 | Improved |
| increase lambda (EDL loss KL coefficient) from 0.001 to 0.005 (ablation) | PAUL · GTA-UAV cross-area, 30% noise | retrieval performance | Improved |
| increase lambda_EDL beyond 1 (ablation) | PAUL · GTA-UAV cross-area, 30% noise | retrieval performance | Hurt |
| use GMM loss-based partition instead of other partition strategies (ablation) | PAUL co-partition · GTA-UAV cross-area, 30% noise | noisy-set purity | Improved |
| guide mask generation with EDL loss instead of InfoNCE or uncertainty (ablation) | PAUL co-augment · GTA-UAV cross-area, 30% noise | retrieval performance | Improved |
| use direct (zero) masking instead of magnification, Gaussian noise or mean masking (ablation) | PAUL co-augment · GTA-UAV cross-area, 30% noise | retrieval performance | Improved |
| use generic noisy-correspondence methods (NCR, BiCro, ESC) instead of plain InfoNCE | Sample4Geo · GTA-UAV cross-area, 30% noise | R@1: 46.27 -> 44.40 (NCR) | Hurt |
| PAUL vs InfoNCE at 0% noise on same-area split | Sample4Geo · GTA-UAV same-area, 0% noise | R@1: 72.29 -> 73.55 | Improved |
| PAUL vs InfoNCE as noise rises from 0% to 60% | Sample4Geo · GTA-UAV cross-area | R@1: 42.15 -> 52.61 at 60% noise | Improved |
Closest papers by meaning
- Weather-Robust Cross-View Geo-Localization via Prototype-Based Semantic Part Discovery (NeurIPS 2026 · similarity 0.89)
- Where Am I Looking At? Joint Location and Orientation Estimation by Cross-View Matching (CVPR 2020 · similarity 0.89)
- Semantic Ambiguity Modeling and Propagation for Fine-Grained Visual Cross View Geo-Localization (AAAI 2025 · similarity 0.89)
- From Coarse to Fine: A Matching and Alignment Framework for Unsupervised Cross-View Geo-Localization (AAAI 2025 · similarity 0.88)
- View From Above: Orthogonal-View aware Cross-view Localization (CVPR 2024 · similarity 0.87)
- Aligning Geometric Spatial Layout in Cross-View Geo-Localization via Feature Recombination (AAAI 2024 · similarity 0.87)
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.
