Aerial, Satellite & BEV Perception
Robustness and Degradation Benchmarking of Map-Conditioned Perception Under Imperfect OpenStreetMap Priors
Generated automatically from the limitations stated in 3 papers (CVPR, NeurIPS, ICCV), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.
The problem
Contemporary BEV perception, aerial localization, and landmark-guided navigation models increasingly condition their visual representations on OpenStreetMap (OSM) vector geometry and metadata. In practice, OSM data exhibits severe geographic disparities, missing road classes, topological errors, and incomplete landmark tags. Because existing methods assume clean and complete vector maps as an operational precondition, system performance degrades unpredictably in under-mapped or rural areas, preventing deployment outside densely curated metropolitan regions.
Why it matters
Enables map-conditioned autonomous navigation and visual localization systems to reliably detect when OSM priors are missing or erroneous, dynamically falling back to visual perception rather than failing catastrophically.
Ways to approach it
Prior-work checks are free with an account. Results someone already ran are shown to everyone.
- 1
Controlled Map-Corruption Stress Testing: Synthetically corrupt OSM metadata and vector topology (e.g., dropping road segments, perturbing node coordinates, masking landmark tags) across existing datasets like nuScenes, CityNav, and OpenStreetView-5M to measure the degradation curves of current state-of-the-art map-conditioned models.
- 2
Cross-Regional Heterogeneity Evaluation: Evaluate navigation and localization performance across a globally diverse geographic spectrum where OSM completeness naturally varies, measuring the exact correlation between OSM feature density metrics and perception failure modes.
- 3
Fallback and Uncertainty-Aware Conditioning: Implement an uncertainty-gated fusion module that detects discrepancies between online visual observations and OSM priors, measuring localization accuracy and route completion rate when relying primarily on vision under corrupted map segments.
Have a different approach?
Describe how you would tackle this problem and we'll look for papers that already do it. Free; your text stays private.
Why it might fail
If commercial standard-definition mapping platforms achieve near-universal global completeness and automated verification before academic solutions mature, or if purely visual end-to-end foundation models render external vector map conditioning obsolete.
Evidence
Each paper's own statement of the limitation, verbatim.
- RHO: Robust Holistic OSM-Based Metric Cross-View Geo-LocalizationCVPR 2026
Performance is bounded by the quality and availability of OpenStreetMap vector features in the target region.
- SDTagNet: Leveraging Text-Annotated Navigation Maps for Online HD Map ConstructionNeurIPS 2025
Performance depends on SD map accuracy and coverage: discrepancies between OSM and ground-truth HD maps, such as missing tunnels or small service roads, hurt prediction (Figure 6b)
- CityNav: A Large-Scale Dataset for Real-World Aerial NavigationICCV 2025
Navigation performance is dependent on the completeness and accuracy of OpenStreetMap metadata for landmark retrieval.
Nearest existing work
- U-ViLAR: Uncertainty-Aware Visual Localization for Autonomous Driving via Differentiable Association and RegistrationICCV 2025
- SafeMap: Robust HD Map Construction from Incomplete ObservationsICML 2025
- Uncertainty-Aware Vision-Based Metric Cross-View GeolocalizationCVPR 2023
- OptiMVMap: Offline Vectorized Map Construction via Optimal Multi-vehicle PerspectivesCVPR 2026
- Enhancing Vectorized Map Perception with Historical Rasterized MapsECCV 2024
- HTNav: A Hybrid Navigation Framework with Tiered Structure for Urban Aerial Vision-and-Language NavigationCVPR 2026
- MapUQ: Map with Uncertainty Quantification for Robust BEV Vectorized ConstructionICML 2026
- Enhancing Road Maps by Parsing Aerial Images Around the WorldICCV 2015
- CityNav: A Large-Scale Dataset for Real-World Aerial NavigationICCV 2025
- OpenStreetView-5M: The Many Roads to Global Visual GeolocationCVPR 2024
- Spatial Retrieval Augmented Autonomous DrivingCVPR 2026
- Leveraging Crowdsourced GPS Data for Road Extraction From Aerial ImageryCVPR 2019
- MV-Map: Offboard HD-Map Generation with Multi-view ConsistencyICCV 2023
- Parse, Search, and Confirmation: Training-Free Aerial Vision-and-Dialog Navigation with Chain-of-Thought Reasoning and Structured Spatial MemoryCVPR 2026
- RHO: Robust Holistic OSM-Based Metric Cross-View Geo-LocalizationCVPR 2026
Related open problems
Aerial, Satellite & BEV Perception
Robust Cross-View Perception and Localization Without Paired Ground-Level Street Imagery
Current cross-view visual geolocalization, 3D scene reconstruction, and BEV perception models strictly depend on dense, spatially calibrated ground-level panoramas paired with satellite imagery or pre-existing 3D city meshes. In rural, sparsely populated, or underdeveloped regions, ground-level street-view coverage is either completely absent or uncalibrated. This reliance on dense ground-level imagery structurally restricts cross-view perception algorithms to major metropolitan areas with dedicated mapping fleets.
Aerial, Satellite & BEV Perception
Benchmarking and Mitigating UAV Perception Failure Under Rapid Motion Blur, Scale Shifts, and Occlusion
Current perception models deployed on unmanned aerial vehicles achieve low absolute accuracy on standard downstream tasks (e.g., recall under 7.5% in scene understanding, under 40% in action recognition, and under 30% m_vIoU in tracking). These failures are driven by real-world flight conditions—rapid camera motion blur, extreme resolution and altitude shifts, severe occlusion, and small target sizes—that standard benchmarks do not systematically couple. As a consequence, aerial vision pipelines remain brittle and cannot be reliably trusted for autonomous navigation, search-and-rescue, or monitoring in complex flight regimes.