Research map / 3D vision / Depth and stereo
Leaving the City: A Large-Scale Aerial Dataset for Cross-Season Localization in Unstructured Environments
Michael Schleiss, Henry Hölzemann, Fahmi Rouatbi, Torsten Fiolka, Thomas Pany, Roger Förstner, Daniel Cremers
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 |
|---|---|---|---|
| Evaluate on forest instead of artificial surfaces | SE2-LoFTR · LTC visual localization | Recall@(5m,5deg) %: 66.1 -> 33.3 | Hurt |
| Evaluate on agricultural instead of artificial surfaces | SE2-LoFTR · LTC visual localization | Recall@(5m,5deg) %: 66.1 -> 9.4 | Hurt |
| Use rotation-equivariant SE2-LoFTR instead of LoFTR | LoFTR · LTC visual localization | Recall@(5m,5deg) %: 18.8 -> 34.4 | Improved |
| Use dense matchers instead of sparse at tight thresholds | LoFTR/SE2-LoFTR · LTC visual localization | terrain gap at R1/1 | No clear effect |
| Use DINOv2-based AnyLoc instead of ground-trained CosPlace/MixVPR | VPR · LTC cross-domain VPR | Recall@1 %: 7.0 -> 19.2 | Improved |
| Evaluate on forest instead of artificial surfaces | AnyLoc · LTC cross-domain VPR | Recall@1 %: 25.8 -> 11.3 | Hurt |
| Use in-flight instead of ground initialization | VINS-Mono · LTC odometry | ATE (m): 281.5 -> 271.5 (short); 6834.5 -> 6733.4 (long 06-14) | Mixed |
| Use VINS-Mono VIO instead of ORB-SLAM VIO | visual-inertial odometry · LTC odometry 2022-02-23 11h | ATE (m): 884.9 -> 271.5 | Improved |
| Use ORB-SLAM VO instead of DROID-SLAM VO | visual odometry · LTC odometry | ATE (m): 1268.6 -> 435.2 (short 02-23 11h); 287.9 vs 490.2 reverse on 06-10 | Mixed |
| Move from small to large trajectory | VINS-Mono · LTC odometry | MPD: <=1% -> 2%-6% | Hurt |
| Water-body terrain in view | VINS-Mono/ORB-SLAM · LTC odometry failure cases | rotational error | Hurt |
| Control for image scale, reference age, season distance (ablation) | SP+LightGlue · LTC visual localization | terrain gap | No clear effect |
Closest papers by meaning
- Weather-Robust Cross-View Geo-Localization via Prototype-Based Semantic Part Discovery (NeurIPS 2026 · similarity 0.88)
- OpenCVL: An Open, Diverse, and Large-Scale Dataset for Fine-Grained Cross-View Localization (ECCV 2026 · similarity 0.87)
- Game4Loc: A UAV Geo-Localization Benchmark from Game Data (AAAI 2025 · similarity 0.86)
- OrthoLoC: UAV 6-DoF Localization and Calibration Using Orthographic Geodata (NeurIPS 2025 · similarity 0.86)
- Seeing Across Skies and Streets: Feedforward 3D Reconstruction from Satellite, Drone, and Ground Images (NeurIPS 2026 · similarity 0.86)
- View From Above: Orthogonal-View aware Cross-view Localization (CVPR 2024 · similarity 0.85)
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.
