Research map / 3D vision / Depth and stereo
LandscapeAR: Large Scale Outdoor Augmented Reality by Matching Photographs with Terrain Models Using Learned Descriptors
Jan Brejcha, Michal Lukáč, Yannick Hold-Geoffroy, Oliver Wang, Martin Čadík
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 |
|---|---|---|---|
| use adaptive semi-hard negative mining instead of random semi-hard (ablation) | Ours (two-branch descriptor) · GeoPose3K (516 test images) | fraction localized, position error <= 100 m: 0.29 -> 0.30 | Improved |
| use adaptive semi-hard negative mining instead of random semi-hard (ablation) | Ours (two-branch descriptor) · GeoPose3K (516 test images) | fraction localized, position error <= 500 m: 0.61 -> 0.63 | Improved |
| use adaptive semi-hard negative mining instead of random semi-hard (ablation) | Ours (two-branch descriptor) · GeoPose3K (516 test images) | fraction localized, position error <= 900 m: 0.67 -> 0.70 | Improved |
| use adaptive semi-hard negative mining instead of random semi-hard (ablation) | Ours (two-branch descriptor) · GeoPose3K (516 test images) | fraction localized, rotation error <= 1 deg: 0.34 -> 0.39 | Improved |
| use adaptive semi-hard negative mining instead of random semi-hard (ablation) | Ours (two-branch descriptor) · GeoPose3K (516 test images) | fraction localized, rotation error <= 9 deg: 0.64 -> 0.69 | Improved |
| add auxiliary photo-photo and render-render triplet losses (ablation) | Ours (two-branch descriptor) · GeoPose3K (516 test images) | cumulative pose error (Fig. 6) | Improved |
| use separate photo and render branches instead of a single shared encoder (ablation) | Ours (two-branch descriptor) · GeoPose3K (516 test images) | cumulative pose error (Fig. 6) | Improved |
| add pose refinement step (re-match across wider rendered FOV) (ablation) | HardNet++ and Ours · GeoPose3K (516 test images) | position error up to about 300 m | Improved |
| add pose refinement step (ablation) | D2Net · GeoPose3K (516 test images) | position error | Hurt |
| add pose refinement step (ablation) | all three descriptors · GeoPose3K (516 test images) | rotation error up to 5 deg | Improved |
| use fine-tuned larger VGG-16 single branch instead of two-branch network | Ours · Andes Huascaran (126 images) | fraction localized | Improved |
| compare two-branch Ours-aux with D2Net | Ours-aux · Andes Huascaran (126 images) | fraction localized | No clear effect |
Closest papers by meaning
- HDR Environment Map Estimation for Real-Time Augmented Reality (CVPR 2021 · similarity 0.83)
- LaMAR: Benchmarking Localization and Mapping for Augmented Reality (ECCV 2022 · similarity 0.83)
- Visual Localization Using Imperfect 3D Models From the Internet (CVPR 2023 · similarity 0.83)
- Learnable Earth Parser: Discovering 3D Prototypes in Aerial Scans (CVPR 2024 · similarity 0.82)
- LiDAR-LLM: Exploring the Potential of Large Language Models for 3D LiDAR Understanding (AAAI 2025 · similarity 0.81)
- AerialMegaDepth: Learning Aerial-Ground Reconstruction and View Synthesis (CVPR 2025 · similarity 0.81)
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.
