Research map / Video / Video generation
Video2BEV: Transforming Drone Videos to BEVs for Video-based Geo-localization
Hao Ju, Shaofei Huang, Si Liu, Zhedong Zheng
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 Video2BEV transformation (3DGS BEV) to drone-video baseline (ablation) | first-stage baseline · UniV 45deg D->S | AP: 91.28 -> 95.64 | Improved |
| add second-stage re-ranking (ablation) | Video2BEV BEVs · UniV 45deg D->S | AP: 95.64 -> 96.48 | Improved |
| add diffusion-synthesized hard negatives to second stage (ablation) | Video2BEV two-stage · UniV 45deg D->S | AP: 96.48 -> 96.80 | Improved |
| add diffusion-synthesized hard negatives (ablation) | Video2BEV two-stage · UniV 45deg S->D | AP: 95.78 -> 96.57 | Improved |
| freeze first stage instead of training both stages from ImageNet weights (ablation) | Video2BEV · UniV 45deg D->S | AP: 79.29 -> 96.80 | Improved |
| freeze first stage instead of fine-tuning it (ablation) | Video2BEV · UniV 45deg S->D | AP: 95.99 -> 96.57 | Improved |
| re-rank top-32 instead of top-8 (ablation) | Video2BEV · UniV 45deg D->S | AP: 96.52 -> 96.80 | Improved |
| re-rank top-512 instead of top-32 (ablation) | Video2BEV · UniV 45deg D->S | AP: 96.80 -> 96.99 | No clear effect |
| set contrastive loss weight lambda=1 instead of 0.10 (ablation) | Video2BEV first stage · UniV 45deg D->S | AP: 95.25 -> 95.64 | Improved |
| lower drone elevation (45deg to 30deg) with more occlusions | Sample4Geo · UniV D->S | AP: 96.75 -> 86.00 | Hurt |
| lower drone elevation (45deg to 30deg) | Video2BEV · UniV D->S | AP: 96.80 -> 93.01 | Hurt |
| apply rainy augmentation before Video2BEV transformation instead of after | Video2BEV · UniV rainy OOD D->S | AP: 67.49 -> 33.75 | Hurt |
Closest papers by meaning
- Game4Loc: A UAV Geo-Localization Benchmark from Game Data (AAAI 2025 · similarity 0.90)
- Weather-Robust Cross-View Geo-Localization via Prototype-Based Semantic Part Discovery (NeurIPS 2026 · similarity 0.88)
- Where Does It Exist from the Low-Altitude: Spatial Aerial Video Grounding (NeurIPS 2025 · similarity 0.88)
- Seeing Across Skies and Streets: Feedforward 3D Reconstruction from Satellite, Drone, and Ground Images (NeurIPS 2026 · similarity 0.86)
- OrthoLoC: UAV 6-DoF Localization and Calibration Using Orthographic Geodata (NeurIPS 2025 · similarity 0.85)
- LoD-Loc: Aerial Visual Localization using LoD 3D Map with Neural Wireframe Alignment (NeurIPS 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.
