Research map / Video / Video segmentation and tracking
GAReT: Cross-view Video Geolocalization with Adapters and Auto-Regressive Transformers
Manu S Pillai, Mamshad Nayeem Rizve, Mubarak Shah
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 GeoAdapter on top of the image encoder (vs Baseline 1 averaging frame embeddings) (ablation) | GAReT · GAMa sequence-to-image | R@1: 40.07 -> 50.69 | Improved |
| patch the large aerial image into small tiles for the aerial encoder instead of resizing to 512x512 with an unadapted aerial encoder (ablation) | GAReT · GAMa sequence-to-image | R@1: 15.39 -> 50.69 | Improved |
| use image-level pretraining (Baseline 1) instead of general video models | TransGeo-style encoder · GAMa sequence-to-image | R@1: 20.37 -> 40.07 | Improved |
| replace nearest neighbor retrieval with TransRetriever (k=10) (ablation) | GAReT frame-to-frame · GAMa frame-to-frame | R@1: 54.64 -> 67.66 | Improved |
| replace Dominant Sets with TransRetriever (k=20) (ablation) | GAReT frame-to-frame · GAMa frame-to-frame | R@1: 60.67 -> 70.70 | Improved |
| increase number of candidate neighbors from k=10 to k=20 (ablation) | TransRetriever · GAMa frame-to-frame | R@1: 67.66 -> 70.70 | Improved |
| replace AIM adapter with GeoAdapter (temporal attention on CLS tokens) (ablation) | GAReT · GAMa sequence-to-image | R@1: 48.06 -> 50.69 | Improved |
| replace ST-Adapter with GeoAdapter (ablation) | GAReT · GAMa sequence-to-image | R@1: 34.29 -> 50.69 | Improved |
| temporal attention on CLS tokens only instead of all patches in both branches (ablation) | GeoAdapter · GAMa sequence-to-image | retrieval recall | Improved |
| apply GAReT to SeqGeo versus prior SeqGeo method | GAReT · SeqGeo sequence-to-image | R@1: 1.80 -> 3.34 | Improved |
Closest papers by meaning
- Video Geo-Localization Employing Geo-Temporal Feature Learning and GPS Trajectory Smoothing (ICCV 2021 · similarity 0.90)
- VidTAG: Temporally Aligned Video to GPS Geolocalization with Denoising Sequence Prediction at a Global Scale (CVPR 2026 · similarity 0.89)
- GeoDTR+: Toward Generic Cross-View Geolocalization via Geometric Disentanglement (TPAMI 2024 · similarity 0.89)
- GAMa: Cross-view Video Geo-localization (ECCV 2022 · similarity 0.89)
- CVGL: Causal Learning and Geometric Topology (NeurIPS 2025 · similarity 0.87)
- GeoRouter: Dynamic Paradigm Routing for Worldwide Image Geolocalization (NeurIPS 2026 · 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.
