Research map / Vision-language and multimodal / Human and embodied modeling
GT-Loc: Unifying When and Where in Images Through a Joint Embedding Space
David G. Shatwell, Ishan Rajendrakumar Dave, Sirnam Swetha, 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 |
|---|---|---|---|
| jointly train image-time and image-location alignment (GT-Loc vs image-time only TimeLoc) (ablation) | GT-Loc · SkyFinder unseen cameras | TPS: 75.49 -> 77.00 | Improved |
| replace regression/classification heads with retrieval-based time prediction | CLIP ViT-L/14 · SkyFinder unseen cameras | TPS: 74.33 (reg. head) -> 75.49 (TimeLoc) -> 77.00 (GT-Loc) | Improved |
| use temporal metric learning with soft targets instead of CLIP-style hard-positive contrastive loss (ablation) | GT-Loc · SkyFinder unseen cameras | TPS: 71.12 -> 77.00 | Improved |
| use toroidal cyclic distance instead of Euclidean distance in TML (ablation) | GT-Loc · SkyFinder unseen cameras | TPS: 75.88 -> 77.00 | Improved |
| use CLIP ViT-L/14 backbone instead of DINOv2-L or OpenCLIP-G (ablation) | TimeLoc · SkyFinder | TPS: 68.71 (DINOv2-L), 74.65 (OpenCLIP-G) -> 75.49 | Improved |
| use RFF time encoder instead of Time2Vec or circular decomposition (ablation) | GT-Loc · SkyFinder | TPS: 75.99 (Time2Vec) -> 77.00 | Improved |
| use monthly instead of daily time-of-year scale (ablation) | GT-Loc · SkyFinder | TPS: 76.61 -> 77.00 | No clear effect |
| reduce training data to 5% (ablation) | GT-Loc · SkyFinder | TPS: 77.00 -> 72.07 | Hurt |
| add Gaussian label noise to training timestamps (ablation) | GT-Loc · SkyFinder | TPS: 77.00 -> 76.00 (sigma 1), 73.81 (sigma 2), 69.92 (sigma 3) | Mixed |
| joint image-location alignment for geo-localization | GT-Loc vs GeoCLIP · Im2GPS3k 1 km | accuracy (%): 14.11 -> 14.41 | Improved |
| use GT-Loc unified embedding for composed time+location retrieval | GT-Loc · SkyFinder composed retrieval | R@10: 29.46 -> 38.54 | Improved |
Later papers that compared with this method
1 later accepted paper reports GT-Loc as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- TIGER: A Unified Framework for Time, Images and Geo-location Retrieval (CVPR 2026 · better on 3 of 5 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- G3: An Effective and Adaptive Framework for Worldwide Geolocalization Using Large Multi-Modality Models (NeurIPS 2024 · similarity 0.85)
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.85)
- TimeSpot: Benchmarking Geo-Temporal Understanding in Vision–Language Models in Real-World Settings (ICML 2026 · similarity 0.85)
- Geolocation Representation from Large Language Models Are Generic Enhancers for Spatio-Temporal Learning (AAAI 2025 · similarity 0.84)
- Weatherproofing Retrieval for Localization with Generative AI and Geometric Consistency (ICLR 2024 · similarity 0.83)
- LocDiff: Identifying Locations on Earth by Diffusing in the Hilbert Space (NeurIPS 2025 · similarity 0.83)
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.
