Research map / Image generation / Autoregressive and latent generation
PlaNet - Photo Geolocation with Convolutional Neural Networks
Tobias Weyand, Ilya Kostrikov, James Philbin
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 |
|---|---|---|---|
| average single-image predictions within each album | PlaNet · Google+ photo albums | % within 1 km: 14.9% -> 22.2% | Improved |
| replace averaging with a unidirectional LSTM over the album | PlaNet avg · Google+ photo albums | % within 1 km: 22.2% -> 32.0% | Improved |
| shift labels by 1 step (label offset) (ablation) | LSTM · Google+ photo albums | % within 1 km: 32.0% -> 30.9% | Hurt |
| shift labels by 2 steps (label offset) (ablation) | LSTM · Google+ photo albums | % within 1 km: 32.0% -> 29.9% | Hurt |
| repeat the sequence twice and predict on second pass (ablation) | LSTM · Google+ photo albums | % within 1 km: 32.0% -> 34.5% | Improved |
| use a bidirectional LSTM instead of repeated-sequence LSTM (25-image albums) (ablation) | LSTM rep 25 · Google+ photo albums | % within 1 km: 28.3% -> 33.0% | Improved |
| use PlaNet LSTM over single-image model on albums | PlaNet · Google+ photo albums | % within 1 km: 14.9% -> 32.0% | Improved |
| classification with adaptive cells instead of nearest-neighbour retrieval | Im2GPS (new) · Im2GPS test (237 photos) | % within 1 km: 2.5% -> 8.4% | Improved |
| classification with adaptive cells instead of nearest-neighbour retrieval | Im2GPS (new) · Im2GPS test (237 photos) | % within 750 km: 35.4% -> 53.6% | Improved |
| cell discretization for landmark images | PlaNet vs Im2GPS (new) · Im2GPS test (237 photos), manmade landmark | median error km: 61.1 -> 74.5 | Hurt |
| PlaNet vs Im2GPS on city scenes | PlaNet · Im2GPS test (237 photos), city scene | median error km: 3375.3 -> 212.6 | Improved |
| add augmentation and spatial search to PlaNet features | PlaNet features · INRIA Holidays | mAP: 73.3 -> 89.9 | Improved |
Later papers that compared with this method
1 later accepted paper reports PlaNet [36] as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Revisiting IM2GPS in the Deep Learning Era (ICCV 2017 · better on 3 of 3 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
- PIGEON: Predicting Image Geolocations (CVPR 2024 · similarity 0.88)
- Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation (CVPR 2025 · similarity 0.87)
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.87)
- LocDiff: Identifying Locations on Earth by Diffusing in the Hilbert Space (NeurIPS 2025 · similarity 0.86)
- EarthLoc: Astronaut Photography Localization by Indexing Earth from Space (CVPR 2024 · similarity 0.86)
- OpenStreetView-5M: The Many Roads to Global Visual Geolocation (CVPR 2024 · similarity 0.86)
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.
