{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/planet-photo-geolocation-with-convolutional","title":"PlaNet - Photo Geolocation with Convolutional Neural Networks","arxiv_id":"1602.05314","date":"2016-02-17","proceeding":null,"authors":["Tobias Weyand","Ilya Kostrikov","James Philbin"],"abstract":"Is it possible to build a system to determine the location where a photo was\ntaken using just its pixels? In general, the problem seems exceptionally\ndifficult: it is trivial to construct situations where no location can be\ninferred. Yet images often contain informative cues such as landmarks, weather\npatterns, vegetation, road markings, and architectural details, which in\ncombination may allow one to determine an approximate location and occasionally\nan exact location. Websites such as GeoGuessr and View from your Window suggest\nthat humans are relatively good at integrating these cues to geolocate images,\nespecially en-masse. In computer vision, the photo geolocation problem is\nusually approached using image retrieval methods. In contrast, we pose the\nproblem as one of classification by subdividing the surface of the earth into\nthousands of multi-scale geographic cells, and train a deep network using\nmillions of geotagged images. While previous approaches only recognize\nlandmarks or perform approximate matching using global image descriptors, our\nmodel is able to use and integrate multiple visible cues. We show that the\nresulting model, called PlaNet, outperforms previous approaches and even\nattains superhuman levels of accuracy in some cases. Moreover, we extend our\nmodel to photo albums by combining it with a long short-term memory (LSTM)\narchitecture. By learning to exploit temporal coherence to geolocate uncertain\nphotos, we demonstrate that this model achieves a 50% performance improvement\nover the single-image model.","url_abs":"http://arxiv.org/abs/1602.05314v1","url_pdf":"http://arxiv.org/pdf/1602.05314v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"planet-photo-geolocation-with-convolutional","repo_url":"https://github.com/gjacopo/poppysite","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"EUPL-1.1"}}],"tasks":[{"task_slug":"image-retrieval","task_name":"Image Retrieval"},{"task_slug":"photo-geolocation-estimation","task_name":"Photo geolocation estimation"},{"task_slug":"retrieval","task_name":"Retrieval"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps","task":"Photo geolocation estimation","dataset":"Im2GPS","model":"PlaNet (91M)","rank_in_archive_order":8,"of":11,"metrics":{"City level (25 km)":"24.5","Continent level (2500 km)":"71.3","Country level (750 km)":"53.6","Reference images":"0","Region level (200 km)":"37.6","Street level (1 km)":"8.4","Training images":"91M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps","task":"Photo geolocation estimation","dataset":"Im2GPS","model":"PlaNet (6.2M)","rank_in_archive_order":10,"of":11,"metrics":{"City level (25 km)":"18.1","Continent level (2500 km)":"65.8","Country level (750 km)":"45.6","Reference images":"0","Region level (200 km)":"30.0","Street level (1 km)":"6.3","Training images":"6.2M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-yfcc26k","task":"Photo geolocation estimation","dataset":"YFCC26k","model":"PlaNet","rank_in_archive_order":6,"of":6,"metrics":{"City level (25 km)":"11.0","Continent level (2500 km)":"47.7","Country level (750 km)":"28.5","Region level (200 km)":"16.9","Street level (1 km)":"4.4","Training Images":"30.3M"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1602.05314","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}