{"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/cplanet-enhancing-image-geolocalization-by","title":"CPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of Maps","arxiv_id":"1808.02130","date":"2018-08-06","proceeding":"ECCV 2018 9","authors":["Paul Hongsuck Seo","Tobias Weyand","Jack Sim","Bohyung Han"],"abstract":"Image geolocalization is the task of identifying the location depicted in a\nphoto based only on its visual information. This task is inherently challenging\nsince many photos have only few, possibly ambiguous cues to their geolocation.\nRecent work has cast this task as a classification problem by partitioning the\nearth into a set of discrete cells that correspond to geographic regions. The\ngranularity of this partitioning presents a critical trade-off; using fewer but\nlarger cells results in lower location accuracy while using more but smaller\ncells reduces the number of training examples per class and increases model\nsize, making the model prone to overfitting. To tackle this issue, we propose a\nsimple but effective algorithm, combinatorial partitioning, which generates a\nlarge number of fine-grained output classes by intersecting multiple\ncoarse-grained partitionings of the earth. Each classifier votes for the\nfine-grained classes that overlap with their respective coarse-grained ones.\nThis technique allows us to predict locations at a fine scale while maintaining\nsufficient training examples per class. Our algorithm achieves the\nstate-of-the-art performance in location recognition on multiple benchmark\ndatasets.","url_abs":"http://arxiv.org/abs/1808.02130v1","url_pdf":"http://arxiv.org/pdf/1808.02130v1.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":[],"tasks":[{"task_slug":"photo-geolocation-estimation","task_name":"Photo geolocation estimation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps","task":"Photo geolocation estimation","dataset":"Im2GPS","model":"CPlaNet (1-5, PlaNet)","rank_in_archive_order":3,"of":11,"metrics":{"City level (25 km)":"37.1","Continent level (2500 km)":"78.5","Country level (750 km)":"62.0","Reference images":"0","Region level (200 km)":"46.6","Street level (1 km)":"16.5","Training images":"30.3M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps3k","task":"Photo geolocation estimation","dataset":"Im2GPS3k","model":"CPlaNet (1-5, PlaNet)","rank_in_archive_order":8,"of":14,"metrics":{"City level (25 km)":"26.5","Continent level (2500 km)":"64.4","Country level (750 km)":"48.6","Region level (200 km)":"34.6","Street level (1 km)":"10.2","Training Images":"30.3M"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1808.02130","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}