{"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/revisiting-im2gps-in-the-deep-learning-era","title":"Revisiting IM2GPS in the Deep Learning Era","arxiv_id":"1705.04838","date":"2017-05-13","proceeding":"ICCV 2017 10","authors":["Nam Vo","Nathan Jacobs","James Hays"],"abstract":"Image geolocalization, inferring the geographic location of an image, is a\nchallenging computer vision problem with many potential applications. The\nrecent state-of-the-art approach to this problem is a deep image classification\napproach in which the world is spatially divided into cells and a deep network\nis trained to predict the correct cell for a given image. We propose to combine\nthis approach with the original Im2GPS approach in which a query image is\nmatched against a database of geotagged images and the location is inferred\nfrom the retrieved set. We estimate the geographic location of a query image by\napplying kernel density estimation to the locations of its nearest neighbors in\nthe reference database. Interestingly, we find that the best features for our\nretrieval task are derived from networks trained with classification loss even\nthough we do not use a classification approach at test time. Training with\nclassification loss outperforms several deep feature learning methods (e.g.\nSiamese networks with contrastive of triplet loss) more typical for retrieval\napplications. Our simple approach achieves state-of-the-art geolocalization\naccuracy while also requiring significantly less training data.","url_abs":"http://arxiv.org/abs/1705.04838v1","url_pdf":"http://arxiv.org/pdf/1705.04838v1.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":"classification-1","task_name":"Classification"},{"task_slug":"deep-learning","task_name":"Deep Learning"},{"task_slug":"density-estimation","task_name":"Density Estimation"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"photo-geolocation-estimation","task_name":"Photo geolocation estimation"},{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":null,"task_name":"Triplet"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps","task":"Photo geolocation estimation","dataset":"Im2GPS","model":"Im2GPS (... 28m database)","rank_in_archive_order":5,"of":11,"metrics":{"City level (25 km)":"33.3","Continent level (2500 km)":"73.4","Country level (750 km)":"61.6","Reference images":"28M","Region level (200 km)":"47.7","Street level (1 km)":"14.4","Training images":"6M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps","task":"Photo geolocation estimation","dataset":"Im2GPS","model":"Im2GPS ([L] KNN, sigma=4)","rank_in_archive_order":7,"of":11,"metrics":{"City level (25 km)":"33.3","Continent level (2500 km)":"71.3","Country level (750 km)":"57.4","Reference images":"0","Region level (200 km)":"44.3","Street level (1 km)":"12.2","Training images":"6M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps","task":"Photo geolocation estimation","dataset":"Im2GPS","model":"Im2GPS ([L] 7011C)","rank_in_archive_order":9,"of":11,"metrics":{"City level (25 km)":"21.9","Continent level (2500 km)":"63.7","Country level (750 km)":"49.4","Reference images":"0","Region level (200 km)":"34.6","Street level (1 km)":"6.8","Training images":"6M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps3k","task":"Photo geolocation estimation","dataset":"Im2GPS3k","model":"Im2GPS (kNN, sigma = 4)","rank_in_archive_order":11,"of":14,"metrics":{"City level (25 km)":"19.4","Continent level (2500 km)":"55.9","Country level (750 km)":"38.9","Region level (200 km)":"26.9","Street level (1 km)":"7.2","Training Images":"6M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps3k","task":"Photo geolocation estimation","dataset":"Im2GPS3k","model":"Im2GPS ([L] 7011C)","rank_in_archive_order":12,"of":14,"metrics":{"City level (25 km)":"14.8","Continent level (2500 km)":"52.4","Country level (750 km)":"32.6","Region level (200 km)":"21.4","Street level (1 km)":"4.0","Training Images":"6M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-im2gps3k","task":"Photo geolocation estimation","dataset":"Im2GPS3k","model":"Im2GPS ([M] 7011C)","rank_in_archive_order":13,"of":14,"metrics":{"City level (25 km)":"14.2","Continent level (2500 km)":"52.7","Country level (750 km)":"33.5","Region level (200 km)":"21.3","Street level (1 km)":"3.7","Training Images":"6M"},"uses_additional_data":false},{"leaderboard":"/sota/photo-geolocation-estimation-on-yfcc4k","task":"Photo geolocation estimation","dataset":"YFCC4k","model":"[L]kNN, σ = 4","rank_in_archive_order":4,"of":4,"metrics":{"City (25 km)":"5.7","Continent (2500 km)":"42.0","Country (750 km)":"23.5","Region (200 km)":"11.0","Street (1 km)":"2.3"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1705.04838","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}