{"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/learning-aerial-image-segmentation-from","title":"Learning Aerial Image Segmentation from Online Maps","arxiv_id":"1707.06879","date":"2017-07-21","proceeding":null,"authors":["Pascal Kaiser","Jan Dirk Wegner","Aurelien Lucchi","Martin Jaggi","Thomas Hofmann","Konrad Schindler"],"abstract":"This study deals with semantic segmentation of high-resolution (aerial)\nimages where a semantic class label is assigned to each pixel via supervised\nclassification as a basis for automatic map generation. Recently, deep\nconvolutional neural networks (CNNs) have shown impressive performance and have\nquickly become the de-facto standard for semantic segmentation, with the added\nbenefit that task-specific feature design is no longer necessary. However, a\nmajor downside of deep learning methods is that they are extremely data-hungry,\nthus aggravating the perennial bottleneck of supervised classification, to\nobtain enough annotated training data. On the other hand, it has been observed\nthat they are rather robust against noise in the training labels. This opens up\nthe intriguing possibility to avoid annotating huge amounts of training data,\nand instead train the classifier from existing legacy data or crowd-sourced\nmaps which can exhibit high levels of noise. The question addressed in this\npaper is: can training with large-scale, publicly available labels replace a\nsubstantial part of the manual labeling effort and still achieve sufficient\nperformance? Such data will inevitably contain a significant portion of errors,\nbut in return virtually unlimited quantities of it are available in larger\nparts of the world. We adapt a state-of-the-art CNN architecture for semantic\nsegmentation of buildings and roads in aerial images, and compare its\nperformance when using different training data sets, ranging from manually\nlabeled, pixel-accurate ground truth of the same city to automatic training\ndata derived from OpenStreetMap data from distant locations. We report our\nresults that indicate that satisfying performance can be obtained with\nsignificantly less manual annotation effort, by exploiting noisy large-scale\ntraining data.","url_abs":"http://arxiv.org/abs/1707.06879v1","url_pdf":"http://arxiv.org/pdf/1707.06879v1.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":"learning-aerial-image-segmentation-from","repo_url":"https://github.com/Anirudh0707/Roads-and-Building-Segmentation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"learning-aerial-image-segmentation-from","repo_url":"https://github.com/alpemek/aerial-segmentation","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-segmentation","task_name":"Image Segmentation"},{"task_slug":"segmentation","task_name":"Segmentation"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1707.06879","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}