{"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/expecting-the-unexpected-training-detectors","title":"Expecting the Unexpected: Training Detectors for Unusual Pedestrians with Adversarial Imposters","arxiv_id":"1703.06283","date":"2017-03-18","proceeding":"CVPR 2017 7","authors":["Shiyu Huang","Deva Ramanan"],"abstract":"As autonomous vehicles become an every-day reality, high-accuracy pedestrian\ndetection is of paramount practical importance. Pedestrian detection is a\nhighly researched topic with mature methods, but most datasets focus on common\nscenes of people engaged in typical walking poses on sidewalks. But performance\nis most crucial for dangerous scenarios, such as children playing in the street\nor people using bicycles/skateboards in unexpected ways. Such \"in-the-tail\"\ndata is notoriously hard to observe, making both training and testing\ndifficult. To analyze this problem, we have collected a novel annotated dataset\nof dangerous scenarios called the Precarious Pedestrian dataset. Even given a\ndedicated collection effort, it is relatively small by contemporary standards\n(around 1000 images). To allow for large-scale data-driven learning, we explore\nthe use of synthetic data generated by a game engine. A significant challenge\nis selected the right \"priors\" or parameters for synthesis: we would like\nrealistic data with poses and object configurations that mimic true Precarious\nPedestrians. Inspired by Generative Adversarial Networks (GANs), we generate a\nmassive amount of synthetic data and train a discriminative classifier to\nselect a realistic subset, which we deem the Adversarial Imposters. We\ndemonstrate that this simple pipeline allows one to synthesize realistic\ntraining data by making use of rendering/animation engines within a GAN\nframework. Interestingly, we also demonstrate that such data can be used to\nrank algorithms, suggesting that Adversarial Imposters can also be used for\n\"in-the-tail\" validation at test-time, a notoriously difficult challenge for\nreal-world deployment.","url_abs":"http://arxiv.org/abs/1703.06283v2","url_pdf":"http://arxiv.org/pdf/1703.06283v2.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":"expecting-the-unexpected-training-detectors","repo_url":"https://github.com/huangshiyu13/RPNplus","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"autonomous-vehicles","task_name":"Autonomous Vehicles"},{"task_slug":"pedestrian-detection","task_name":"Pedestrian Detection"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}