[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-128439-en":3,"doc-seo-128439-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},128439,962084925290,"Ophelia","https://ap-avatar.wpscdn.com/davatar_085a072bc5b1113ac321206ff7593b45",8,"Research & Report","Efficient Sample Mining for Object Detection","Object detection systems based on sliding-window classifiers are commonly trained in two stages: an initial model is learned from positive object images and randomly sampled negatives, then a bootstrapping or hard-mining step scans scenes with the initial classifier to collect negatives that were incorrectly judged as positives. This process yields “hard” samples that tighten the decision boundary. The work targets the inefficiency of dense bootstrapping by identifying scene regions that can be discarded without further computation. Experiments on pedestrian and face detection demonstrate substantial multi-fold speedups.","View metadata, citation and similar [papers at ](papers at core.ac.uk)[core.ac.uk](papers at core.ac.uk) brought to you by CORE  \nprovided by Infoscience- École polytechnique fédérale de Lausanne  \nJMLR: Workshop and Conference Proceedings 37:1{15, 2014 ACML 2014  \nE􀀎cient Sample Mining for Object Detection  \nOlivier Can􀀓evet [olivier.canevet@idiap.ch](olivier.canevet@idiap.ch)  \nFran􀀘cois Fleuret [francois.fleuret@idiap.ch](francois.fleuret@idiap.ch)  \nComputer Vision and Learning group, Idiap Research Institute, Martigny, Switzerland  \n􀀓  \nEcole Polytechnique F􀀓ed􀀓erale de Lausanne, Lausanne, Switzerland  \nEditor: Dinh Phung and Hang Li  \nAbstract  \nObject detectors based on the sliding window technique are usually trained in two successive steps: 􀀌rst, an initial classi􀀌er is trained on a population of positive samples (i. e. images of the object to detect) and negative samples randomly extracted from scenes which do not contain the object to detect. Then, the scenes are scanned with that initial classi􀀌er to enrich the initial set with negative samples incorrectly classi􀀌ed as positive. This bootstrapping process provides the learning algorithm with \\hard\" samples, which help to improve the decision boundary.  \nLittle work has been done on how to e􀀎ciently enrich the training set. While the standard bootstrapping approach densely visits the scenes, we propose to evaluate which regions of scenes can be discarded without any further computation to concentrate the search on promising areas.  \nWe apply our method to two standard object detection settings, pedestrian and face detection, and show that it provides a multi-fold speed up.  \nKeywords: Computer Vision, Object Detection, Hard Mining, Bootstrapping  \n1. Introduction  \nWith many real world applications such as surveillance, tra􀀎c analysis or content based indexing, object detection remains one of the core problems in computer vision. Slidingwindow detectors are based on a binary classi􀀌er that scans the image at all possible scalesand locations, and outputs a score related to the presence or absence of the object of interest. Multiple detections are then removed by non-maxima suppression to output the 􀀌nal detections for the test image.  \nThis binary classi􀀌er is usually trained in two steps, both using positives samples (i. e. images of the object) and negative samples (i. e. object-free images) . In the 􀀌rst step, negative examples are randomly extracted out of scenes that do not contain the object of interest. The training produces a classi􀀌er f1 which already discriminates well between objects and background, but still raises too many false alarms to be used.  \nThe second step is based on a bootstrapping procedure, or hard mining, which consists of augmenting the training set with \\hard\" negative samples. A hard negative sample x is a negative sample on which the binary classi􀀌er f1 has a strongly positive response, orin the case of SVM, that lies inside the margin of the predictor (f1 (x) 􀀕 􀀀1, Felzenszwalbet al. 2010b) . Object-free scenes are thus densely parsed until enough samples have been found and a second classi􀀌er f2 is trained on this augmented set.  \n􀀍c 2014 O. Can􀀓evet & F. Fleuret.  \nCanvet Fleuret  \nMany works over the past years have focused on proposing new features to improve the performance of the 􀀌nal detector. Haar-wavelets proposed by Viola and Jones (2001) have shown to be extremely e􀀎cient combined with AdaBoost to detect faces in images. Histograms of Oriented Gradients (HOGs, Dalal and Triggs 2005) are currently used by almost all state-of-the-art detectors. Walk et al. (2010) proposed features based on the optical 􀀍owin videos and on the self-similarity between color channels. Integral or aggregated channel features (Doll􀀓ar et al. (2009); Doll􀀓ar et al. (2014) are currently the best and fastest for pedestrian detection when combined with 2-depth trees.  \nOther works have concentrated on accelerating detection at test time. Making use of c","cbCaijCvpg4B5vO4","https://ap.wps.com/l/cbCaijCvpg4B5vO4","pdf",1533985,1,16,"English","en",105,"# Introduction\n## Sliding-window detectors and two-stage training\n## Bootstrapping (hard negative mining) and its motivation\n## Prior work on features and faster detection pipelines","[{\"question\":\"How is training performed for sliding-window object detectors in the described approach?\",\"answer\":\"Training uses two stages: first a classifier is trained on positives and randomly sampled negatives, then scenes are scanned to collect hard negatives that the current classifier misclassifies, forming an augmented set for a second training stage.\"},{\"question\":\"What is hard negative mining, and why does it help?\",\"answer\":\"Hard negatives are object-free samples that trigger strongly positive responses from the current classifier (or fall inside an SVM margin). Adding them concentrates learning on difficult cases and improves the decision boundary.\"},{\"question\":\"What inefficiency does the paper address in standard bootstrapping?\",\"answer\":\"Standard bootstrapping densely visits scenes, evaluating many regions even when they are unlikely to contribute useful hard negatives. The paper proposes discarding certain scene regions without additional computation to focus on promising areas.\"}]","Efficient Sample Mining for Object Detection | PDF",1785947712,40,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"efficient-sample-mining-for-object-detection","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/efficient-sample-mining-for-object-detection/128439/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-23","2026-08-05",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"How is training performed for sliding-window object detectors in the described approach?","Question",{"text":76,"@type":77},"Training uses two stages: first a classifier is trained on positives and randomly sampled negatives, then scenes are scanned to collect hard negatives that the current classifier misclassifies, forming an augmented set for a second training stage.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"What is hard negative mining, and why does it help?",{"text":81,"@type":77},"Hard negatives are object-free samples that trigger strongly positive responses from the current classifier (or fall inside an SVM margin). Adding them concentrates learning on difficult cases and improves the decision boundary.",{"name":83,"@type":74,"acceptedAnswer":84},"What inefficiency does the paper address in standard bootstrapping?",{"text":85,"@type":77},"Standard bootstrapping densely visits scenes, evaluating many regions even when they are unlikely to contribute useful hard negatives. The paper proposes discarding certain scene regions without additional computation to focus on promising areas.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":93},[94,98,102,106,111,116,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":46,"category_name":113,"show_sort_weight":114,"slug":115},6,"Technology",50,"technology",{"id":117,"doc_module":4,"doc_module_name":46,"category_name":118,"show_sort_weight":29,"slug":119},7,"Healthcare","healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":107,"slug":138},19,"General","general"]