[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86444-en":3,"doc-seo-86444-105":30,"detail-sidebar-cat-0-en-105":95},{"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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},86444,5909877438554,"Maeve","https://ap-avatar.wpscdn.com/avatar/5600025385ad2bf12a7?_k=1778553567797529272",8,"Research & Report","A Strong Balanced-Softmax Classifier-Retraining Baseline for Long-Tailed Recognition","Long-tailed recognition approaches often address class imbalance by altering losses, margins, or representations so that frequent head classes do not dominate learning. This work tests whether tail errors that remain after Balanced Softmax training can be reduced through classifier-only retraining. It introduces BS-cRT, a two-stage pipeline that freezes the backbone trained with Balanced Softmax and retrains only a cosine classifier using balanced episodic batches. Experiments on CIFAR-100-LT, CIFAR-10-LT, ImageNet-LT, and Places-LT show consistent Few-shot accuracy improvements over the matched Balanced Softmax checkpoint, including +5.15, +5.83, +6.92, and +9.78 points at IF=100. An analysis of CBRM highlights failure modes and motivates boundary supervision that accounts for class frequency.","A STRONG BALANCED-SOFTMAX CLASSIFIER-RETRAINING BASELINE FOR LONG-TAILED RECOGNITION  \nA PREPRINT  \n Juan R. Terven  \nCICATA Querétaro Instituto Politecnico Nacional Mexico  \n Diana M. Córdova-Esparza  \nFacultad de Informática Universidad Autónoma de Querétaro Mexico  \n Julio Alejandro Romero-Gonzalez  \nFacultad de Informática Universidad Autónoma de Querétaro Mexico  \narXiv :2607 .09832v 1 [ cs .LG] 10 Jul 2026  \n Edgar Arturo Chávez-Urbiola  \nCICATA Querétaro Instituto Politecnico Nacional Mexico  \n Francisco Javier Willars-Rodríguez  \nCICATA Querétaro Instituto Politecnico Nacional Mexico  \n Juan Bautista Hurtado Ramos  \nCICATA Querétaro Instituto Politecnico Nacional Mexico  \n Alfonso Ramirez Pedraza  \nCICATA Querétaro Instituto Politecnico Nacional Mexico  \n Gendry Alfonso Francia  \nEscuela Nacional de Estudios Superiores  \nUniversidad Nacional Autónoma de México  \nMexico  \nJuly 10, 2026  \nABSTRACT  \nLong-tailed recognition methods often modify losses, margins, or representations to reduce the dominance of frequent classes. We ask whether, after Balanced Softmax training, the remaining tail error can be reduced by retraining only the classifier. We evaluate BS-cRT, a two-stage procedure that trains a backbone and cosine classifier with Balanced Softmax, freezes the backbone, and updates only the classifier on balanced episodic batches. The second stage keeps the empirical-prior Balanced Softmax objective and uses raw cosine logits at inference. Across CIFAR-100-LT, CIFAR-10-LT, ImageNet-LT, and Places-LT, this classifier-only step consistently improves Few-shot accuracy over the matched Balanced Softmax checkpoint. At imbalance factor 100, Few-shot gains are +5 .15 points on CIFAR-100-LT and +5 .83 on CIFAR-10-LT; on ImageNet-LT and Places-LT, gains are +6 .92 and +9 .78 points, respectively, with a Top-1/Few-shot trade-off on ImageNet-LT. We also analyze Counterfactual Boundary Risk Minimization (CBRM), a boundary-probe extension using prototypebased features near decision boundaries. CBRM identifies two failure modes: scaled-logit cosine margins destabilize training, and corrected hardest-negative probes remain head-class anchored. The results support BS-cRT as a practical classifier-side baseline and indicate that boundary supervision must account for class frequency.  \nKeywords Long-tailed recognition · Imbalanced classification · Classifier retraining · Balanced Softmax · Few-shot recognition · Decision boundaries · Classifier bias  \n1 Introduction  \nReal-world visual recognition rarely enjoys the balanced label distributions assumed by standard benchmarks. Instead, categories typically follow a long-tailed frequency profile: a small number of head classes dominate the training set, while many tail classes appear only a handful of times. This imbalance is especially problematic because the rare classes are often the most important ones in practice, such as unusual medical findings, rare species, safety-critical failures, or infrequent industrial defects. Models trained naively on such data tend to allocate excessive decision space to head classes, yielding deceptively strong overall accuracy while failing on large parts of the label space (Yang et al., 2022) . A large body of work addresses long-tailed recognition by modifying the training signal induced by observed examples. Re-weighting methods amplify the loss of rare-class samples (Cui et al., 2019); focal-style objectives reduce the contribution of easy examples (Lin et al., 2017); margin and logit-adjustment methods reshape the classifier geometry according to class frequency (Cao et al., 2019; Ren et al., 2020); and decoupled training recipes separate representation learning from classifier learning, often retraining the classifier under a more balanced sampling distribution (Kang et al., 2020) . Among these, Balanced Softmax has become a particularly strong and simple training-time baseline. At the same time, classifier retraining (cRT) has shown that much of","cbCaiitS4KrTQBOA","https://ap.wps.com/l/cbCaiitS4KrTQBOA","pdf",466275,2,1,18,"English","en",105,"# Introduction\n## Problem setting: long-tailed visual recognition\n## Prior approaches and motivation\n## Proposed method: BS-cRT\n## Experimental findings and baseline strength\n## Follow-up: Counterfactual Boundary Risk Minimization (CBRM)","[{\"question\":\"What is BS-cRT, and how does it use Balanced Softmax?\",\"answer\":\"BS-cRT is a two-stage procedure: train a backbone with a cosine classifier using Balanced Softmax, then freeze the backbone and retrain only the cosine classifier using Balanced Softmax on balanced episodic batches.\"},{\"question\":\"What datasets and evaluation settings are used to measure BS-cRT performance?\",\"answer\":\"The paper evaluates on CIFAR-100-LT, CIFAR-10-LT, ImageNet-LT, and Places-LT, reporting both Top-1 and Few-shot accuracy under different imbalance factors.\"},{\"question\":\"How much does classifier-only retraining improve Few-shot accuracy?\",\"answer\":\"Across benchmarks, BS-cRT improves Few-shot accuracy over the matched Balanced Softmax stage-1 checkpoint, with gains including +5.15 (CIFAR-100-LT), +5.83 (CIFAR-10-LT), +6.92 (ImageNet-LT), and +9.78 (Places-LT) at IF=100.\"},{\"question\":\"What issues does the CBRM boundary-probe analysis reveal?\",\"answer\":\"CBRM identifies two failure modes: scaled-logit cosine margins can destabilize training, and corrected hardest-negative probes may remain anchored to head-class behavior, indicating boundary supervision must reflect class frequency.\"}]",1784211776,45,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":90,"head_meta":92,"extra_data":94,"updated_unix":28},"a-strong-balanced-softmax-classifier-retraining-baseline-for-long-tailed-recognition","",{"@graph":36,"@context":89},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/a-strong-balanced-softmax-classifier-retraining-baseline-for-long-tailed-recognition/86444/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-24","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81,85],{"name":72,"@type":73,"acceptedAnswer":74},"What is BS-cRT, and how does it use Balanced Softmax?","Question",{"text":75,"@type":76},"BS-cRT is a two-stage procedure: train a backbone with a cosine classifier using Balanced Softmax, then freeze the backbone and retrain only the cosine classifier using Balanced Softmax on balanced episodic batches.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What datasets and evaluation settings are used to measure BS-cRT performance?",{"text":80,"@type":76},"The paper evaluates on CIFAR-100-LT, CIFAR-10-LT, ImageNet-LT, and Places-LT, reporting both Top-1 and Few-shot accuracy under different imbalance factors.",{"name":82,"@type":73,"acceptedAnswer":83},"How much does classifier-only retraining improve Few-shot accuracy?",{"text":84,"@type":76},"Across benchmarks, BS-cRT improves Few-shot accuracy over the matched Balanced Softmax stage-1 checkpoint, with gains including +5.15 (CIFAR-100-LT), +5.83 (CIFAR-10-LT), +6.92 (ImageNet-LT), and +9.78 (Places-LT) at IF=100.",{"name":86,"@type":73,"acceptedAnswer":87},"What issues does the CBRM boundary-probe analysis reveal?",{"text":88,"@type":76},"CBRM identifies two failure modes: scaled-logit cosine margins can destabilize training, and corrected hardest-negative probes may remain anchored to head-class behavior, indicating boundary supervision must reflect class frequency.","https://schema.org",{"og:url":51,"og:type":91,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":93,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":96},[97,101,105,109,114,119,124,127,132,135,139],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":106,"show_sort_weight":107,"slug":108},"Exam",70,"exam",{"id":110,"doc_module":4,"doc_module_name":46,"category_name":111,"show_sort_weight":112,"slug":113},5,"Comic",60,"comic",{"id":115,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},6,"Technology",50,"technology",{"id":120,"doc_module":4,"doc_module_name":46,"category_name":121,"show_sort_weight":122,"slug":123},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":125,"slug":126},30,"research-report",{"id":128,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":130,"slug":131},9,"Religion & Spirituality",20,"religion-spirituality",{"id":130,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":130,"slug":134},"World Cup","world-cup",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":136,"slug":138},10,"Lifestyle","lifestyle",{"id":140,"doc_module":4,"doc_module_name":46,"category_name":141,"show_sort_weight":110,"slug":142},19,"General","general"]