[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-327162-105":59,"doc-detail-327162-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","procal-inference-time-proposal-calibration-for-open-vocabulary-object-detection","ProCal - Inference-Time Proposal Calibration for Open-Vocabulary Object Detection","","Open-vocabulary object detection localizes and classifies objects beyond the fixed training categories. Recent approaches use a frozen vision-language model (VLM) as a detector backbone, but VLM scores do not reliably reflect an object’s position and scale within the image. Observing that pretrained VLMs can distinguish foreground and background regions, ProCal introduces an inference-time proposal calibration that reranks proposals using localization-aware foreground scoring and background-aware suppression. Experiments on CLIPSelf ViT-L/14 improve APr by 2.5 on OV-LVIS, mitigating novel-object miscalibration.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/procal-inference-time-proposal-calibration-for-open-vocabulary-object-detection/327162/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/procal-inference-time-proposal-calibration-for-open-vocabulary-object-detection/327162.png","ImageObject",300,407,{"name":92,"@type":93},"Ivy","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-29","2026-09-21",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",14,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does ProCal address in open-vocabulary object detection?","Question",{"text":113,"@type":114},"It addresses miscalibration where existing frozen-VLM detectors fail to properly rank proposals for novel objects, often suppressing true novel-object proposals due to base-class-biased training.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How does ProCal compute its proposal calibration signal?",{"text":118,"@type":114},"ProCal combines two scores: a localization-aware foreground score that estimates whether a proposal contains an object area, and a background-aware suppression score that measures how much a proposal resembles background.",{"name":120,"@type":111,"acceptedAnswer":121},"What experimental outcome does ProCal achieve?",{"text":122,"@type":114},"Applied to CLIPSelf ViT-L/14, ProCal improves APr by 2.5 on OV-LVIS through proposal-level reranking effects that mitigate ranking miscalibration.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},327162,1790375268,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":29,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":144,"read_time":145,"preview_page_count":4},549758252649,"https://ap-avatar.wpscdn.com/avatar/8000253669c5317157?_k=1778319167496531819","ProCal: Inference-Time Proposal Calibration for Open-Vocabulary  \nObject Detection  \nJae-Ryung Hong 1 , Ho-Joong Kim 1 , and Seong-Whan Lee 1  \narXiv :2607 .0 1759v 1 [ cs .CV] 2 Jul 2026  \nAbstract—Open-vocabulary object detection aims to localize and classify objects beyond the fixed set of categories seen during training. Recent open-vocabulary object detection methods improve localization and classification for unseen categories by leveraging a frozen VLM as a detector backbone. However, VLM classification score lacks recognizing position and scale of the object in an image. We observe that pretrained VLMs enable to classify foreground and background regions. According to this observation, we propose a simple inference-time Proposal Calibration (ProCal) that improves localization quality of the classification score. ProCal computes a proposal prior by combining two scores: localization-aware foreground score and background-aware suppression score. Localization-aware foreground score captures whether a proposal contains an object area. Background-aware suppression score measures the extent to which the proposal resembles background. We analyze that ProCal suppresses false novel activation on background proposals and consistently ranks true novel proposals above background and partial novel proposals. Applied to CLIPSelf ViT-L/14, ProCal improves APr +2.5 on OV-LVIS. The analyses show that proposal-level localization-aware reranking effects to mitigate ranking miscalibration for novel objects.  \nI. INTRODUCTION  \nCategory-level generalization has been studied across zero-shot and transfer-learning recognition settings [1], [2],[3] . Open-vocabulary object detection (OVD) instantiates this issue in object detection, where detectors trained with basecategory annotations are required to localize and classify categories that are not annotated during detector training. By leveraging text embeddings from pretrained vision-language models (VLMs) such as CLIP [4], OVD enables the localization and classification of unseen objects. Through pretrained image-text alignment, the detector associates object regions with semantic class descriptions. Transferring this image-level alignment of pretrained VLMs to region-level recognition for object detection remains challenging.  \nPrevious methods [5], [6], [7], [8] adapt vision-language models to detection through additional pretraining, transfer learning, or fine-tuning to improve region-text alignment. FVLM [9] incorporates the image encoder of a pretrained VLM into the detector backbone, motivated by the observation that pretrained VLM features capture object boundaries. Building on this idea, recent open-vocabulary detectors [10],  \n*This work was supported by the Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. RS-2019-II190079, Artificial Intelligence Graduate School Program, Korea University), and by the Information Technology Research Center (ITRC) support program (No. IITP-2026-RS- 2024-00436857) .  \n1J.-R. Hong, H.-J. Kim, and S.-W. Lee are with the Department of Artificial Intelligence, Korea University, Anam-dong, Seongbuk-ku, Seoul 02841, Korea. {h   jr, hojoong kim, [sw.lee](sw.lee}@korea.ac.kr)[}](sw.lee}@korea.ac.kr)[@korea.ac.kr](sw.lee}@korea.ac.kr)  \n“a photo of a background” “a photo of an object in the background”  \n“a photo of an object in the background”  \n- “a photo of an object”  \n(a) (b) (c)  \nFig. 1: Qualitative results of similarities computed by three different prompt strategies. Each visualized heatmap highlights the similarities between cropped images and classagnostic prompts. (a): leveraging the background prompt highlights the background area. (b): prompt including both object and background highlights boundary regions of the object. (c): leveraging the difference between background and object prompts yields object-containing regions.  \n[11] follow this approach that freezes a p","cbCaiis7b4J4YbC9","https://ap.wps.com/l/cbCaiis7b4J4YbC9","pdf",726601,"English","# Abstract\n# Introduction\n## Category-level generalization in open-vocabulary detection\n## Limitations of existing frozen-VLM detectors\n## ProCal motivation from foreground/background recognition","[{\"question\":\"What problem does ProCal address in open-vocabulary object detection?\",\"answer\":\"It addresses miscalibration where existing frozen-VLM detectors fail to properly rank proposals for novel objects, often suppressing true novel-object proposals due to base-class-biased training.\"},{\"question\":\"How does ProCal compute its proposal calibration signal?\",\"answer\":\"ProCal combines two scores: a localization-aware foreground score that estimates whether a proposal contains an object area, and a background-aware suppression score that measures how much a proposal resembles background.\"},{\"question\":\"What experimental outcome does ProCal achieve?\",\"answer\":\"Applied to CLIPSelf ViT-L/14, ProCal improves APr by 2.5 on OV-LVIS through proposal-level reranking effects that mitigate ranking miscalibration.\"}]","ProCal - Inference-Time Proposal Calibration for Open-Vocabulary Object Detection | PDF",1789976571,15]