[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85401-en":3,"doc-seo-85401-105":29,"detail-sidebar-cat-0-en-105":91},{"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":13,"seo_description":14,"update_tm":27,"read_time":28},85401,1099513958607,"Jiven","https://ap-avatar.wpscdn.com/avatar/100002390cf8733938c?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778829742770036399",8,"Research & Report","Cascaded Multi-Scale Attention for Enhanced Multi-Scale Feature Extraction and Interaction with Low-Resolution Images","Real-world image recognition often relies on low-resolution inputs, which makes extracting and using multi-scale features difficult and can degrade inference accuracy in tasks requiring fine spatial detail. This work proposes cascaded multi-scale attention (CMSA) for CNNViT hybrid architectures, enabling multi-scale feature extraction and seamless integration without downsampling inputs or intermediate feature maps. CMSA combines grouped multi-head self-attention with window-based local attention and cascaded fusion across scales, improving performance for pose estimation and related problems while using fewer parameters.","Cascaded Multi-Scale Attention for Enhanced Multi-Scale Feature Extraction and Interaction with Low-Resolution Images  \nXiangyong Lu∗ , Masanori Suganuma∗†, Takayuki Okatani∗†,  \narXiv :2412 .02197v4 [ cs .CV] 13 Jul 2026  \nAbstract—In real-world applications of image recognition tasks, such as human pose estimation, cameras often capture objects, like human bodies, at low resolutions. This scenario poses a challenge in extracting and leveraging multi-scale features, which is often essential for precise inference. To address this challenge, we propose a new attention mechanism, named cascaded multi-scale attention (CMSA), tailored for use in CNNViT hybrid architectures, to handle low-resolution inputs effectively. The design of CMSA enables the extraction and seamless integration of features across various scales without necessitating the downsampling of the input image or feature maps. This is achieved through a novel combination of grouped multi-head self-attention mechanisms with window-based local attention and cascaded fusion of multi-scale features over different scales. This architecture allows for the effective handling of features across different scales, enhancing the model’s ability to perform tasks such as human pose estimation, head pose estimation, and more with low-resolution images. Our experimental results show that the proposed method outperforms existing state-of-the-art methods in these areas with fewer parameters, showcasing its potential for broad application in real-world scenarios where capturing high-resolution images is not feasible. Code is available at [https://github.com/xyongLu/CMSA](https://github.com/xyongLu/CMSA).  \nIndex Terms—Article submission, IEEE, IEEEtran, journal, LATEX, paper, template, typesetting.  \nI. INTRODUCTION  \nIN recent years, the application of deep learning for image  \nrecognition has significantly broadened its scope in the real world. At the forefront, there is the use under tougher conditions for image recognition. One of these is the application to low-resolution images. In this paper, we discuss how to maximize the estimation accuracy for several typical image recognition problems using images of a lower resolution than those primarily considered in previous studies, as illustrated in the left panel of Fig. 1.  \nThere are several scenarios where handling low-resolution images is necessary. One is when ideal image capture conditions are not achievable, resulting in objects not being adequately resolved in images. For instance, this can occur when trying to estimate specific information, such as the pose of a person located at a relative distance from a fixed camera, like a surveillance camera. Another scenario involves conducting image recognition on low-performance edge devices. Being able to make estimations from lower-resolution images could lead to reductions in device costs and computational demands. In this paper, we explore network architectures suitable for the above applications. Specifically, we build upon the recently  \n∗Graduate School of Information Sciences, Tohoku University.†RIKEN Center for AIP  \nFig. 1. Left: Estimating pose from low-resolution images. Left-Upper: human pose estimation from COCO2017 . Left-Lower: head pose estimation from AFLW2000 . Right: resolution-accuracy trade-off for human pose estimation evaluated on COCO 2017 val dataset.  \nevolved CNN-ViT hybrid models to design an architecture that can effectively handle low-resolution inputs. In particular, our goal is to perform multi-scale feature extraction from images and to facilitate appropriate interactions among these features.  \nWhy do we aim to do this? First, multi-scale feature extraction combined with their interactions is an essential component for several tasks. The estimation of human poses mentioned earlier is a prime example of this necessity, as evidenced by previous studies [1]–[3] . While the level of necessity may differ, this strategy is fundamentally beneficial across ","cbCaifplbnCJC83l","https://ap.wps.com/l/cbCaifplbnCJC83l","pdf",2132901,1,12,"English","en",105,"# Introduction\n## Problem: low-resolution inputs\n## Motivation: multi-scale features and interactions\n## Limitation of downsampling for low-resolution inputs\n## Proposed approach (grouped attention + window-based local attention + cascaded fusion)","[{\"question\":\"What problem does the proposed CMSA method address?\",\"answer\":\"CMSA targets the difficulty of extracting and leveraging multi-scale features when images are low-resolution, which hurts accuracy in precision tasks such as pose estimation.\"},{\"question\":\"How does CMSA obtain multi-scale features without downsampling?\",\"answer\":\"CMSA groups multi-head self-attention to process different scales and uses window-based local attention (inspired by Swin Transformer) to extract scale-specific features, then cascades fusion from lower to higher scales via feature transfer between groups.\"},{\"question\":\"Which tasks and outcomes are reported for validating CMSA?\",\"answer\":\"Experiments are reported for human pose estimation and head pose estimation, showing the method outperforms state-of-the-art approaches while achieving better parameter efficiency for low-resolution scenarios.\"}]",1784203136,30,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"cascaded-multi-scale-attention-for-enhanced-multi-scale-feature-extraction-and-interaction-with-low-resolution-images","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/cascaded-multi-scale-attention-for-enhanced-multi-scale-feature-extraction-and-interaction-with-low-resolution-images/85401/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does the proposed CMSA method address?","Question",{"text":75,"@type":76},"CMSA targets the difficulty of extracting and leveraging multi-scale features when images are low-resolution, which hurts accuracy in precision tasks such as pose estimation.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does CMSA obtain multi-scale features without downsampling?",{"text":80,"@type":76},"CMSA groups multi-head self-attention to process different scales and uses window-based local attention (inspired by Swin Transformer) to extract scale-specific features, then cascades fusion from lower to higher scales via feature transfer between groups.",{"name":82,"@type":73,"acceptedAnswer":83},"Which tasks and outcomes are reported for validating CMSA?",{"text":84,"@type":76},"Experiments are reported for human pose estimation and head pose estimation, showing the method outperforms state-of-the-art approaches while achieving better parameter efficiency for low-resolution scenarios.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":28,"slug":121},"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]