[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86131-en":3,"doc-seo-86131-105":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":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},86131,962075114765,"Quinn","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","CUST Clustered Unit-level Similarity Transformer for Lightweight Image Super-Resolution","CUST (Clustered Unit-level Similarity Transformer) is proposed for lightweight single-image super-resolution, targeting the efficiency bottlenecks of Vision Transformer models whose quadratic complexity with image resolution leads to high latency and large memory use. CUST combines CANA to cluster and attend to semantically similar patches across widened regions while keeping practical computation, and MEDA to extract multi-frequency error signals for improved high-frequency detail restoration. Experiments show faster inference and lower GPU memory under realistic constraints. Code is released at the provided repository.","arXiv :2607 . 11088v1 [ cs .CV] 13 Jul 2026  \nCUST : Clustered Unit-level Similarity Transformer for Lightweight Image Super-Resolution  \nJeongsoo Kim 1   \nIndependent Researcher  \n[jeongskim512@gmail.com](jeongskim512@gmail.com)  \nAbstract. Recently, Vision Transformer (ViT)-based models have exhibited remarkable performance in image super-resolution. However, the quadratic computational complexity of ViTs with respect to spatial resolution severely constrains their efficiency, leading to high latency and massive memory consumption. To alleviate this, various window-based attention mechanisms have been proposed; yet, they inherently compromise the long-range dependency modeling that is the primary advantage of ViTs. To overcome these limitations, we propose the Clustered Unitlevel Similarity Transformer (CUST), a novel architecture that efficiently integrates global and local information. Specifically, CUST enables each patch to aggregate and attend to similar patches within a broadened regional scope outside its local window, thereby capturing extensive contextual understanding. Furthermore, it employs overlapping attention windows to capture local dependencies, while explicitly extracting highfrequency details by computing the residual difference between the original features and their downsampled-upsampled counterparts. Comprehensive experiments demonstrate that our proposed model achieves a practical balance between computational efficiency and restoration performance. It achieves a lower memory footprint and faster inference speed compared to recent global context or lightweight models under realistic constraints. Code is available at [https://github.com/jwgdmkj/CUST](https://github.com/jwgdmkj/CUST).  \nKeywords: Single Image Super-Resolution · Vision Transformer · Efficient Image Super-Resolution  \n1 Introduction  \nSingle Image Super-Resolution (SISR), which aims to reconstruct a HR (HighResolution) image from a given LR (Low-Resolution) counterpart, is a fundamental and actively researched area in computer vision. SISR has been widely applied in various real-world scenarios, ranging from medical imaging requiring precise diagnosis and digital photography for restoring old photos to high-magnification digital zoom in smartphones [14, 24] .  \nSince the pioneering work of SRCNN [8], which first introduced Convolutional Neural Networks (CNNs) to SISR, numerous CNN-based models have  \n2 J. Kim  \nPerformance vs. Latency vs. GPU Memory  \n120 140 160 180 200 220 240 Latency (ms)  \nFig. 1: Efficiency-Performance trade-off analysis on Urban100 (×4) .  \nbeen proposed [7, 20, 28, 58] . However, CNNs suffer from a fundamental limitation regarding their restricted receptive fields, making it difficult to capture longrange dependencies effectively. Although various approaches have attempted to overcome this issue by stacking deeper layers [26,31,59], this strategy often leads to increased computational costs and model complexity [42] .  \nFollowing the remarkable success of the attention mechanism in Natural Language Processing (NLP) [46], various models adopting this paradigm have emerged in the computer vision domain [10, 11, 35, 47, 50] . The attention mechanism effectively captures long-range dependencies, addressing the shortcomings of CNNs. However, Vision Transformers (ViTs) suffer from quadratic computational complexity relative to input resolution [5, 6, 21, 55] . To alleviate this, Window-based Self-Attention (WSA) divides images into fixed-size windows [30, 62]; however, this restricts the receptive field and undermines the capture of long-range dependencies [4, 19, 25, 29] . Subsequent attempts to mitigate these limitations—such as generating super-tokens [34, 52] or employing large-window lightweighting strategies [36, 44, 54]—introduce new bottlenecks. Super-token methods are often hindered by slow iterative generation or memory-intensive similarity calculations, while many lightweight models prioritize parameter or FLOP","cbCaimTFd8nL30Yj","https://ap.wps.com/l/cbCaimTFd8nL30Yj","pdf",5098186,6,1,28,"English","en",105,"# 1 Introduction\n## Single Image Super-Resolution and existing CNN/Transformer limits\n## Motivation for resource-aware efficient SR\n# 2 Proposed Method: CUST\n## Cross-window Affinity Neighbor Attention (CANA)\n## Multi-frequency Errordriven Dense Attention (MEDA)\n# Efficiency-Performance Trade-off","[{\"question\":\"Why do Vision Transformer models face efficiency issues in image super-resolution?\",\"answer\":\"Their attention cost grows quadratically with spatial resolution, causing high latency and large memory consumption. This limits deployment efficiency for super-resolution tasks.\"},{\"question\":\"How does CUST capture long-range context while remaining lightweight?\",\"answer\":\"CUST’s CANA clusters and attends to semantically similar patches beyond local window boundaries using affinities to pooled window tokens. This broadens effective context without fully global attention.\"},{\"question\":\"What role does MEDA play in restoring image details?\",\"answer\":\"MEDA extracts multi-frequency error signals from multi-scale feature differences, using dilated convolutions to guide overlap window attention toward salient structures such as edges. This improves local fidelity and high-frequency detail reconstruction.\"}]",1784208756,71,{"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":87,"head_meta":89,"extra_data":91,"updated_unix":28},"cust-clustered-unit-level-similarity-transformer-for-lightweight-image-super-resolution","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"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/cust-clustered-unit-level-similarity-transformer-for-lightweight-image-super-resolution/86131/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"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","2026-07-27","2026-07-16",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},"Why do Vision Transformer models face efficiency issues in image super-resolution?","Question",{"text":76,"@type":77},"Their attention cost grows quadratically with spatial resolution, causing high latency and large memory consumption. This limits deployment efficiency for super-resolution tasks.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does CUST capture long-range context while remaining lightweight?",{"text":81,"@type":77},"CUST’s CANA clusters and attends to semantically similar patches beyond local window boundaries using affinities to pooled window tokens. This broadens effective context without fully global attention.",{"name":83,"@type":74,"acceptedAnswer":84},"What role does MEDA play in restoring image details?",{"text":85,"@type":77},"MEDA extracts multi-frequency error signals from multi-scale feature differences, using dilated convolutions to guide overlap window attention toward salient structures such as edges. This improves local fidelity and high-frequency detail reconstruction.","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":25},{"code":4,"msg":5,"data":93},[94,98,102,106,111,115,120,123,128,131,135],{"id":21,"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":20,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"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"]