[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81516-en":3,"doc-seo-81516-105":30,"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":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},81516,1099513958762,"Logic","https://ap-avatar.wpscdn.com/avatar/1000023916a998db790?x-image-process=image/resize,m_fixed,w_180,h_180&k=1784791008015729253",8,"Research & Report","Looking Locally: Object-Centric Vision Transformers as Foundation Models for Efficient Segmentation","Current state-of-the-art segmentation models encode whole images before focusing on specific objects, wasting computation. FLIP (Fovea-Like Input Patching) is a parameter-efficient vision foundation model for object-centric segmentation using biologically inspired top-down attention. It samples multi-resolution, object-centered patches, allocating high-resolution processing to object centers while preserving coarse peripheral context. FLIP substantially outperforms SAM2 variants on six benchmarks with far fewer parameters, and accurately segments extremely small objects on ObjaScale.","Looking Locally: Object-Centric Vision Transformers as Foundation Models for Efficient Segmentation  \nManuel Traub 1 Martin V. Butz 1  \narXiv :2502 .02763v3 [ cs .CV] 10 Jul 2026  \nAbstract  \nCurrent state-of-the-art segmentation models encode entire images before focusing on specific objects. This wastes computational resources. We introduce FLIP (Fovea-Like Input Patching), a parameter-efficient vision model that realizes object segmentation through biologically-inspired top-down attention. FLIP selectively samples multi-resolution patches centered on objects of interest from the input. As a result, it allocates high-resolution processing to object centers while maintaining coarser peripheral context. This off-grid, scale-invariant design enables FLIP to outperform META’s Segment Anything models (SAM, SAM2 and fast variants) by large margins: With more than 440 × fewer parameters, FLIP-Tiny (0.51M parameters) reaches a mean IoU of 79.90% while SAM2-L reaches 75.87% IoU (224.45M parameters) . FLIP-Large even achieves 83.26% mean IoU (96.6M parameters), still running about 2 × faster than SAM2-L. We evaluate on six benchmarks in total. In five established benchmarks (Hypersim, KITTI-360, OpenImages, COCO, LVIS) FLIP consistently outperforms SAM and various variants of it. In our novel ObjaScale dataset, which stress-tests scale invariance with objects ranging from 0.0001% up to 25% of the image area, we show that FLIP segments even very small objects accurately, where existing models fail severely. FLIP opens new possibilities for real-time, object-centric vision and offers much higher energy efficiency. We believe that FLIP can act as a powerful foundation model, as it is very well-suited to track objects over time, for example, when being integrated into slot-based scene segmentation architectures.  \nProject Page: [https://cognitivemodeling.github.io/FLIP/](https://cognitivemodeling.github.io/FLIP/)[ ](https://cognitivemodeling.github.io/FLIP/)[1](1Neuro-Cognitive Modeling Group)[Neuro-Cognitive Modeling Group](1Neuro-Cognitive Modeling Group), [University of T](University of T)¨ubingen, Germany. Correspondence to: \u003C[manuel.traub@uni-tuebingen.de](manuel.traub@uni-tuebingen.de) >.  \nAccepted at ICML 2026.  \n1. Introduction  \nObject-centric models have emerged as a powerful paradigm for structured perception in visual tasks. They offer the potential to represent complex scenes in a more interpretable and compositional manner. While traditional architectures such as Convolutional Neural Networks (CNNs) (Liu et al., 2022) and Vision Transformers (ViTs) (Dosovitskiy et al., 2020) have demonstrated impressive performance on largescale datasets, they often lack the nuanced object-level understanding required for robust scene parsing. Furthermore, these models typically require massive amounts of labeled data and exhibit vulnerabilities to adversarial perturbations.  \nRecent advances in object-centric learning, such as Slot Attention (Locatello et al., 2020), have sought to address these challenges by enabling the model to discover and represent objects within a scene as distinct entities. Models like  \nMean_IoU  \n80  \n75  \n\n|  |  |  | FLIP−Mi |  |  |  |  FLIP−Large |  |\n| --- | --- | --- | --- | --- | --- | --- | --- | --- |\n|  | FLIP−Small |  |  |  |  |  |  |  |\n| FLIP−Tiny |  |  |  |  | 440.1x fe |  | wer params |  |\n|  |  |  |  | SAM2−S |  | SA\u003Cbr>\u003Cbr> | M2−B+\u003Cbr>\u003Cbr>\u003Cbr>SA | M2−L\u003Cbr>SAM−H\u003Cbr> |\n|  |  |  |  | SAM2\u003Cbr>\u003Cbr>\u003Cbr>EfficientSAM−S |  |  | SAM−B | SAM−L |\n|  |  | Eff | \u003Cbr>icientSAM−T | \u003Cbr> MobileSAM |  |  |  |  |\n\nddle  \n1 10 100  \nParams_M  \nFigure 1. Comparison of mean IoU across six diverse datasets (Hypersim, KITTI-360, OpenImages, COCO, LVIS, ObjaScale) plotted against model size (in millions of parameters) . FLIP variants consistently outperform SAM variants while using several orders of magnitude fewer parameters.  \n Looking Locally: Object-Centric Vision Transformers   \n|  | Encoder module |  |\n| --- | --- | -","cbCaivlxVSx33CO7","https://ap.wps.com/l/cbCaivlxVSx33CO7","pdf",4392579,4,1,20,"English","en",105,"# Introduction\n## Problem with current segmentation models\n## Object-centric learning background\n## Segment Anything (SAM) and SAM2 overview\n## Limitations of existing models","[{\"question\":\"What computational inefficiency do current segmentation models have?\",\"answer\":\"They encode entire images first, then focus on specific objects, which wastes resources when only parts of the image are relevant.\"},{\"question\":\"How does FLIP achieve object-centric segmentation more efficiently?\",\"answer\":\"FLIP uses Fovea-Like Input Patching to selectively sample multi-resolution patches centered on objects of interest, then processes them with top-down attention to produce instance segmentation with pixel-level precision.\"},{\"question\":\"How does FLIP compare to SAM and SAM2 variants in performance and model size?\",\"answer\":\"FLIP variants achieve higher mean IoU while using far fewer parameters; for example, FLIP-Tiny reaches 79.90% mean IoU versus SAM2-L at 75.87% with much larger parameter counts.\"}]",1784173945,50,{"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":86,"head_meta":88,"extra_data":90,"updated_unix":28},"looking-locally-object-centric-vision-transformers-as-foundation-models-for-efficient-segmentation","",{"@graph":36,"@context":85},[37,53,68],{"@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":20},"https://docshare.wps.com/document/looking-locally-object-centric-vision-transformers-as-foundation-models-for-efficient-segmentation/81516/",{"url":52,"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],{"name":72,"@type":73,"acceptedAnswer":74},"What computational inefficiency do current segmentation models have?","Question",{"text":75,"@type":76},"They encode entire images first, then focus on specific objects, which wastes resources when only parts of the image are relevant.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does FLIP achieve object-centric segmentation more efficiently?",{"text":80,"@type":76},"FLIP uses Fovea-Like Input Patching to selectively sample multi-resolution patches centered on objects of interest, then processes them with top-down attention to produce instance segmentation with pixel-level precision.",{"name":82,"@type":73,"acceptedAnswer":83},"How does FLIP compare to SAM and SAM2 variants in performance and model size?",{"text":84,"@type":76},"FLIP variants achieve higher mean IoU while using far fewer parameters; for example, FLIP-Tiny reaches 79.90% mean IoU versus SAM2-L at 75.87% with much larger parameter counts.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,114,119,122,126,129,133],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":29,"slug":113},6,"Technology","technology",{"id":115,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":22,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":127,"show_sort_weight":22,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":46,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":46,"category_name":135,"show_sort_weight":106,"slug":136},19,"General","general"]