[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85733-en":3,"doc-seo-85733-105":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},85733,5909877438554,"Maeve","https://ap-avatar.wpscdn.com/avatar/5600025385ad2bf12a7?_k=1778553567797529272",8,"Research & Report","TDSal: A Task-Based Top-Down Saliency Prediction Model","Visual saliency predicts image regions likely to attract human gaze, but real observers often follow explicit task goals rather than free-viewing assumptions. TDSal addresses this by generating task-dependent saliency maps from natural-language task descriptions, conditioning attention shifts on sentence-level semantic embeddings fused with spatial visual features. Quantitative and qualitative evaluations show that incorporating explicit task semantics better models goal-directed fixation behavior and improves alignment with observed attention patterns.","arXiv :2607 .09827v1 [ cs .CV] 10 Jul 2026  \nTDSal: A Task-Based Top-Down Saliency Prediction Model  \nCan Mizrakli 1 ,2[0009−0000−1990−3996] and Tolga K. Capin2[0000−0002−7843−6336]  \n1 Karlsruhe Institute of Technology, Germany  \n2 TED University, Türkiye  \n[can.mizrakli@student.kit.edu](can.mizrakli@student.kit.edu) , [tolga.capin@tedu.edu.tr](tolga.capin@tedu.edu.tr)  \nAbstract. Visual saliency aims to predict the regions of an image most likely to attract human visual attention. While most saliency models assume free-viewing conditions, human attention is often shaped by explicit task goals. In this work, we address task-driven saliency prediction by proposing a model that conditions visual attention on natural-language task descriptions. The model produces task-dependent saliency maps that reflect how attention shifts under different viewing intents. Through quantitative and qualitative analysis, we show that incorporating explicit task semantics enables more faithful modeling of goal-directed visual attention.  \nKeywords: Visual Saliency · Top-Down Saliency · Task-Based Saliency  \n· Visual Attention · Gaze  \n1 Introduction  \nVisual saliency aims to predict where humans attend in an image, often modeled as a spatial probability map of fixation likelihood. Classical bottom-up approaches rely on low-level cues such as contrast, color, and orientation under free-viewing assumptions [6, 5] . However, in many real-world settings attention is goal-directed, and task demands can substantially shift where observers look.  \nTask-conditioned saliency prediction is becoming increasingly important in domains such as autonomous driving, human–robot interaction, and assistive vision. However, existing approaches often fall short in two critical aspects: (1) they do not explicitly incorporate high-level task semantics into the saliency generation process, and (2) they lack a lightweight mechanism for combining such semantic cues with spatial visual features. Approaches based on complex transformer or diffusion architectures [14, 23] can suffer from reduced spatial resolution and high computational cost, whereas methods without explicit task conditioning may fail to capture goal-directed human attention patterns.3  \nTo address these limitations, we structure our study around the following research questions:  \n3 Supplementary material associated with this work is publicly available through the project repository: [https://github.com/canmizrakli/TDSal-2026](https://github.com/canmizrakli/TDSal-2026)  \n2 C. Mizrakli and T. K. Capin  \nFig. 1. Conceptual overview of task-driven saliency prediction. Different task prompts lead to distinct human fixation patterns on the same image. We model this by fusing visual features extracted from the image with semantic representations of the task prompt, producing a task-based saliency map that reflects where a viewer would look given a specific goal.  \n– RQ1: How can textual task definitions, encoded through sentence-level embeddings, be fused with spatial visual features to guide attention maps?  \n– RQ2: Can explicit vision–language conditioning support fixation alignment in task-driven saliency prediction?  \n– RQ3: How can a pretrained object-detection backbone be leveraged to extract spatially rich visual features that support task-oriented saliency modeling?  \nFigure 1 provides an overview of our task-driven formulation. The model combines (i) spatial visual features extracted from the input image and (ii) a semantic representation of the task prompt, and outputs a task-conditioned saliency map. This schematic summarizes the core fusion pathway that is detailed in Section 3 .  \n1.1 Contributions  \nOur work introduces TDSal, a modular vision-language architecture for taskdriven saliency prediction. Building upon the research questions outlined above, our main contributions are as follows:  \n– We introduce TDSal, a modular top-down saliency prediction network that fuses YOLO-derived spatial visual fe","cbCaisw4tkIE07sY","https://ap.wps.com/l/cbCaisw4tkIE07sY","pdf",6062558,1,15,"English","en",105,"# Introduction\n## Contributions\n# Related Work\n## Bottom-Up vs. Task-Driven Saliency Models","[{\"question\":\"What problem does TDSal address in visual saliency prediction?\",\"answer\":\"TDSal targets the gap between free-viewing saliency models and real human behavior, where attention is guided by explicit task goals.\"},{\"question\":\"How does TDSal incorporate task information into saliency maps?\",\"answer\":\"It encodes the textual task prompt with sentence-level embeddings and fuses these semantic representations with spatial visual features to produce task-conditioned saliency maps.\"},{\"question\":\"What model components and evaluation approach does TDSal use?\",\"answer\":\"TDSal uses a modular vision-language architecture that combines YOLO-derived spatial features with Sentence-BERT task embeddings via a transformer-based fusion block, and evaluates using seven saliency metrics plus qualitative visualizations.\"}]",1784205886,38,{"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},"tdsal-a-task-based-top-down-saliency-prediction-model","",{"@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/tdsal-a-task-based-top-down-saliency-prediction-model/85733/",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 TDSal address in visual saliency prediction?","Question",{"text":75,"@type":76},"TDSal targets the gap between free-viewing saliency models and real human behavior, where attention is guided by explicit task goals.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does TDSal incorporate task information into saliency maps?",{"text":80,"@type":76},"It encodes the textual task prompt with sentence-level embeddings and fuses these semantic representations with spatial visual features to produce task-conditioned saliency maps.",{"name":82,"@type":73,"acceptedAnswer":83},"What model components and evaluation approach does TDSal use?",{"text":84,"@type":76},"TDSal uses a modular vision-language architecture that combines YOLO-derived spatial features with Sentence-BERT task embeddings via a transformer-based fusion block, and evaluates using seven saliency metrics plus qualitative visualizations.","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},[93,97,101,105,110,115,120,123,128,131,135],{"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":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]