[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82054-en":3,"doc-seo-82054-105":29,"detail-sidebar-cat-0-en-105":90},{"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":4,"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},82054,7971461740909,"Levi","https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d",8,"Research & Report","Multi-Conditioned Diffusion Synthesis of Sand Boils for Low-Resource Earthen-Levee Inspection","Sand boils on earthen levees are safety-critical defects, but pixel-level detection from inspection imagery is constrained by scarce annotated samples. A diffusion-based synthesis pipeline addresses this low-resource regime using a Stable Diffusion XL backbone fine-tuned with DreamBooth on a small curated reference set and multibranch ControlNet conditioning. Soft-mask inpainting preserves defect pixels while re-rendering context, minimizing seams and color shifts versus prior compositing. A Prompt Atlas expands domain specifications into a CLIP-validated prompt bank, producing 1,020 candidates and 815 CLIP-admissible augmentations. Experiments assess quality, fidelity/diversity, distribution drift, and memorisation; downstream segmentation is deferred.","Multi-Conditioned Diffusion Synthesis of Sand Boils for Low-Resource Earthen-Levee Inspection  \nPadam Jung Thapa 1 , Abdullah Bin Naeem2 , Ayon Dey2 , Anav Katwal2 , Md Tamjidul Hoque2,*  \n1 University of Louisiana at Lafayette, Lafayette, LA, USA  \n2 Department of Computer Science, Louisiana State University New Orleans, New Orleans, LA, USA  \nEmails: [tpadamjung@gmail.com](tpadamjung@gmail.com) ; {aneem, adey, akatwal, [thoque}@lsuneworleans.edu](thoque}@lsuneworleans.edu)  \n* Corresponding author: [thoque@lsuneworleans.edu](thoque@lsuneworleans.edu)  \narXiv :2607 .08794v 1 [ cs .GR] 8 Jul 2026  \nAbstract—Sand boils on earthen levees are safetycritical defects, yet pixel-level detection from inspection imagery is limited by the scarcity of annotated examples. We address this low-resource setting with a diffusion-based synthesis pipeline. A Stable Diffusion XL backbone, fine-tuned with DreamBooth on a small curated reference set and conditioned by a multibranch ControlNet stack, generates synthetic sand-boil imagery. A soft-mask inpainting protocol preserves the real defect region pixel-for-pixel while re-rendering the surrounding scene, avoiding the seams and colour casts of the seamless-cloning compositing used previously. A complementary mask-conditioned ControlNet generates a fresh boil into a chosen mask, so the conditioning mask is the segmentation label by construction, with the Convex Hull Annotator repurposed asa drift-checking quality gate—though certifying that label at scale is not yet solved by the available realtrained gate, so we release the soft-mask preset, not this one, as the default. Text conditioning is supplied by a taxonomy-driven Prompt Atlas that expands a single domain specification into a stratified, CLIPvalidated prompt bank and ports to new defect classes without code changes. From the real training images the pipeline produces 1 ,020 synthetic candidates, of which 815 pass a CLIP admissibility filter to form the augmented dataset. We evaluate image quality with distribution and fidelity/diversity measures against the real reference set and a Poisson baseline, and audit the augmented set for out-of-distribution drift and memorisation. No single preset dominates: the presets trade off fidelity, diversity, and label reliability, so we release the label-reliable preset as the default and treat a curated mixture as the natural augmentation set. We scope our claims to image quality, label provenance, and diversity; downstream segmentation is left to future work. Code and an artefact manifest are released for reproducibility.  \nIndex Terms—Sand boil detection, earthen levees, synthetic image generation, Stable Diffusion XL, DreamBooth, ControlNet, soft-mask inpainting, prompt curation, low-resource computer vision, civil infrastructure.  \nI. Introduction  \nEarthen levees protect large populations and property from flooding, yet their condition is hard to monitor at scale. The cost of inadequate inspection became clear  \nduring the 2005 Hurricane Katrina failures, when independent investigators attributed the catastrophic flooding in New Orleans to a combination of design, construction, and inspection deficiencies [1] . Among the defect classes that inspectors are trained to recognise, sand boils are particularly consequential. A sand boil forms when water under hydraulic pressure carries fine sand upward through a permeable foundation layer, producing a small saturated dome at the surface (Figure 1) . This is the early visible stage of internal erosion (also called piping), which, if left untreated, can progress to a sudden breach during a highwater event. Detecting and segmenting sand boils at the pixel level from routine field photographs therefore offers a direct lever for proactive flood-risk management.  \nTraining a deep segmentation model for this task runs into a stubborn obstacle: annotated sand boil imagery is rare. The defect appears only during high water, is spread across long linear levee n","cbCaigIMf4Zsxt93","https://ap.wps.com/l/cbCaigIMf4Zsxt93","pdf",26234278,1,29,"English","en",105,"# Introduction\n## Motivation: sand boils and the annotation scarcity\n## Approach overview: diffusion-based low-resource augmentation\n## Core problems and proposed solutions","[{\"question\":\"Why is sand boil pixel-level detection difficult in practice?\",\"answer\":\"Annotated sand boil imagery is rare because the defect appears only during high water, spans long levee networks, and requires hydrologic expertise for accurate pixel-level annotation.\"},{\"question\":\"How does the proposed pipeline generate synthetic sand-boil images?\",\"answer\":\"It fine-tunes a Stable Diffusion XL backbone with DreamBooth on a small curated reference set and applies multibranch ControlNet conditioning, then uses soft-mask inpainting to preserve real defect pixels while re-rendering surrounding context.\"},{\"question\":\"How is the quality and usefulness of the synthetic dataset evaluated?\",\"answer\":\"Image quality is assessed with distribution and fidelity/diversity measures against real references and a Poisson baseline, and the augmented set is audited for out-of-distribution drift and memorisation; label reliability is also considered when selecting the default preset.\"}]",1784177854,73,{"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":85,"head_meta":87,"extra_data":89,"updated_unix":27},"multi-conditioned-diffusion-synthesis-of-sand-boils-for-low-resource-earthen-levee-inspection","",{"@graph":35,"@context":84},[36,53,67],{"@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/multi-conditioned-diffusion-synthesis-of-sand-boils-for-low-resource-earthen-levee-inspection/82054/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"Why is sand boil pixel-level detection difficult in practice?","Question",{"text":74,"@type":75},"Annotated sand boil imagery is rare because the defect appears only during high water, spans long levee networks, and requires hydrologic expertise for accurate pixel-level annotation.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does the proposed pipeline generate synthetic sand-boil images?",{"text":79,"@type":75},"It fine-tunes a Stable Diffusion XL backbone with DreamBooth on a small curated reference set and applies multibranch ControlNet conditioning, then uses soft-mask inpainting to preserve real defect pixels while re-rendering surrounding context.",{"name":81,"@type":72,"acceptedAnswer":82},"How is the quality and usefulness of the synthetic dataset evaluated?",{"text":83,"@type":75},"Image quality is assessed with distribution and fidelity/diversity measures against real references and a Poisson baseline, and the augmented set is audited for out-of-distribution drift and memorisation; label reliability is also considered when selecting the default preset.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"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":105,"slug":137},19,"General","general"]