[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-665356-105":3,"doc-detail-665356-en":81,"detail-sidebar-cat-0-en-105":98},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","flowgen-synthesizing-diverse-flowcharts-to-enhance-and-benchmark-mllm-reasoning","FLOWGEN - Synthesizing Diverse Flowcharts to Enhance and Benchmark MLLM Reasoning","","Flowcharts are widely used to depict processes and relationships, yet accurate interpretation is difficult because diagrams vary greatly in structure and visual rendering. Existing datasets provide limited control over properties like complexity and style, constraining training and evaluation for multimodal large language models (MLLMs). FlowGen introduces a controllable synthesizer generating diverse flowcharts with customizable structural features and multiple renderer backends. Experiments show improved parsing and question answering, better generalization, and challenging tests that reveal consistent MLLM weaknesses in high complexity and varied styles.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/flowgen-synthesizing-diverse-flowcharts-to-enhance-and-benchmark-mllm-reasoning/665356/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/flowgen-synthesizing-diverse-flowcharts-to-enhance-and-benchmark-mllm-reasoning/665356.png","ImageObject",300,407,{"name":42,"@type":43},"lilac","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-30","2026-10-29",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",6,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does FlowGen address for MLLM visual reasoning over flowcharts?","Question",{"text":63,"@type":64},"FlowGen targets the lack of fine-grained control in existing flowchart datasets over structural complexity and rendering styles, which limits reliable training and benchmarking of MLLMs on visual reasoning tasks.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How does FlowGen generate flowcharts?",{"text":68,"@type":64},"FlowGen is a controllable flowchart synthesizer that produces diagrams with customizable structural features (e.g., order/size, branched arrows, nested subgraphs) and supports multiple renderer backends.",{"name":70,"@type":61,"acceptedAnswer":71},"What benefits do experiments report when training or testing with FlowGen?",{"text":72,"@type":64},"Experiments indicate that training on FlowGen substantially improves flowchart parsing and question answering, enhances generalization to other public datasets, and provides challenging test sets exposing consistent weaknesses in current MLLMs.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},665356,1793299676,{"code":4,"msg":82,"data":83},"success",{"doc_id":79,"user_id":84,"nickname":42,"user_avatar":85,"doc_module":4,"category_id":86,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":87,"file_id":88,"file_url":89,"file_type":90,"file_size":91,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":92,"language":93,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":94,"faqs":95,"seo_title":96,"seo_description":12,"update_tm":80,"read_time":97},3985747861865,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"FLOWGEN: SYNTHESIZING DIVERSE FLOWCHARTSTO ENHANCE AND BENCHMARK MLLM REASONING  \nKaiwen Shi, Sichen Liu, Ziyue Lin, Hangrui Guo, Gong Cheng∗  \nState Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China {kaiwenshi, sichenliu, 221240094, [221240069](221240069}@smail.nju.edu.cn)[}](221240069}@smail.nju.edu.cn)[@smail.nju.edu.cn](221240069}@smail.nju.edu.cn)[gcheng@nju.edu.cn](gcheng@nju.edu.cn)  \nABSTRACT  \nFlowcharts are widely used to represent processes and relationships through intuitive visual representations. However, accurately interpreting these diagrams remains challenging due to their structural complexity and high visual diversity.  \nExisting flowchart datasets often lack fine-grained control over key properties such as graph complexity and rendering style, limiting their utility for training and testing of multimodal large language models (MLLMs) on visual reasoning tasks. To address these limitations, we introduce FlowGen, a controllable synthesizer that generates flowcharts that have customizable structural features and supports multiple renderer backends. FlowGen enables fine-grained control over graph properties such as graph order and size, branched arrows, and nested subgraphs, facilitating systematic evaluation of MLLMs’ capabilities. Extensive experiments on open-source and proprietary MLLMs show that training on FlowGen substantially improves flowchart parsing and question answering (QA), while also enhancing generalization to other public datasets. Furthermore, FlowGen provides challenging test datasets that expose consistent weaknesses in current MLLMs, particularly related to high structural complexity and varied rendering styles. Our code and data are publicly available at [https:](https:)//[github.com/nju-websoft/FlowGen](github.com/nju-websoft/FlowGen).  \n1 INTRODUCTION  \nFlowcharts, as illustrated in Figure 1, are widely used to represent processes, relationships, and workflows, prevalent in domains such as scientific communication, business process modeling (BPMN), software engineering documentation, and education (Bhushan et al., 2024) . As a graphical language, flowcharts integrate symbolic information with spatial layout, enabling humans to quickly understand complex processes. However, this same structural complexity makes automatic parsing highly challenging (Montalvo, 1990) . Building models to extract structured representations from flowcharts is crucial for enabling downstream applications, including code generation, knowledge extraction from documents, automated reasoning over workflows, and multimodal question answering (Vasudevan et al., 2008 ; Bashir & Giri, 2013 ; Hu et al., 2024) .  \nMotivation. Recent advances in vision language models (VLMs) and MLLMs have shown potential to understand flowcharts, enabling end-to-end reasoning from images (Ye et al., 2024) . However, the development of such capabilities is hindered by the limitations of existing datasets. On the one hand, existing test sets provide limited control over complex structures, such as branching factor and  \nhierarchical nesting. They are also usually ∗ Corresponding Author: Gong Cheng.  \nFigure 1: Example flowchart synthesized by FlowGen.  \nTable 1: Comparison between our proposed synthesizer (FlowGen) and existing flowchart datasets.  \n\n| Dataset | Complexity | Renderer(s) | Source | \\#Train | \\#Test |\n| --- | --- | --- | --- | --- | --- |\n| FlowchartQA (Tannert et al., 2023) | NA | Graphviz | Synthetic | Not released | Not released |\n| FlowVQA (Singh et al., 2024) | NA | Mermaid | Synthetic | 1,319 | 953 |\n| CBD (Bhushan & Lee, 2022) | NA | NA | Web crawling | 300 | 96 |\n| FC A (Awal et al. 2011) | Branching | NA | Hand-drawn flowcharts | 248 | 145 |\n| ,\u003Cbr>FC B (Bresler et al., 2016) hdBPMN (Schfer et al., 2021b) | Branching\u003Cbr>Branching & Nesting | NA\u003Cbr>NA | Hand-drawn flowcharts\u003Cbr>Hand-drawn BPMN diagrams | 280\u003Cbr>0 | 196\u003Cbr>173 |\n| FlowLearn (Pan et al., 2024) | Branching | Mermaid | Syn","cbCaidiFtX5KPuT7","https://ap.wps.com/l/cbCaidiFtX5KPuT7","pdf",4036604,31,"English","# Abstract\n# 1 Introduction\n## Motivation\n## Our Work\n## Table 1 Comparison","[{\"question\":\"What problem does FlowGen address for MLLM visual reasoning over flowcharts?\",\"answer\":\"FlowGen targets the lack of fine-grained control in existing flowchart datasets over structural complexity and rendering styles, which limits reliable training and benchmarking of MLLMs on visual reasoning tasks.\"},{\"question\":\"How does FlowGen generate flowcharts?\",\"answer\":\"FlowGen is a controllable flowchart synthesizer that produces diagrams with customizable structural features (e.g., order/size, branched arrows, nested subgraphs) and supports multiple renderer backends.\"},{\"question\":\"What benefits do experiments report when training or testing with FlowGen?\",\"answer\":\"Experiments indicate that training on FlowGen substantially improves flowchart parsing and question answering, enhances generalization to other public datasets, and provides challenging test sets exposing consistent weaknesses in current MLLMs.\"}]","FLOWGEN - Synthesizing Diverse Flowcharts to Enhance and Benchmark MLLM Reasoning | PDF",78,{"code":4,"msg":82,"data":99},[100,104,108,112,117,121,126,129,134,137,141],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":101,"show_sort_weight":102,"slug":103},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":105,"show_sort_weight":106,"slug":107},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":109,"show_sort_weight":110,"slug":111},"Exam",70,"exam",{"id":113,"doc_module":4,"doc_module_name":25,"category_name":114,"show_sort_weight":115,"slug":116},5,"Comic",60,"comic",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":118,"show_sort_weight":119,"slug":120},"Technology",50,"technology",{"id":122,"doc_module":4,"doc_module_name":25,"category_name":123,"show_sort_weight":124,"slug":125},7,"Healthcare",40,"healthcare",{"id":86,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":127,"slug":128},30,"research-report",{"id":130,"doc_module":4,"doc_module_name":25,"category_name":131,"show_sort_weight":132,"slug":133},9,"Religion & Spirituality",20,"religion-spirituality",{"id":132,"doc_module":4,"doc_module_name":25,"category_name":135,"show_sort_weight":132,"slug":136},"World Cup","world-cup",{"id":138,"doc_module":4,"doc_module_name":25,"category_name":139,"show_sort_weight":138,"slug":140},10,"Lifestyle","lifestyle",{"id":142,"doc_module":4,"doc_module_name":25,"category_name":143,"show_sort_weight":113,"slug":144},19,"General","general"]