[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-346093-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-346093-en":130},{"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","supplementary-material-for-ocr-free-document-understanding-transformer","Supplementary Material for OCR-free Document Understanding Transformer","","Supplementary material describes the technical components supporting an OCR-free document understanding approach. It details OCR engine behaviors used for OCR-dependent baselines, including Microsoft, CLOVA, EasyOCR, and PaddleOCR, with notes on supported languages and roles in experiments. It also explains the synthetic document generator pipeline (SynthDoG), covering sampled backgrounds, realistic paper textures, randomized layout generation for text regions, multilingual corpus and font rendering, and image post-processing such as color and blur effects.",{"@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/supplementary-material-for-ocr-free-document-understanding-transformer/346093/",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/supplementary-material-for-ocr-free-document-understanding-transformer/346093.png","ImageObject",300,407,{"name":42,"@type":43},"Mary Man","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-16","2026-09-22",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",8,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What is the purpose of the supplementary material?","Question",{"text":63,"@type":64},"It provides detailed technical information used to support OCR-free document understanding, including OCR engine details for baselines and the SynthDoG synthetic data generation pipeline.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How are OCR engines used in OCR-dependent experiments?",{"text":68,"@type":64},"OCR-dependent backbones take OCR outputs as input features, and the supplementary section specifies which OCR engine is used for different datasets and tasks, such as documentation information extraction and timing measurements.",{"name":70,"@type":61,"acceptedAnswer":71},"What components make up SynthDoG?",{"text":72,"@type":64},"SynthDoG samples background images and paper textures, generates text layout regions using heuristic rules, renders multilingual text with appropriate fonts, and applies image post-processing like color/brightness/contrast adjustment and blur effects.","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},346093,1791344229,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,114,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":112,"slug":113},30,"research-report",{"id":115,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},9,"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":55,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":144,"read_time":145,"preview_page_count":33,"preview_urls":146},7421720224475,"https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c","Supplementary Material for OCR-free Document Understanding Transformer  \nGeewook Kim 1 ∗ , Teakgyu Hong4†, Moonbin Yim2†, JeongYeon Nam 1 , Jinyoung Park5†, Jinyeong Yim6†, Wonseok Hwang7†, Sangdoo Yun3 , Dongyoon Han3 , and Seunghyun Park 1  \n1 NAVER CLOVA 2 NAVER Search 3 NAVER AI Lab  \n4 Upstage 5 Tmax 6 Google 7 LBox  \nTable of Contents  \nA Appendix ..................................................... 1  \nA.1 Details of OCR Engines (MS, CLOVA, Easy, Paddle) .......... 1  \nA.2 Details of Synthetic Document Generator (SynthDoG) ......... 2  \nA.3 Details of Document Information Extraction .................. 4  \nA.4 Details of Model Training Scheme and Output Format ......... 5  \nA.5 Implementation and Training Hyperparameters ............... 6  \nA.6 Preliminary Experiments in Smaller Resources ................ 8  \nA.7 Details of OCR-dependent Baseline Models ................... 8  \nA Appendix  \nA.1 Details of OCR Engines (MS, CLOVA, Easy, Paddle)  \nCurrent state-of-the-art visual document understanding (VDU) backbones, such as BROS [6], LayoutLM [17] and LayoutLMv2 [16], are dependent on off-theshelf OCR engines. These backbones take the output of OCR as their (one of) input features. For the OCR-dependent methods, in our experiments, we use state-of-the-art OCR engines that are publicly available, including 2 OCR API products (i.e. , MS OCR 1 and CLOVA OCR2 ) and 2 open-source OCR models (i.e. , Easy OCR3 and Paddle OCR4 ) . In the main paper, Paddle OCR is used for the Chinese train ticket dataset [5] and CLOVA OCR is used for the rest datasets in the document information extraction (IE) tasks. MS OCR is used to measure the running time of the LayoutLM family in document classification and visual question answering (VQA) tasks, following the previous work of Xu et al. [16] . Each OCR engine is explained in the following.  \n∗ Corresponding author: [gwkim.rsrch@gmail.com](gwkim.rsrch@gmail.com)  \n† This work was done while the authors were at NAVER CLOVA.  \n1[https://docs.microsoft.com/en-us/azure/cognitive-services/](https://docs.microsoft.com/en-us/azure/cognitive-services/)[ ](https://docs.microsoft.com/en-us/azure/cognitive-services/)computer-vision/overview-ocr.  \n2[https://clova.ai/ocr/en](https://clova.ai/ocr/en).  \n3[https://github.com/JaidedAI/EasyOCR](https://github.com/JaidedAI/EasyOCR).  \n4[https://github.com/PaddlePaddle/PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR).  \n2 G. Kim et al.  \nMS OCR MS OCR 1 is the latest OCR API product from Microsoft and used in several recent VDU methods, e.g., LayoutLMv2 [16] . This engine supports 164 languages for printed text and 9 languages for handwritten text (until 2022/03) .  \nCLOVA OCR CLOVA OCR2 is an API product from NAVER CLOVA and is specialized in document IE tasks. This engine supports English, Japanese and Korean (until 2022/03) . In the ablation experiments on the CORD dataset [11](Figure 9 in the main paper), the CLOVA OCR achieved the best accuracy.  \nEasy OCR Easy OCR3 is a ready-to-use OCR engine that is publicly available at GitHub. This engine supports more than 80 languages (until 2022/03) . Unlike the aforementioned two OCR products (i.e. , MS OCR and CLOVA OCR), this engine is publicly opened and downloadable.3 The entire model architecture is based on the modern deep-learning-based OCR modules [2, 1] with some modifications to make the model lighter and faster. The total number of model parameters is 27M which is small compared to the state-of-the-art models [2, 1] .  \nPaddle OCR Paddle OCR4 is an open-source OCR engine available at GitHub. We used a lightweight (i.e., mobile) version of the model which is specially designed for a fast and light OCR of English and Chinese texts. The model is served on a CPU environment and the size of the model is extremely small, which is approximately 10M.  \nA.2 Details of Synthetic Document Generator (SynthDoG)  \nIn this section, we explain the components of the proposed Synthetic Document Generator (SynthDoG)","cbCaiqSrmTa1cOl5","https://ap.wps.com/l/cbCaiqSrmTa1cOl5","pdf",2805098,11,"English","# A Appendix\n## A.1 Details of OCR Engines (MS, CLOVA, Easy, Paddle)\n## A.2 Details of Synthetic Document Generator (SynthDoG)","[{\"question\":\"What is the purpose of the supplementary material?\",\"answer\":\"It provides detailed technical information used to support OCR-free document understanding, including OCR engine details for baselines and the SynthDoG synthetic data generation pipeline.\"},{\"question\":\"How are OCR engines used in OCR-dependent experiments?\",\"answer\":\"OCR-dependent backbones take OCR outputs as input features, and the supplementary section specifies which OCR engine is used for different datasets and tasks, such as documentation information extraction and timing measurements.\"},{\"question\":\"What components make up SynthDoG?\",\"answer\":\"SynthDoG samples background images and paper textures, generates text layout regions using heuristic rules, renders multilingual text with appropriate fonts, and applies image post-processing like color/brightness/contrast adjustment and blur effects.\"}]","Supplementary Material for OCR-free Document Understanding Transformer | PDF",1790060066,28,[147,151,155,159],{"page_no":22,"url":148,"width":149,"height":150},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/244922972074568074128d7500382b1f.webp",983,1272,{"page_no":26,"url":152,"width":153,"height":154},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/8f5f8cc35096e54f7afdd80c19118d0a.webp",982,1270,{"page_no":30,"url":156,"width":157,"height":158},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/ad7099beb333b6a1add9996086f21794.webp",981,1269,{"page_no":33,"url":160,"width":161,"height":162},"https://res-academy.cache.wpscdn.com/upload/kdocshare/2026-11-16/5b8517bc3be83ff3613b60962622bb96.webp",980,1268]