[{"@type":"PropertyValue","name":"Data Content","value":"Each data sample consists of one image and one JSON document. The JSON document contains either:OCR text recognition results of the image, or a textual description (caption) of the image, or visual question answering (VQA) based on the image, or visual question answering based on the OCR recognition results of the image，Among them, visual question answering includes at least one round of Q&A."},{"@type":"PropertyValue","name":"Data Scale","value":"89,007 sets in total, including 42,094 sets in Arabic and 46,913 sets in Japanese."},{"@type":"PropertyValue","name":"Category Distribution","value":"The dataset includes two languages, Japanese and Arabic, and covers four task categories for each language: Image Captioning , Visual Question Answering, Optical Character Recognition , and OCR-based Visual Question Answering. Each category is further divided into six domains: ①Business and Finance, ②Coding and Computer Science，③Law, Government, and Politics, ④Science, Technology, Engineering, and Mathematics , ⑤Society, Culture, Humanities, and Religion , ⑥Sports, Lifestyle, and Leisure."},{"@type":"PropertyValue","name":"Data Format","value":"Images in JPG or other common image formats; annotations in JSON format."},{"@type":"PropertyValue","name":"Collection accuracy","value":"The accuracy of image domain classification（per-image accuracy） is above 95%"},{"@type":"PropertyValue","name":"Annotation Accuracy","value":"The matching degree between image and text description is greater than 95%；OCR recognition accuracy (per-sentence accuracy) must exceed 95%. Accuracy is measured by segmenting at punctuation marks (such as commas, semicolons, exclamation marks, etc.) or at titles/headings."}]
{"id":1828,"datatype":"1","titleimg":"https://www.nexdata.ai/shujutang/static/image/index/datatang_tuxiang_default.webp","type1":"226","type1str":null,"type2":"254","type2str":null,"dataname":"89K Japanese-Arabic Image-Text Dataset for Multimodal LLM Training","datazy":[{"title":"Data Content","content":"Each data sample consists of one image and one JSON document. The JSON document contains either:OCR text recognition results of the image, or a textual description (caption) of the image, or visual question answering (VQA) based on the image, or visual question answering based on the OCR recognition results of the image，Among them, visual question answering includes at least one round of Q&A."},{"title":"Data Scale","content":"89,007 sets in total, including 42,094 sets in Arabic and 46,913 sets in Japanese."},{"title":"Category Distribution","content":"The dataset includes two languages, Japanese and Arabic, and covers four task categories for each language: Image Captioning , Visual Question Answering, Optical Character Recognition , and OCR-based Visual Question Answering. Each category is further divided into six domains: ①Business and Finance, ②Coding and Computer Science，③Law, Government, and Politics, ④Science, Technology, Engineering, and Mathematics , ⑤Society, Culture, Humanities, and Religion , ⑥Sports, Lifestyle, and Leisure."},{"title":"Data Format","content":"Images in JPG or other common image formats; annotations in JSON format."},{"title":"Collection accuracy","content":"The accuracy of image domain classification（per-image accuracy） is above 95%"},{"title":"Annotation Accuracy","content":"The matching degree between image and text description is greater than 95%；OCR recognition accuracy (per-sentence accuracy) must exceed 95%. Accuracy is measured by segmenting at punctuation marks (such as commas, semicolons, exclamation marks, etc.) or at titles/headings."}],"datatag":"Japanese,Arabic,Visual Question Answering(VQA),Image Captioning,Optical Character Recognition(OCR)","technologydoc":null,"downurl":null,"datainfo":null,"standard":null,"dataylurl":null,"flag":null,"publishtime":null,"createby":null,"createtime":null,"ext1":null,"samplestoreloc":null,"hosturl":null,"datasize":null,"industryPlan":null,"keyInformation":null,"samplePresentation":[],"officialSummary":"The dataset comprises a total of 89,007 samples, with each sample consisting of an image and a JSON document. The JSON document may contain image descriptions, visual question-answering pairs, OCR results extracted from the image, or visual question-answering pairs based on the OCR results.The dataset covers Arabic and Japanese languages and spans six major domains: ① Business and Finance, ②Coding and Computer Science, ③Law, Government, and Politics, ④Science, Technology, Engineering, and Mathematics (STEM), ⑤Society, Culture, Humanities, and Religion, ⑥ Sports, Lifestyle, and Leisure. Image classification accuracy (per-image) exceeds 95%; image-text matching accuracy is above 95%; OCR recognition accuracy (per-sentence) exceeds 95%.Suitable for multilingual OCR, multimodal Large Language Model (LLM) training, image captioning, and multilingual Visual Question Answering (VQA) tasks.","dataexampl":null,"datakeyword":["image text dataset","multimodal dataset","vision language dataset","image caption dataset","vlm training data","multimodal llm dataset"],"isDelete":null,"ids":null,"idsList":null,"datasetCode":null,"productStatus":null,"tagTypeEn":"Type","tagTypeZh":null,"website":null,"samplePresentationList":null,"datazyList":null,"keyInformationList":null,"dataexamplList":null,"bgimg":null,"datazyScriptList":null,"datakeywordListString":null,"sourceShowPage":"llm","dataShowType":"[{\"code\":\"0\",\"language\":\"ZH\"},{\"code\":\"1\",\"language\":\"ZH\"},{\"code\":\"2\",\"language\":\"EN,JP\"},{\"code\":\"3\",\"language\":\"EN\"},{\"code\":\"4\",\"language\":\"JP\"}]","productNameEn":"89,007 Sets of Japanese–Arabic Image-Text Construction Data","BGimg":"","voiceBg":["/shujutang/static/image/comm/audio_bg.webp","/shujutang/static/image/comm/audio_bg2.webp","/shujutang/static/image/comm/audio_bg3.webp","/shujutang/static/image/comm/audio_bg4.webp","/shujutang/static/image/comm/audio_bg5.webp"]}
89K Japanese-Arabic Image-Text Dataset for Multimodal LLM Training
image text dataset
multimodal dataset
vision language dataset
image caption dataset
vlm training data
multimodal llm dataset
The dataset comprises a total of 89,007 samples, with each sample consisting of an image and a JSON document. The JSON document may contain image descriptions, visual question-answering pairs, OCR results extracted from the image, or visual question-answering pairs based on the OCR results.The dataset covers Arabic and Japanese languages and spans six major domains: ① Business and Finance, ②Coding and Computer Science, ③Law, Government, and Politics, ④Science, Technology, Engineering, and Mathematics (STEM), ⑤Society, Culture, Humanities, and Religion, ⑥ Sports, Lifestyle, and Leisure. Image classification accuracy (per-image) exceeds 95%; image-text matching accuracy is above 95%; OCR recognition accuracy (per-sentence) exceeds 95%.Suitable for multilingual OCR, multimodal Large Language Model (LLM) training, image captioning, and multilingual Visual Question Answering (VQA) tasks.
This is a paid datasets for commercial use, research purpose and more. Licensed ready made datasets help jump-start AI projects.
Specifications
Data Content
Each data sample consists of one image and one JSON document. The JSON document contains either:OCR text recognition results of the image, or a textual description (caption) of the image, or visual question answering (VQA) based on the image, or visual question answering based on the OCR recognition results of the image，Among them, visual question answering includes at least one round of Q&A.
Data Scale
89,007 sets in total, including 42,094 sets in Arabic and 46,913 sets in Japanese.
Category Distribution
The dataset includes two languages, Japanese and Arabic, and covers four task categories for each language: Image Captioning , Visual Question Answering, Optical Character Recognition , and OCR-based Visual Question Answering. Each category is further divided into six domains: ①Business and Finance, ②Coding and Computer Science，③Law, Government, and Politics, ④Science, Technology, Engineering, and Mathematics , ⑤Society, Culture, Humanities, and Religion , ⑥Sports, Lifestyle, and Leisure.
Data Format
Images in JPG or other common image formats; annotations in JSON format.
Collection accuracy
The accuracy of image domain classification（per-image accuracy） is above 95%
Annotation Accuracy
The matching degree between image and text description is greater than 95%；OCR recognition accuracy (per-sentence accuracy) must exceed 95%. Accuracy is measured by segmenting at punctuation marks (such as commas, semicolons, exclamation marks, etc.) or at titles/headings.
Specifications
Sample
Recommended Dataset
