prthm29/Gujarati-Handwritten-Dataset

--- license: apache-2.0 task_categories: - image-to-text language: - gu tags: - handwritten-text-recognition - htr - gujarati - ocr - iiit-indic-hw-words pretty_name: Gujarati Handwritten Dataset (IIIT-INDIC-HW-WORDS) size_categories: - 10K-100K dataset_info: features: - name: file_name dtype: string - name: text dtype: string - name: image dtype: image splits: - name: train num_bytes: 2867138581.533 num_examples: 82563 - name: val num_bytes: 632023288.427 num_examples: 17643 - name: test num_bytes: 597646751.57 num_examples: 16490 download_size: 4016768714 dataset_size: 4096808621.53 configs: - config_name: default data_files: - split: train path: data/train-* - split: val path: data/val-* - split: test path: data/test-* --- # Gujarati Handwritten Word Dataset This dataset is a subset of the **IIIT-INDIC-HW-WORDS** collection, specifically focused on the **Gujarati** language. It is designed for training and evaluating **Handwritten Text Recognition (HTR)** models. ## Dataset Summary The original [IIIT-INDIC-HW-WORDS](https://cvit.iiit.ac.in/usodi/istr.php) is a large-scale benchmark for Indic scripts. This Gujarati subset contains word-level images manually written by multiple annotators to capture natural variations in handwriting styles. ### Key Statistics | Feature | Count | | :--- | :--- | | **Total Word Images** | 82,563 | | **Train Set** | 48,430 | | **Validation Set** | 17,643 | | **Test Set** | 16,490 | --- ## Dataset Structure & Extraction The dataset consists of image folders and corresponding annotation text files. Follow these instructions to map images to their transcriptions: ### 1. Files Overview * **Images:** Located in the `train/`, `val/`, and `test/` folders. * **Labels:** Provided in `train.txt`, `val.txt`, and `test.txt`. * **Lexicon:** `vocab.txt` contains the full list of Unicode strings used in the dataset. ### 2. Label Format Each row in the label files (`train.txt`, `val.txt`, `test.txt`) follows this format: ` <FileName>, <VocabId> ` ### 3. Mapping Logic The `<VocabId>` is a **0-indexed** pointer to the line number in `vocab.txt`. * **Step 1:** Locate the `VocabId` for an image in the split text file. * **Step 2:** Go to that specific line number in `vocab.txt` to extract the Unicode Gujarati string. --- ## Citation If you use this dataset in your research, please cite the following paper: ```bibtex @inproceedings{gongidi2021iiit, title={IIIT-Indic-HW-Words: A Dataset for Indic Handwritten Text Recognition}, author={Gongidi, Santhoshini and Jawahar, CV}, booktitle={Proceedings of the 16th International Conference on Document Analysis and Recognition (ICDAR)}, pages={444--459}, year={2021}, organization={Springer} } ```许可证：Apache-2.0 任务类别： - 图像到文本（image-to-text） 语言： - 古吉拉特语（gu） 标签： - 手写文本识别（Handwritten Text Recognition, HTR） - HTR - 古吉拉特语 - 光学字符识别（Optical Character Recognition, OCR） - iiit-indic-hw-words 数据集展示名：古吉拉特语手写数据集（IIIT-INDIC-HW-WORDS） 样本量区间： - 10K-100K 数据集信息： 特征： - 文件名：字符串类型 - 文本：字符串类型 - 图像：图像类型 拆分： - 训练集：字节数2867138581.533，样本量82563 - 验证集：字节数632023288.427，样本量17643 - 测试集：字节数597646751.57，样本量16490 下载大小：4016768714 数据集总大小：4096808621.53 配置项： - 配置名称：default 数据文件： - 训练集：路径data/train-* - 验证集：路径data/val-* - 测试集：路径data/test-* # 古吉拉特语手写单词数据集 本数据集是**IIIT-INDIC-HW-WORDS**合集的子集，专门针对**古吉拉特语**打造，用于训练和评估**手写文本识别（Handwritten Text Recognition, HTR）**模型。 ## 数据集概览 原始的[IIIT-INDIC-HW-WORDS](https://cvit.iiit.ac.in/usodi/istr.php)是面向印度语系脚本的大规模基准数据集。该古吉拉特语子集包含由多名标注人员手写生成的单词级图像，以覆盖手写风格的自然差异。 ### 关键统计数据 | 特征 | 数量 | | :--- | :--- | | **总单词图像数** | 82,563 | | **训练集** | 48,430 | | **验证集** | 17,643 | | **测试集** | 16,490 | --- ## 数据集结构与提取方式 本数据集由图像文件夹与对应的标注文本文件组成，请按照以下步骤完成图像与转录文本的映射： ### 1. 文件概览 * **图像**：存放在`train/`、`val/`和`test/`文件夹中。 * **标签**：提供于`train.txt`、`val.txt`和`test.txt`文件中。 * **词表**：`vocab.txt`包含数据集中使用的所有Unicode字符串全集。 ### 2. 标签格式 标签文件（`train.txt`、`val.txt`、`test.txt`）中的每一行格式为：` <文件名>, <词表ID> ` ### 3. 映射逻辑 <词表ID>是指向`vocab.txt`中行号的0索引指针。 * **步骤1**：在对应拆分的文本文件中查找图像对应的词表ID。 * **步骤2**：前往`vocab.txt`中的对应行号，提取古吉拉特语Unicode字符串。 --- ## 引用说明 若在研究中使用本数据集，请引用以下论文： bibtex @inproceedings{gongidi2021iiit, title={"IIIT-Indic-HW-Words: A Dataset for Indic Handwritten Text Recognition"}, author={Gongidi, Santhoshini and Jawahar, CV}, booktitle={Proceedings of the 16th International Conference on Document Analysis and Recognition (ICDAR)}, pages={444--459}, year={2021}, organization={Springer} }