Dataset and evaluation for HTR models for Latin and French Medieval Documentary Manuscripts

<strong>1. Dataset presentation.</strong> This is the dataset used to produce the HTR models applied to documentary Latin and French manuscripts presented in the paper: Sergio Torres Aguilar, Vincent Jolivet. <strong>Handwritten Text Recognition for Documentary Medieval<br> Manuscripts. </strong>2022. https://hal.science/hal-03892163 The dataset contains mostly charters and registers from the Late-medieval period (12th-15th). The training and evaluation, entailing 1855 pages, 120k lines of text and almost 1M tokens, were conducted using three freely available ground-truth corpora : <strong>The Alcar-HOME database </strong>: https://zenodo.org/record/5600884 <strong>The e-NDP corpus </strong>: https://github.com/chartes/e-NDP_HTR <strong>The Himanis project </strong>: https://zenodo.org/record/5535306 The final model operates in a multilingual environment (Latin and French) and it is able to recognize several Latin script families (mostly <em>Textualis</em> and <em>Cursiva</em>) in documents produced in ca. 12th - 15th centuries. During the evaluation the models shows an accuracy of <strong>94.01%</strong> on the validation set and a CER (character error ratio) of about <strong>0.12</strong> to <strong>0.17</strong> on four external unseen datasets. A fine-tuning exercise using 10 ground-truth pages can raise these results to a CER between <strong>0.06</strong> to <strong>0.10</strong> respectively. <strong>2. Dataset contents .</strong> a) <em>GT_list : </em>List containing the GT file names which constitute the training, evaluation and test sets. The images and transcriptions can be downloaded from their original repositories. b) <em>Training :</em> Contains the training and testing results (evaluation and prediction files) presented in the original paper for the two training phases: Regular (Textualis and Cursiva separated training) and Quartiles (mixed training by quartiles). c) <em>Useful_scripts :</em> Scripts to produce the HTR metrics (CER, WER, SER) and plot the model's accuracy. d) <em>Best_model :</em> Contains the best multilingual and multi-script model.