The Molino dataset for transcription

This repository provides the Molino dataset together with pretrained FP-THD models and evaluation resources used in our experiments. The main dataset is distributed as a compressed archive: Molino_lines.tbz – contains the historical text line images and corresponding annotations used for training and evaluation. In addition, the repository includes the following resources: train.ln : training split containing line-level annotations. val.ln : validation split used for model evaluation during training. The repository also provides several pretrained FP-THD models trained on different datasets and configurations: FP-THD-pretrained-model-BenthamDataset_batchz64 : pretrained model trained on the Bentham dataset with batch size 64. FP-THD-pretrained-model-Medieval-Latin : pretrained model trained on a Medieval Latin handwriting dataset. FP-THD-pretrained-model-Rodrigo-batchz64 : pretrained model trained on the Rodrigo dataset with batch size 64. FP-THD-pretrained-model-Rodrigo-batchz128 : pretrained model trained on the Rodrigo dataset with batch size 128. The test collection for the Molino dataset consists of 10 images with ground truths , provided in the file test_images_collection.tbz This dataset and the associated models are intended for research on Historical Text Recognition and document analysis, particularly for experiments involving historical line-level recognition and end-to-end document recognition. Molino dataset is released as part of the Miguel del Molino project (https://migueldelmolino.es/) supported by the Aragon Regional Government (Spain) [grant number PROY_S11_24]