Muharaf-public

Manuscripts of Handwritten Arabic dataset (Muharaf) for cursive text recognition.  The following files are present in this repositoriy:

1. public_data_files.zip: Contains the public part of Muharaf dataset. It has the images and the corresponding annotation files in JSON and XML format.
2. public_line_images.zip: Contains the line images and their corresponding transcriptions.
3. public_summary_and_keywords.zip: Contains the summary and keywords extracted from the ground truth transcriptions of each image.
4. sfr_files.zip: Contains the preprocessed files for the start_follow_read_arabic system for training the public part of Muharaf dataset.
5. public_1100_untrained.zip: Contains an initiailized trial folder with 3 different random splits of (train, validation, test) to reproduce the experiments reported in the paper on Muharaf-public. 
6. public_1100_trained.zip: Contains the results and models weights after training on Muharaf-public. It has results of three different random splits of (train, validation, test) sets.
7. trial_15_untrained.zip: Contains an intialized trial folder with 3 different random splits of (train, validation, test) to reproduce the experiments reported in the paper on training all the files of Muharaf dataset (1500 training images). 
8. trial_15.zip: Contains the results and model weights after training on Muharaf. It has results of three different random splits of (train, validation, test) sets.