Tesseract Training

Training workflow for Tesseract 4 as a Makefile for dependency tracking and building the required software from source.

New features:

• Process for finetuning LSTM network
• Additional tool: Convert dataset based ICDAR format to the format of this tool.

Install

To build leptonica and tesseract, additional data and install it to a subdirectory ./usr in the repo:

  make leptonica tesseract langdata

Provide ground truth

Place ground truth consisting of line images and transcriptions in the folder data/train for training and data/eval for evaluation.

Images must be TIFF and have the extension .tif.

Transcriptions must be single-line plain text and have the same name as the line image but with .tif replaced by .gt.txt.

The repository contains a ZIP archive with sample ground truth, see ocrd-testset.zip. Extract it to ./data/train and run make training.

Running

To train Tesseract from scratch, use the command:

 make training MODEL_NAME=model_name

which is basically a shortcut for

   make unicharset lists proto-model training

To finetune Tesseract LSTM, use the command:

   make finetune MODEL_NAME=model_name TRAINED_MODEL=path/to/traineddata_file

Noted that only trained files in tessdata_best can be finetuned, models in tessdata and tessdata_fast will cause segmentation dump because they are integer models.

Run make help to see all the possible targets and variables:

  Targets

    unicharset       Create unicharset
    lists            Create lists of lstmf filenames for training and eval
    training         Start training
    finetune         Start training from an existing model
    proto-model      Build the proto model
    leptonica        Build leptonica
    tesseract        Build tesseract
    tesseract-langs  Download tesseract-langs
    langdata         Download langdata
    clean            Clean all generated files
    convert          Convert ICDAR formated dataset to the useful format

  Variables

    MODEL_NAME         Name of the model to be built
    CORES              No of cores to use for compiling leptonica/tesseract
    LEPTONICA_VERSION  Leptonica version. Default: 1.75.3
    TESSERACT_VERSION  Tesseract commit. Default: 9ae97508aed1e5508458f1181b08501f984bf4e2
    LANGDATA_VERSION   Tesseract langdata version. Default: master
    TESSDATA_REPO      Tesseract model repo to use. Default: _fast
    TRAIN              Train directory
    RATIO_TRAIN        Ratio of train / eval training data
    ICDAR_DATA         ICDAR Data folder to be converted
    TRAINED_MODEL      Existing model to finetune

License

Software is provided under the terms of the Apache 2.0 license.

Original repository provided by OCR-D