Fine Tuning Tesseract on Windows

Install Tesseract OCR and tool

1. Download the setup from this repo and make sure tesseract is in your PATH.
2. Install Python version 3.x
3. Install Git
4. Install winget/Windows Package Manager and then run winget install ezwinports.make and winget install wget.

Clone repo tesstrain

git clone https://github.com/tesseract-ocr/tesstrain.git

and clone the trained model repo

git clone https://github.com/tesseract-ocr/tessdata_best.git

also clone the langdata cause we will need the wordlist, punctation and number files

git clone https://github.com/tesseract-ocr/langdata

or download the trained model you are interested to fine tune in a folder tessdata_best

1. cd tesstrain

Create virtual environment

2. Create virtual environment

python -m venv .venv

Activate virtual environment

.venv\scripts\activate

Install requirements

python -m pip install -r requirements.txt

Create Folders

Create data folder

Open Git Bash and cd into the directory tesstrain and run this:

>>> DO NOT NEED THIS

make tesseract-langdata

in the data a langdata folder is create with a bunch of unicharset files

Ground Truth Data

Tesseract 5 using lines of data so we need to provide a image with the line (png or tif) and a text file with the content of the image. You can find a ZIP file ocrd-testset.zip with some ground truth data we can use to fine tuning.

1. unzip the file in a folder inside the data folder giving the name of the model you are going to create + ground-truth

IE: lft-ground-truth

Fine tuning

using this command to start the training process

make training MODEL_NAME=lft START_MODEL=eng TESSDATA=../tessdata_best LANGDATA_DIR=../langdata LEARNING_RATE=0.001 RATIO_TRAIN=0.80 MAX_ITERATIONS=5000

the fine tuning process will start and a folder with your model name will be created inside the data folder with the name of your model.

In the ground truth folder lft-ground-truth the box files will be created for each image.

At the end of the training process you can

make traineddata MODEL_NAME=lft

make plot MODEL_NAME=lft