Under construction! Code and data are available and functional, though. Get in touch if you have any questions or comments!
$ git clone https://github.com/nlpAThits/docimg2mmax.git
$ cd docimg2mmax
$ conda create --name docimg2mmax python=3.8 --file reqs.txt
$ source activate docimg2mmax
$ pip install pypdf2 opencv-python
$ wget https://github.com/nlpAThits/pyMMAX2/archive/refs/tags/v0.65.zip
$ unzip v0.65.zip
$ pip install pyMMAX2-0.65/.The console tool pdftocairo is required if you want to use pdf2png.py for document preparation. Also, tesseract is required for the actual recognition.
PDF documents have to be converted to collections of PNG files, one per page, before extraction. pdf2png is a simple utility which generates files with the correct naming conventions.
$ python pdf2png.py --pdf_file_path ./data/in/pdf/PMC6742607_w_markup_300dpi.pdf --out_path ./data/in/png/ --num_sort_split_char . --decolor
Note that --tessdata_dir needs to be set to your local tesseract model installation folder. Improved models can be downloaded from here.
$ python docimg2mmax.py --img_folders ./data/in/png/PMC6742607_w_markup_300dpi@300dpi/ --mmax2_target_folder ./data/out/ --tessdata_dir /usr/share/tesseract-ocr/4.00/tessdata/ --oem 1 --psm 3 --dpi 300 --workers 1 --verbose --detect_markup --html_target_folder ./data/out/html/ --min_markup_percentage 15