Prerequisites

• Virtual Environment with Python < 3.13 is needed, to make tensorflow work
• Always execute the main script from the root folder, aka python3 src/main.py!
• Start WebApp with:

streamlit run src/app.py

Modules

Modules are the building blocks of the pipeline. Each Module should execute (in theory) exactly one task. Currently following modules exist:

• PdfConverter: Converts PDF to jpg pages

  • Input
    • output_folder: str
  • Output: jpg images in output_folder

• TableRotator: Rotates jpg images

  • Input
    • output_folder: str
  • Output: rotated jpg images in output_folder

• TatrExtractor: Extract table structure from jpg images

  • Input
    • output_folder: str
  • Output: extracted table images in output_folder

• ColumnExtractor: Tries to extract all columns from table images

  • Input
    • minFoundColumns: int: Min amount of columns to be found to be an "eligible" page
    • try_experimental_unify: bool: Experimental feature. Tries to map columns based on their extracted width
  • Output list[ColumnExtractorResult]
    • columns_rgb: list[np.ndarray]: Cut-out rgb columns
    • columns_gray: list[np.ndarray]: Cut-out gray-scaled columns
    • split_widths: list[float]: Column widths

• RowExtractor: Tries to extract all rows inside a column

  • Output list[RowExtractorResult]
    • columns: list[list[np.ndarray]]: Cut-out gray cell for each column and row

• CellDenoiser: Denoises each cell

  • Output list[CellDenoiserResult]
    • columns: list[list[np.ndarray]]: Denoised gray cell for each column and row