What is table extraction and what models are there (TableNet, TATR)?

Table extraction is the task of extracting tabular data from documents, images, or PDFs, often using OCR for text recognition.

Models:

• TableNet — a neural network architecture that segments the image into table and non-table regions, and also highlights the table structure (cells, rows, columns). It uses convolutional networks to understand the visual structure.

• TATR (Table Transformer) — a transformer-based model that processes the document as a sequence and predicts the table structure, including cell boundaries and content. It works well with complex and heterogeneous tables.

Both models help automate the process of extracting and structuring tabular data from unstructured sources.