cells2table 0.2.0

Table image parsing with cell detection models

cells2table

Parsing tables in document images with cell detection models

Implemented pipelines

PaddlePaddle models

• Classification model (wired / wireless)
• Cell detection model with different weights for each class

Using ONNX weights (downloaded automatically on first use with huggingface_hub)

Instalation

With uv, add to your project with:

uv add git+https://github.com/jspast/cells2table

ONNX models need a ONNX Runtime installed to run. You can install one on your own or use one of the optionals already configured.

Optional	Description
cuda	For NVIDIA GPUs
openvino	For Intel GPUs and CPUs
cpu	Default CPU runtime
docling	For docling usage

Usage

cells2table only extract structural information from the tables. Another library is needed to extract content from the cells.

Docling

A docling plugin is provided to allow integrating cells2table in a complete pipeline.

Usage example:

from cells2table.docling import CustomDoclingTableStructureOptions

pipeline_options = PdfPipelineOptions(
    allow_external_plugins=True,
    table_structure_options=CustomDoclingTableStructureOptions(),
)

converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options),
        InputFormat.IMAGE: PdfFormatOption(pipeline_options=pipeline_options),
    }
)

result = converter.convert("path/to/document.pdf")
print(result.document.export_to_markdown())