CommonForms

🪄 Automatically convert a PDF into a fillable form.

💻 Hosted Models (detect.semanticdocs.org) | 📄 CommonForms Paper | 🤗 Dataset | 🤗 FFDNet-L | 🤗 FFDNet-S

This repo contains three things:

1. the pip-installable commonforms package, which has a CLI and API for converting PDFs into fillable forms
2. the FFDNet-S and FFDNet-L models from the paper CommonForms: A Large, Diverse Dataset for Form Field Detection
3. the preprocessing code for the CommonForms dataset, which is hosted on HuggingFace: https://huggingface.co/datasets/jbarrow/CommonForms

Installation

CommonForms can be installed with either uv or pip, feel free to choose your package manager flavor:

uv pip install commonforms

Once it's installed, you should be able to run the CLI command on ~any PDF.

CommonForms CLI

The simplest usage will run inference on your CPU using the default suggested settings:

commonforms <input.pdf> <output.pdf>

Input	Output

Command Line Arguments

Argument	Type	Default	Description
input	Path	Required	Path to the input PDF file
output	Path	Required	Path to save the output PDF file
--model	str	FFDNet-L	Model name (FFDNet-L/FFDNet-S) or path to custom .pt file
--keep-existing-fields	flag	False	Keep existing form fields in the PDF
--use-signature-fields	flag	False	Use signature fields instead of text fields for detected signatures
--device	str	cpu	Device for inference (e.g., cpu, cuda, 0)
--image-size	int	1600	Image size for inference
--confidence	float	0.3	Confidence threshold for detection
--fast	flag	False	If running on a CPU, you can trade off accuracy for speed and run in about half the time

CommonForms API

In addition to the CLI, you can use

from commonforms import prepare_form

prepare_form(
    "path/to/input.pdf",
    "path/to/output.pdf"
)

All of the above arguments are keyword arguments to the prepare_form function.

Interactive Streamlit App

For fine-tuned control over field detection, use the interactive web interface:

streamlit run streamlit_app.py

The Streamlit app provides:

• Adjustable confidence threshold - Lower the confidence score if desirable fields aren't detected
• Visual field selection - Preview detected fields on each page and choose which ones to make fillable
• Interactive editing - Uncheck fields you don't want to include in the final PDF

This is especially useful when the default settings miss important form fields or detect unwanted elements.

Dataset Prep

🚧 Code for dataset prep exists in the dataset folder.

Troubleshooting

If form fields aren't being detected:

1. Use the Streamlit app to adjust the confidence threshold (try lowering from 0.3 to 0.1-0.2)
2. Visually inspect detected fields and manually select which ones to include
3. For better accuracy, use --model FFDNet-L instead of FFDNet-S

Citation

If you use the tool, models, or code in an academic paper, please cite the CommonForms paper:

@misc{barrow2025commonforms,
  title        = {CommonForms: A Large, Diverse Dataset for Form Field Detection},
  author       = {Barrow, Joe},
  year         = {2025},
  eprint       = {2509.16506},
  archivePrefix= {arXiv},
  primaryClass = {cs.CV},
  doi          = {10.48550/arXiv.2509.16506},
  url          = {https://arxiv.org/abs/2509.16506}
}

If you use it in a non-academic setting, please reach out to the author (joseph.d.barrow [at] gmail.com)! I love to hear when people are using my work!