CommonForms – open models to auto-detect PDF form fields (github.com)

CommonForms is an open-source toolkit (pip-installable) that automatically converts PDFs into fillable forms via a simple CLI and Python API. The repo bundles two trained models (FFDNet-S and FFDNet-L) from the CommonForms paper, preprocessing code, and a link to the dataset hosted on Hugging Face (jbarrow/CommonForms). After installing, you can run a one-shot command (commonforms <input.pdf> <output.pdf>) or call prepare_form(...) from Python. Key runtime options include model choice (FFDNet-L default, or a custom .pt), device selection (cpu/cuda), image-size (default 1600), confidence threshold (default 0.3), flags to keep existing fields or use signature fields, and a “fast” CPU mode that trades accuracy for speed. This release is significant for document-understanding and data-extraction workflows because it provides end-to-end tools, pretrained models, and the underlying dataset—lowering the barrier to building automated form digitization, data-entry automation, and downstream layout/OCR pipelines. Technically, the package supports CPU/GPU inference, model swapping, and dataset repro steps for researchers who want to fine-tune or benchmark on the large, diverse CommonForms dataset (see arXiv:2509.16506). The combination of a ready-to-run CLI, accessible API, and open dataset/models makes it practical to deploy and iterate on form-field detection in production and research settings.