Open-Source PDF Toolkit Delivers High-Accuracy Layout and Formula Detection

PDF-Extract-Kit is an open‑source toolkit that provides layout detection, formula detection, formula recognition, and OCR for PDF documents.

Usage Examples

Layout Detection

Formula Detection

Exam Paper Detection

Evaluation Metrics

Layout Detection

Models compared: DocXchain, Surya, two models from 360LayoutAnalysis, and LayoutLMv3‑SFT (fine‑tuned from LayoutLMv3‑base‑chinese). Validation sets: 402 academic‑paper pages and 587 textbook pages.

Formula Detection

Models evaluated: Pix2Text‑MFD (open‑source) and a YOLOv8‑l fine‑tuned version (YOLOv8‑Trained). Validation sets: 255 academic‑paper pages and 789 mixed pages (textbooks, books, etc.).

Formula Recognition

Uses Unimernet weights without additional fine‑tuning.

Quick Start (macOS M‑series)

Clone the repository

git clone https://github.com/opendatalab/PDF-Extract-Kit.git

Create and activate a virtual environment

python -m venv venv
source venv/bin/activate

Install dependencies

pip install -r requirements+cpu.txt
# Detectron2 requires compilation; see https://github.com/facebookresearch/detectron2/issues/5114
# or install the pre‑built wheel
pip install https://github.com/opendatalab/PDF-Extract-Kit/raw/main/assets/whl/detectron2-0.6-cp310-cp310-macosx_11_0_arm64.whl

Configure MPS acceleration by editing configs/model_configs.yaml and setting device to mps:

model_args:
  device: mps
  img_size: 1888
  conf_thres: 0.25
  iou_thres: 0.45
  pdf_dpi: 200
  layout_weight: models/Layout/model_final.pth
  mfd_weight: models/MFD/weights.pt
  mfr_weight: models/MFR/UniMERNet

Run the program

python pdf_extract.py --pdf demo/demo1.pdf

https://github.com/opendatalab/PDF-Extract-Kit

References

PDF-Extract-Kit: https://github.com/opendatalab/PDF-Extract-Kit

Unimernet: https://github.com/opendatalab/UniMERNet