DocModel is a state-of-the-art document understanding model designed to extract both textual content and 2D spatial relationships from complex documents. Built on the RoBERTa architecture, DocModel is fine-tuned for tasks such as form understanding, entity extraction, and layout-based document analysis.
2D Spatial Modeling: Captures text layout and spatial relationships within documents, ideal for complex document structures such as forms, tables, and scans. RoBERTa-based Architecture: Built on a robust architecture for token-level tasks with powerful self-attention mechanisms. Fine-tuned for Document Understanding: Specifically trained on datasets like FUNSD to handle noisy and complex document layouts.
DocModel has been evaluated on the FUNSD dataset, achieving competitive results in extracting meaningful information from challenging, real-world documents.
Evaluation Loss: 1.36752
F1-Score: 0.84126
To install and use DocModel, follow these steps:
- Clone the repository:
git clone https://github.com/tobiadefami/docmodel.git
cd docmodel- Install the package using setup.py:
python setup.py install- You can also install the model dependencies using pip:
pip install -r requirements.txtDocModel can be used for a variety of document understanding tasks, including:
- Form understanding: Extracting key-value pairs from structured forms.
- Entity extraction: Identifying important information from documents with diverse layouts.
- Layout-based analysis: Handling complex layouts involving tables, scanned images, and multi-column formats.
Model Hub: DocModel on Hugging Face Hub
This project is licensed under the Mozilla Public License 2.0. See the LICENSE file for details.
For any questions or inquiries, feel free to reach out!