DocLayout

DocumentLayout

DocumentLayout is an object detection model used to analyze document layout. A total of 11 types were detected in the layout of the document.

• Caption
• Footnote
• Formula
• List-item
• Page-footer
• Page-header
• Picture
• Section-header
• Table
• Text
• Title

Model Training

Data

DocLayNet: DS4SD/DocLayNet: DocLayNet: A Large Human-Annotated Dataset for Document-Layout Analysis (github.com)

Data Preprocessing

coco format data needs to be converted to Yolov8 input format in advance.

Please refer to dataset.py

Training

Please refer to train.py

Model Predicting

Please refer to test.py

Gradio UI

Please refer to gradio.py