cdhu-uu/SweMPer-layout-lite

Historical Document Layout Detection Model

A fine-tuned Mask R-CNN model (via LayoutParser/Detectron2) for detecting layout elements in historical Swedish medical journal pages. This is a lighter model and the more complex model can be found here Swemper-layout.

This model was developed as part of the research project:
Communicating Medicine (SweMPer): Digitalisation of Swedish Medical Periodicals, 1781–2011
(Project ID: IN22-0017), funded by Riksbankens Jubileumsfond.

Project page:

https://www.uu.se/en/department/history-of-science-and-ideas/research/research-projects-and-programmes/communicating-medicine-swemper

Model Details

• Model type: Mask R-CNN (ResNet backbone)
• Framework: Detectron2 / LayoutParser
• Fine-tuned for: Historical document layout analysis
• Language of source documents: Swedish

Label Map

ID	Label
0	Advertisement
1	Author
2	Header or Footer
3	Image
4	List
5	Page Number
6	Table
7	Text
8	Title

Evaluation Metrics

The evaluation metrics for this model are as follows:

AP	AP50	AP75	APs	APm	APl
64.325	88.948	69.214	40.350	55.117	67.543

Usage

Installation

Follow instructions at:
https://detectron2.readthedocs.io/en/latest/tutorials/install.html

Finetuning

Follow instructions at:
https://detectron2.readthedocs.io/en/latest/tutorials/training.html

Inference

import cv2
import layoutparser as lp
import matplotlib.pyplot as plt

# Configuration
model_config_path = "config_mask_rcnn_resized.yaml"
model_path = "SweMPer-layout-lite.pth"

label_map = {
    0: "advertisement",
    1: "author",
    2: "header_or_footer",
    3: "image",
    4: "list",
    5: "page_no",
    6: "table",
    7: "text",
    8: "title",
}

# Load model
model = lp.models.Detectron2LayoutModel(
    config_path=model_config_path,
    model_path=model_path,
    extra_config=["MODEL.ROI_HEADS.SCORE_THRESH_TEST", 0.8],
    label_map=label_map,
)

# Load and process image
image = cv2.imread("<path_to_image>")
image = image[..., ::-1]  # BGR to RGB

# Detect layout
layout = model.detect(image)

# Print detected elements
for block in layout:
    print(f"Type: {block.type}, Score: {block.score:.3f}, Box: {block.coordinates}")

# Visualize results
viz = lp.draw_box(image, layout, box_width=3, show_element_type=True)
plt.figure(figsize=(12, 16))
plt.imshow(viz)
plt.axis("off")
plt.show()

Acknowledgements

This work was carried out within the project:
Communicating Medicine (SweMPer): Digitalisation of Swedish Medical Periodicals, 1781–2011
(Project ID: IN22-0017), funded by Riksbankens Jubileumsfond.

We gratefully acknowledge the support of the funder and project collaborators.

This model builds upon the excellent work of:

• Detectron2
• LayoutParser

We thank the contributors and maintainers of these projects for making their tools publicly available and supporting research.