Skip to content
/ PdfDet Public

PdfDet aims to simplify PDF layout detect tasks for users.

License

Notifications You must be signed in to change notification settings

pleb631/PdfDet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

pdfdet: PDF Layout Detection Toolbox

效果1

Requirements

python>=3.10

python -m pip install -r requirements.txt
pip install .

Usage

Simple Operations

python main.py --path "pdf_path"

You can also import a model directly via its full name and then call its __call__ method with pdf path or image path.

  • Layer

    from pdfdet import uni_model
    model = uni_model(name='yolov8m_cdla')
    layer = model(path="image_path")
    content  = layer.to_json()
    """
    content format:
    {
        "image": numpy.ndarray,
        "boxes": [{"box": [x1, y1, x2, y2], "label": str, "score": float}, ...],
    }
    """
  • Document

    from pdfdet import uni_model
    import cv2
    model = uni_model(name='yolov8m_cdla')
    doc = model(path="pdf_path")
    layers = sorted(doc.layers, key=lambda x: int(x))
    for i in layers:
        layer = getattr(doc, i)
        im = layer.imshow()
        im = cv2.resize(im, (640, 640))
        cv2.imshow("im", im)
        cv2.waitKey(0)
    cv2.destroyAllWindows()

Batch processing

# batch predict
python tools/batch_process.py --model "model_name" --src "image_root" --save "res_root"
# generate visualize result
python tools/visualize.py "image_path" "res_path"
#
# evaluate cdla dataset(labelme format)
python tools/eval_map50.py "gt_root" "res_root"

Models

Model Source Associated Dataset optional model
paddle_pub PaddlePaddle PubLayNet(English)
paddle_cdla PaddlePaddle CDLA(Chinese)
cnstd_yolov7 CNSTD CDLA
yolov8l_doc huggingface DocLayNet(English, German,French, Japanese) yolov8n_doc,yolov8s_doc
yolov8m_cdla layout_analysis CDLA yolov8n_cdla

Evaluation Code Source

Note: Labels and annotation strategies vary across different datasets. Visual comparison should be the primary method for evaluating effectiveness.

CDLA

CDLA

Model map50 map50:95 p r
paddle_cdla 0.9675 0.8359 0.9602 0.9347
cnstd_yolov7 0.9058 0.6662 0.9543 0.8321
yolov8m_cdla 0.9436 0.8086 0.9449 0.8980

mnbvc

Test Dataset

Model map50:95 p r
paddle_cdla 0.5717 0.5853 0.6248
cnstd_yolov7 0.5034 0.6278 0.5651
yolov8m_cdla 0.4783 0.5266 0.5922