ketanmore/ArabicDoc-layout-Detection

Setup Instructions

Clone the Surya OCR GitHub Repository

git clone https://github.com/VikParuchuri/surya.git
cd surya

Switch to v0.4.14

git checkout f7c6c04

Install Dependencies

The author has not provided requirements.txt file, but environment.yml from our conda environment has been uploaded, This file can be used to recreate environment for arabic_layout_model model.

ArabicDoc Pipeline

Download ArabicDoc.cpython-310-x86_64-linux-gnu.so , 10x_best.pt and surya folder from the Repository. Place ArabicDoc.cpython-310-x86_64-linux-gnu.so, 10x_best.pt and surya folder in same directory (They are dependent on each other).

from ArabicDoc import arabic_layout_model # This import will originate from ArabicDoc.cpython-310-x86_64-linux-gnu.so , which is present in the repo. Also this works with Linux based OS only.
from surya.postprocessing.heatmap import draw_bboxes_on_image
from PIL import Image

image_path = "sample.jpg"
image  = Image.open(image_path)
bboxes = arabic_layout_model(image_path)
plotted_image  = draw_bboxes_on_image(bboxes,image)

Refer to benchmark.ipynb for comparison between Traditional Surya Layout Model and New Layout Model.

Refer to results folder to visualize images obtained from both the models.