culturalheritagenus/Jawi-layout-YOLO-v1

Yolo for historical Jawi newspapers

An YOLO-based model for detecting different regions in historical newspapers written in Malay in the Jawi script.

Note: You must scale down the size of the image for higher accuracy. Ideally the image sizes should scaled down be around 400 × 600 to 900 x 1100 for both models. Images too large (ie above 2000x2000) will produce erroneous results.

A helper function for image scaling would be as follows:

def scale_down_image(image, scale_factor=1/4):
    resized_img = image.resize((int(image.width * scale_factor), int(image.height * scale_factor)))
    return resized_img

Classes are as follows:

calligraphy
footer
header
image
row
text