Losing an order of subheadings in the document

[TomekPro]

Hi, I noticed the following issue when working with doctr - It messes up an order of subheadings. I checked it on multiple documents. Here is a document example:

It returns:

Any idea how to overcome this issue? @felixdittrich92 ?

Code:

from doctr.models import ocr_predictor

model = ocr_predictor(pretrained=True)

path = "./data/01_segmentation.jpg"
doc = DocumentFile.from_images(path)
result = model(doc)
text = result.render()

print(text)```

Thanks in advance!
Tomek

[TomekPro]

Issue is connected to wrong segmentation:

import numpy as np
import matplotlib.pyplot as plt
import cv2
from bs4 import BeautifulSoup
import re
import os
os.environ['USE_TORCH'] = '1'

from doctr.io import DocumentFile
from doctr.models import ocr_predictor

predictor = ocr_predictor(pretrained=True)

# path = "/content/430629_medicalDocumentatnionFile_2__0.jpg"
path = "/content/01_segmentation_tomek.jpg"
doc = DocumentFile.from_images(path)
result = predictor(doc)
text = result.render()
hocr = result.export()
xml = result.export_as_xml()[0][0].decode("utf-8")

# parse the html
html = re.sub(r"\s+", " ", xml)
soup = BeautifulSoup(html)

blocks_list = []
for el in soup.find_all("div", {"class": "ocr_carea"}):
    blocks_list.append(
        {
            "id":el["id"].split("_")[1],
            "bbox": [int(i) for i in el["title"].split(";")[0].split()[1:]],
        }
    )


img_to_draw = cv2.imread(path)
def draw_contours(contours_list, img, color, size):
    for contour in contours_list:
        x,y, xx,yy = contour["bbox"]
        cv2.rectangle(img, (x,y), (xx,yy), color, size) # -1

draw_contours(blocks_list, img_to_draw, (255,160,122), 2)

plt.figure(figsize=(30,30))
plt.imshow(img_to_draw)
plt.show()