improving OCR

crawler/91_recalculate_floorplan.py

@@ -0,0 +1,13 @@
+from data_access import Listing
+from tqdm import tqdm
+
+listings = Listing.get_all_listings()
+recalculate_listings = []
+
+for listing in listings:
+    sqm = listing.sqm_ocr
+    if sqm is None or sqm < 10 or sqm > 200:
+        recalculate_listings.append(listing)
+
+for listing in tqdm(recalculate_listings):
+    listing.calculate_sqm_ocr(recalculate=True)

crawler/rec/floorplan.py

@@ -1,5 +1,7 @@
 import re
 from PIL import Image
+import cv2
+import numpy as np
 from transformers import Pix2StructProcessor, Pix2StructForConditionalGeneration
 import pytesseract
 
@@ -32,9 +34,22 @@ def calculate_model(image_path):
     estimated_sqm = extract_total_sqm(output)
     return estimated_sqm, output, predictions_tensor
 
+def improve_img_for_ocr(img: Image):
+    img2 = np.array(img.convert('L'))
+    cv2.resize(img2, None, fx=1.2, fy=1.2, interpolation=cv2.INTER_CUBIC)
+    thresh = cv2.adaptiveThreshold(img2,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY,11,2)
+    return Image.fromarray(thresh)
 
 def calculate_ocr(image_path):
     img = Image.open(image_path)
     text = pytesseract.image_to_string(img)
     estimated_sqm = extract_total_sqm(text)
+    if estimated_sqm is None:
+        improved_img = improve_img_for_ocr(img)
+        text2 = pytesseract.image_to_string(improved_img)
+        estimated_sqm2 = extract_total_sqm(text2)
+        with open("recalculating.log", "a") as f:
+            f.write(f"before: {estimated_sqm} after: {estimated_sqm2} - {image_path}\n")
+        return estimated_sqm2, text2
+        
     return estimated_sqm, text