Spaces:
Running
Running
Update ocr_engine.py
Browse files- ocr_engine.py +17 -9
ocr_engine.py
CHANGED
@@ -1,36 +1,44 @@
|
|
1 |
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
|
2 |
-
from PIL import Image
|
3 |
import torch
|
4 |
import re
|
5 |
|
6 |
-
# Load model
|
7 |
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
|
8 |
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")
|
9 |
|
10 |
def clean_ocr_text(text):
|
11 |
print("[RAW OCR]", text)
|
12 |
-
|
13 |
-
text =
|
|
|
14 |
print("[CLEANED OCR]", text)
|
15 |
return text
|
16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
def extract_weight(image):
|
18 |
try:
|
|
|
19 |
pixel_values = processor(images=image, return_tensors="pt").pixel_values
|
20 |
generated_ids = model.generate(pixel_values)
|
21 |
raw_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
22 |
|
23 |
cleaned = clean_ocr_text(raw_text)
|
24 |
|
25 |
-
#
|
26 |
-
match = re.search(r'(\d{1,
|
27 |
if match:
|
28 |
return f"{match.group(1)} {match.group(2)}"
|
29 |
|
30 |
-
# Fallback: only number
|
31 |
-
fallback = re.search(r'(\d{1,
|
32 |
if fallback:
|
33 |
-
return f"{fallback.group(1)} g"
|
34 |
|
35 |
return f"No valid weight found | OCR: {cleaned}"
|
36 |
except Exception as e:
|
|
|
1 |
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
|
2 |
+
from PIL import Image, ImageFilter
|
3 |
import torch
|
4 |
import re
|
5 |
|
6 |
+
# Load TrOCR model
|
7 |
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
|
8 |
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")
|
9 |
|
10 |
def clean_ocr_text(text):
|
11 |
print("[RAW OCR]", text)
|
12 |
+
# Fix common misreads
|
13 |
+
text = text.replace(",", ".").replace("s", "5").replace("o", "0").replace("O", "0")
|
14 |
+
text = re.sub(r"[^\d\.kg]", "", text.lower()) # Keep digits, dot, kg
|
15 |
print("[CLEANED OCR]", text)
|
16 |
return text
|
17 |
|
18 |
+
def preprocess_image(image):
|
19 |
+
# Enlarge + sharpen for better OCR
|
20 |
+
image = image.resize((image.width * 2, image.height * 2), Image.BICUBIC)
|
21 |
+
image = image.filter(ImageFilter.SHARPEN)
|
22 |
+
return image
|
23 |
+
|
24 |
def extract_weight(image):
|
25 |
try:
|
26 |
+
image = preprocess_image(image)
|
27 |
pixel_values = processor(images=image, return_tensors="pt").pixel_values
|
28 |
generated_ids = model.generate(pixel_values)
|
29 |
raw_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
|
30 |
|
31 |
cleaned = clean_ocr_text(raw_text)
|
32 |
|
33 |
+
# Try matching decimal weight with unit
|
34 |
+
match = re.search(r'(\d{1,3}\.\d{1,3})\s*(kg|g)', cleaned)
|
35 |
if match:
|
36 |
return f"{match.group(1)} {match.group(2)}"
|
37 |
|
38 |
+
# Fallback: match only decimal number
|
39 |
+
fallback = re.search(r'(\d{1,3}\.\d{1,3})', cleaned)
|
40 |
if fallback:
|
41 |
+
return f"{fallback.group(1)} g" # Default to grams
|
42 |
|
43 |
return f"No valid weight found | OCR: {cleaned}"
|
44 |
except Exception as e:
|