Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import torch
|
3 |
+
from PIL import Image
|
4 |
+
import pytesseract
|
5 |
+
from torchvision import transforms
|
6 |
+
from model import UTRNet # Assuming the UTRNet model is defined in a file `model.py`
|
7 |
+
|
8 |
+
# Load model
|
9 |
+
def load_model():
|
10 |
+
model = UTRNet() # Initialize the model (ensure it is defined in a separate model.py)
|
11 |
+
model.load_state_dict(torch.load('saved_models/UTRNet-Large/best_norm_ED.pth'))
|
12 |
+
model.eval()
|
13 |
+
return model
|
14 |
+
|
15 |
+
# Image preprocessing
|
16 |
+
def preprocess_image(image):
|
17 |
+
transform = transforms.Compose([
|
18 |
+
transforms.ToTensor(),
|
19 |
+
transforms.Resize((320, 320)),
|
20 |
+
])
|
21 |
+
return transform(image).unsqueeze(0)
|
22 |
+
|
23 |
+
# OCR prediction function
|
24 |
+
def predict_ocr(image, model):
|
25 |
+
image_tensor = preprocess_image(image)
|
26 |
+
with torch.no_grad():
|
27 |
+
output = model(image_tensor)
|
28 |
+
# Post-process the output to get text (This depends on how the model is structured)
|
29 |
+
return output # You might need to decode the output to actual text
|
30 |
+
|
31 |
+
# Streamlit App
|
32 |
+
def main():
|
33 |
+
st.title("Urdu Text Extraction Using UTRNet")
|
34 |
+
st.write("Upload an image containing Urdu text for OCR extraction.")
|
35 |
+
|
36 |
+
uploaded_image = st.file_uploader("Upload Image", type=["jpg", "png", "jpeg"])
|
37 |
+
|
38 |
+
if uploaded_image is not None:
|
39 |
+
# Load and display the image
|
40 |
+
image = Image.open(uploaded_image)
|
41 |
+
st.image(image, caption="Uploaded Image", use_column_width=True)
|
42 |
+
|
43 |
+
# Load the model
|
44 |
+
model = load_model()
|
45 |
+
|
46 |
+
# Get predictions
|
47 |
+
if st.button("Extract Text"):
|
48 |
+
output = predict_ocr(image, model)
|
49 |
+
st.write("Extracted Text:")
|
50 |
+
st.write(output) # You will need to process `output` to display text properly
|
51 |
+
|
52 |
+
if __name__ == "__main__":
|
53 |
+
main()
|