ocr2 / app.py
shayansjm's picture
Update app.py
d3be5ae verified
import gradio as gr
from paddleocr import PaddleOCR
from PIL import Image, ImageOps
import numpy as np
# Global model holder
ocr_instance = None
def process_bank_form(image):
global ocr_instance
if image is None:
return "Please upload an image."
try:
# 1. LAZY LOAD MODEL (Standard 2026 init)
if ocr_instance is None:
ocr_instance = PaddleOCR(
lang='en',
ocr_version='PP-OCRv5', # Using latest v5 for best handwriting
use_angle_cls=True
)
# 2. IMAGE STANDARDIZATION
img = image.convert("RGB")
img = ImageOps.exif_transpose(img) # Prevents 'sideways' photo errors
img_array = np.array(img)
# 3. RUN OCR
result = ocr_instance.ocr(img_array)
# 4. DEEP DEFENSIVE PARSING (Fixes 'index out of range')
if not result or not isinstance(result, list) or len(result) == 0 or result[0] is None:
return "No text detected. Try a closer, clearer photo."
extracted_text = []
# result[0] is the list of detected lines/boxes
for line in result[0]:
# Each 'line' must be a list: [ [coordinates], [text, confidence] ]
if isinstance(line, list) and len(line) >= 2:
content = line[1] # This should be [text, confidence]
# Check if 'content' has a list with at least the text string
if isinstance(content, (list, tuple)) and len(content) > 0:
text_val = content[0]
# Ensure it's a valid string and skip if empty
if text_val and isinstance(text_val, str):
extracted_text.append(text_val.strip())
if not extracted_text:
return "AI found text boxes but couldn't read characters. Try cropping the image."
return "\n".join(extracted_text)
except Exception as e:
return f"System Error: {str(e)}\nTip: If it persists, use 'Factory Reboot' in Space Settings."
# UI
demo = gr.Interface(
fn=process_bank_form,
inputs=gr.Image(type="pil", label="Upload Bank Form"),
outputs=gr.Textbox(label="Extracted Text", lines=20),
title="🏦 Global Bank OCR"
)
if __name__ == "__main__":
demo.launch(max_threads=1)