app.py

import gradio as gr
import numpy as np
import cv2
import onnxruntime as ort
import os
from huggingface_hub import hf_hub_download

# Function to download model files from Hugging Face
def download_model_files():
    model_repo = "SWHL/RapidOCR"
    det_model_filename = "PP-OCRv4/en_PP-OCRv3_det_infer.onnx"
    rec_model_filename = "PP-OCRv4/ch_PP-OCRv4_rec_server_infer.onnx"
    cls_model_filename = "PP-OCRv3/ch_ppocr_mobile_v2.0_cls_train.onnx"

    det_model_path = hf_hub_download(repo_id=model_repo, filename=det_model_filename)
    rec_model_path = hf_hub_download(repo_id=model_repo, filename=rec_model_filename)
    cls_model_path = hf_hub_download(repo_id=model_repo, filename=cls_model_filename)

    return det_model_path, rec_model_path, cls_model_path

# Download model files
det_model_path, rec_model_path, cls_model_path = download_model_files()

# Load models
det_session = ort.InferenceSession(det_model_path)
rec_session = ort.InferenceSession(rec_model_path)
cls_session = ort.InferenceSession(cls_model_path)

def preprocess_image(image):
    # Convert to grayscale
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    # Resize or pad image to desired size if necessary
    return gray

import cv2
from rapidocr_onnxruntime import RapidOCR

# Initialize the RapidOCR engine
ocr_engine = RapidOCR()

def ocr_predict(image):
    """
    Perform OCR on the input image and return the extracted text.
    """
    # Convert the image from RGB to BGR format as OpenCV uses BGR
    image_bgr = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)

    # Perform OCR
    result, _ = ocr_engine(image_bgr)

    # Extract text from the result
    extracted_text = "\n".join([item[1] for item in result])

    return extracted_text

# Define Gradio interface
iface = gr.Interface(
    fn=ocr_predict,
    inputs=gr.Image(type="numpy", label="Upload Image"),
    outputs=gr.Textbox(label="Detected Text"),
    title="RapidOCR Gradio Demo",
    description="Upload an image to extract text using RapidOCR."
)

if __name__ == "__main__":
    iface.launch()