main.py

from mmocr.ocr import MMOCR
import gradio as gr
import cv2

model_dir = 'model'
ocr = MMOCR(det_config=f'{model_dir}/config.py',
            det_ckpt=f'{model_dir}/epoch_40.pth', device='cpu')
def get_rec(points):
    xs = []
    ys = []
    for ix, iv in enumerate(points):
        if ix % 2:
            ys.append(iv)
        else:
            xs.append(iv)
    return (min(xs), min(ys)), (max(xs), max(ys))
        
    
def predict(image_input, score_threshold):
    draw_img = image_input.copy()
    print(image_input.shape)
    output = ocr.readtext(image_input)
    polygons = output['det_polygons']
    scores = output['det_scores']
    for polygon, score in zip(polygons, scores):
        if score < score_threshold:
            continue
        
        p0, p1 = get_rec([int(i) for i in polygon])
        draw_img = cv2.rectangle(draw_img, p0, p1, (255,0,0), 2)
        draw_img = cv2.putText(draw_img, '%.3f' % score, p0, cv2.FONT_HERSHEY_PLAIN, 2)
    return draw_img

def run():
    demo = gr.Interface(
        fn=predict,
        inputs=[gr.components.Image(), gr.Slider(0, 1, 0.8)],
        outputs=gr.components.Image(),
    )

    demo.launch(server_name="0.0.0.0", server_port=7860)


if __name__ == "__main__":
    run()