import torch
from PIL import Image
import gradio as gr
from lavis.models import load_model_and_preprocess
from lavis.processors import load_processor
from gradio.themes.utils import sizes


theme = gr.themes.Default(radius_size=sizes.radius_none).set(
    block_label_text_color = '#4D63FF',
    block_title_text_color = '#4D63FF',
    button_primary_text_color = '#4D63FF',
    button_primary_background_fill='#FFFFFF',
    button_primary_border_color='#4D63FF',
    button_primary_background_fill_hover='#EDEFFF',
)


raw_image = Image.open("./merlion.png").convert("RGB")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model, vis_processors, text_processors = load_model_and_preprocess("blip_image_text_matching", "large", device=device, is_eval=True)

def image_text_match_compute(image, text):
    raw_image = Image.open(image).convert("RGB")
    img = vis_processors["eval"](raw_image).unsqueeze(0).to(device)
    txt = text_processors["eval"](text)

    itm_output = model({"image": img, "text_input": txt}, match_head="itm")
    itm_scores = torch.nn.functional.softmax(itm_output, dim=1)

    return f'The image and text are matched with a probability of {itm_scores[:, 1].item():.3%}' 


with gr.Blocks(theme=theme, css="footer {visibility: hidden}") as demo:
    gr.Markdown("""
        <div align='center' ><font size='60'>图片文本相似度计算</font></div>
    """)
    with gr.Row():
        with gr.Column():
            image = gr.Image(label="图片", type="filepath")
            text = gr.Textbox(label="问题")
            with gr.Row():
                button = gr.Button("提交", variant="primary")
        box2 = gr.Textbox(label="文本")

    button.click(fn=image_text_match_compute, inputs=[image, text], outputs=box2)
    examples = gr.Examples(examples=[['merlion.png', 'merlion in Singapore']],  inputs=[image, text], label="例子")


if __name__ == "__main__":
    demo.queue().launch(server_name = "0.0.0.0")