import torch
from PIL import Image
from transformers import BlipProcessor, BlipForQuestionAnswering
import gradio as gr
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',
)


processor = BlipProcessor.from_pretrained("ybelkada/blip-vqa-capfilt-large")
model = BlipForQuestionAnswering.from_pretrained("ybelkada/blip-vqa-capfilt-large", torch_dtype=torch.float16).to("cuda")

def vqa(image, question):
    inp = Image.fromarray(image.astype('uint8'), 'RGB')
    inputs = processor(inp, question, return_tensors="pt").to("cuda", torch.float16)
    out = model.generate(**inputs)

    return processor.decode(out[0], skip_special_tokens=True)

with gr.Blocks(theme=theme, css="footer {visibility: hidden}") as demo:
    gr.Markdown("""
    <div align='center' ><font size='60'>图片问答</font></div>
    """)
    with gr.Row():
        with gr.Column():
            image = gr.Image(label="图片")
            question = gr.Textbox(label="问题")
            with gr.Row():
                button = gr.Button("提交", variant="primary")
        box2 = gr.Textbox(label="文本")

    button.click(fn=vqa, inputs=[image, question], outputs=box2)
    examples = gr.Examples(examples=[['demo.jpg', 'how many dogs are in the picture?']],  inputs=[image, question], label="例子")


if __name__ == "__main__":
    demo.queue(concurrency_count=3).launch(server_name = "0.0.0.0", server_port = 7025)