#!/usr/bin/env python

import gradio as gr
import PIL.Image
import spaces
import torch
from transformers import AutoModelForCausalLM, AutoProcessor

DESCRIPTION = "# Capabara - Image Captioning with GIT"

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model_id = "microsoft/git-large-coco"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id).to(device)


@spaces.GPU
def run(image: PIL.Image.Image) -> str:
    inputs = processor(images=image, return_tensors="pt").to(device)
    generated_ids = model.generate(
        pixel_values=inputs.pixel_values, num_beams=3, max_length=20, min_length=5
    )
    return processor.batch_decode(generated_ids, skip_special_tokens=True)[0]


with gr.Blocks(
    title="Capabara - Image Captioning with GIT",
    theme=gr.themes.Default(font=[gr.themes.GoogleFont("Work Sans"), "sans-serif"]),
    css_paths="static/css/style.css",
) as demo:
    gr.Image(
        elem_id="capa-logo",
        value="./static/assets/TM_Capabara_210622.png",
        container=False,
        height=139,
        width=176,
        show_download_button=False,
        show_fullscreen_button=False,
    )

    gr.HTML(
        elem_id="capa-title",
        value="""
        <h1>Capabara - Image Captioning with GIT</h1>
        """,
    )

    input_image = gr.Image(type="pil")
    run_button = gr.Button(value="Caption", elem_classes=["capa-btn"])
    output = gr.Textbox(label="Result")

    gr.HTML("""<br/>""")
    gr.HTML(
        """
            <h4>Reference</h4>
            <ul>
                <li>
                    <p>
                        This project uses code from <a href="https://huggingface.co/spaces/hysts/image-captioning-with-git/tree/main" target="_blank">Hysts</a>. Thank you for your contributions!
                    </p>
                </li>
            </ul>
        """
    )

    run_button.click(
        fn=run,
        inputs=input_image,
        outputs=output,
        api_name="caption",
    )

if __name__ == "__main__":
    demo.queue(max_size=20).launch()