app.py

from transformers import pipeline, AutoTokenizer
import io
import base64
from PIL import Image
import gradio as gr

model = "Salesforce/blip-image-captioning-large"
tokenizer = AutoTokenizer.from_pretrained(model, use_fast=True)

pipe = pipeline(task="image-to-text",
                model=model, 
                tokenizer=tokenizer) 

def image_to_base64(image: Image) -> str:
    """
    Convert an image to a base64 string.
    """
    bytearray= io.BytesIO()
    image.save(bytearray, format="PNG")
    return str(base64.b64encode(bytearray.getvalue()).decode('utf-8'))    


def caption_image(image):
    result = pipe(
        image_to_base64(image),
        #Temperature=0.7,
        # max_length=130,
        # min_length=30,
        #do_sample=True
    )
    return result[0]['generated_text'].upper()
    
if __name__ == "__main__":
    gr.close_all()
    
    with gr.Blocks() as interface:
        gr.Markdown("### Image Captioning using BLIP Large")
        with gr.Row():
            image_input = gr.Image(type="pil", label="Image")
        with gr.Row():
            caption_output = gr.Textbox(lines=2, label="Caption")
        with gr.Row():
            clear_button = gr.ClearButton()
            caption_button = gr.Button("Caption", variant="primary")

        with gr.Row():
            example_images = gr.Examples(
                examples=[
                    "data/image1.jpg",
                    "data/image2.png",
                    "data/image3.jpg",
                    "data/image4.jpg",
                    "data/image5.jpg",
                    "data/image6.png",
                    "data/image7.png",
                    "data/image8.jpeg",
                    "data/image9.jpeg",
                    "data/image10.jpg",
                ],
                inputs=[image_input],
                label="Example Images"
            )
        
        
        caption_button.click(fn=caption_image,
            inputs=[image_input],
            outputs=[caption_output]
        )
        
        clear_button.click(fn=lambda: [None,""],
                            inputs=[],
                            outputs=[image_input, caption_output])

    interface.launch(share=True, server_port=7860)