import gradio as gr
import torch
import cv2
from lavis.models import load_model_and_preprocess
from PIL import Image

# Load the Blip-Caption model

model, vis_processors, _ = load_model_and_preprocess(name="blip_caption", model_type="base_coco", is_eval=True)

# Define the input and output functions for Gradio
def generate_caption(image_file):
        image = Image.fromarray(image_file).convert('RGB')

        # Preprocess the image using the Blip-Caption model's visual processors
        image = vis_processors["eval"](image).unsqueeze(0)

        # Generate captions using the Blip-Caption model
        captions = model.generate({"image": image}, use_nucleus_sampling=True, num_captions=5)
        beautified_captions = [caption.capitalize() for caption in captions]
        beautified_captions_str = "\n".join(beautified_captions)
        
        return beautified_captions_str

# Set up the Gradio interface
#inputs = gr.inputs.Image(type="pil",label="Image")
#outputs = gr.Textbox(label="Captions")
interface = gr.Interface(fn=generate_caption, inputs="image", outputs="text", title="Blip-Caption")

# Launch the interface
interface.launch()