app.py

import torch
from transformers import AutoProcessor, AutoModelForVision2Seq
from PIL import Image
import gradio as gr

# Define the folder where the model and processor are saved
saved_folder_path = "model_folder"  # Replace with the path to your model folder

# Load processor and model
processor = AutoProcessor.from_pretrained(saved_folder_path)  # Processor (e.g., feature extractor + tokenizer)
model = AutoModelForVision2Seq.from_pretrained(saved_folder_path)  # Pre-trained BLIP model
model.eval()  # Set model to evaluation mode

# Define the caption generation function
def generate_caption(image):
    # Convert the input image to PIL format (if necessary)
    image = Image.fromarray(image)

    # Preprocess the image using the processor
    inputs = processor(images=image, return_tensors="pt")
    pixel_values = inputs.pixel_values

    # Generate caption
    generated_ids = model.generate(pixel_values=pixel_values, max_length=50)
    generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

    return generated_caption

# Define the Gradio interface
interface = gr.Interface(
    fn=generate_caption,  # Function to process input
    inputs=gr.Image(),    # Input as image
    outputs=gr.Textbox(), # Output as text
    live=True             # Enable live prediction
)

# Launch the Gradio app
interface.launch()