app.py

import gradio as gr
from transformers import VisionEncoderDecoderModel, ViTImageProcessor, AutoTokenizer
import torch
from PIL import Image

# Load model and tokenizer
model = VisionEncoderDecoderModel.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
feature_extractor = ViTImageProcessor.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
tokenizer = AutoTokenizer.from_pretrained("nlpconnect/vit-gpt2-image-captioning")

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

# Define generation parameters
max_length = 16
num_beams = 4
gen_kwargs = {"max_length": max_length, "num_beams": num_beams}

def predict_step(image):
    # Convert image from numpy array to PIL image
    i_image = Image.fromarray(image)
    if i_image.mode != "RGB":
        i_image = i_image.convert(mode="RGB")
    
    # Prepare image for the model
    pixel_values = feature_extractor(images=[i_image], return_tensors="pt").pixel_values
    pixel_values = pixel_values.to(device)

    # Generate prediction
    output_ids = model.generate(pixel_values, **gen_kwargs)
    preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
    preds = [pred.strip() for pred in preds]

    return preds[0]

# Define Gradio interface
interface = gr.Interface(
    fn=predict_step,
    inputs=gr.Image(type="numpy"),  # Updated to use gr.Image
    outputs="text",
    title="Image-to-Text Conversion",
    description="Upload an image, and this model will generate a textual description of the image."
)

# Launch the interface
interface.launch()