from PIL import Image
from transformers import VisionEncoderDecoderModel, ViTFeatureExtractor, PreTrainedTokenizerFast
import requests

model = VisionEncoderDecoderModel.from_pretrained("Zayn/vit2distilgpt2")
vit_feature_extractor = ViTFeatureExtractor.from_pretrained("google/vit-base-patch16-224-in21k")
tokenizer = PreTrainedTokenizerFast.from_pretrained("distilgpt2")

def vit2distilgpt2(img):
  pixel_values = vit_feature_extractor(images=img, return_tensors="pt").pixel_values
  encoder_outputs = model.generate(pixel_values.to('cpu'),num_beams=5)
  generated_sentences = tokenizer.batch_decode(encoder_outputs, skip_special_tokens =True)

  return(generated_sentences[0].split('.')[0])

import gradio as gr

inputs = [
    gr.inputs.Image(type="pil", label = "Original Image")
]

outputs = [
    gr.outputs.Textbox(label = 'Caption')
]   
title = "Image Captioning using Vision Transformer and GPT-2"
description = "Developed by Zayn"
article = "< a href='https://huggingface.co/Zayn/vit2distilgpt2'>Hugging Face AI Community</a>" 
examples = [
    ["car.jpg"]
]
gr.Interface(
    vit2distilgpt2,
    inputs,
    outputs,
    title = title,
    description = description,
    article = article,
    examples = examples,
    theme = "huggingface",   
).launch(debug=True,enable_queue=True)