import streamlit as st
from transformers import pipeline
from PIL import Image
import torch
from transformers import ViTFeatureExtractor, AutoTokenizer, VisionEncoderDecoderModel

loc = "ydshieh/vit-gpt2-coco-en"
pipeline = pipeline(model=loc)

feature_extractor = ViTFeatureExtractor.from_pretrained(loc)
tokenizer = AutoTokenizer.from_pretrained(loc)
model = VisionEncoderDecoderModel.from_pretrained(loc)
model.eval()

def predict(image):

    pixel_values = feature_extractor(images=image, return_tensors="pt").pixel_values

    with torch.no_grad():
        output_ids = model.generate(pixel_values, max_length=1000, num_beams=4, return_dict_in_generate=True).sequences

    preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
    preds = [pred.strip() for pred in preds]

    return preds

file_name = st.file_uploader("Upload")

    
if file_name is not None:
    col1, col2 = st.columns(2)

    image = Image.open(file_name)
    col1.image(image, use_column_width = True)
    

    col2.header("Description")
    st.write(predict(image))