app.py

import streamlit as st
import pandas as pd
import numpy as np
import os
import torch
from grobidmonkey import reader

from transformers import pipeline
from transformers import BartTokenizer, BartModel, BartForConditionalGeneration
from transformers import T5Tokenizer, T5ForConditionalGeneration

from document import Document
from BartSE import BARTAutoEncoder


def save_uploaded_file(uploaded_file):
    file_path = os.path.join("./uploads", uploaded_file.name)
    os.makedirs("./uploads", exist_ok=True)  # Create 'uploads' directory if it doesn't exist
    with open(file_path, "wb") as f:
        f.write(uploaded_file.getbuffer())
    return file_path  # Return the file path as a string

st.title('Paper2Slides')

st.subheader('Upload paper in pdf format')

col1, col2 = st.columns([3, 1])
with col1:
    uploaded_file = st.file_uploader("Choose a file")
with col2:
    option = st.selectbox(
        'Select parsing method.',
        ('monkey', 'x2d', 'lxml'))

if uploaded_file is not None:
    
    st.write(uploaded_file.name)
    bytes_data = uploaded_file.getvalue()
    st.write(len(bytes_data), "bytes")
    
    saved_file_path = save_uploaded_file(uploaded_file)
    monkeyReader = reader.MonkeyReader(option)
    outline = monkeyReader.readOutline(saved_file_path)
    for pre, fill, node in outline:
        st.write("%s%s" % (pre, node.name))
    
    
    # read paper content
    essay = monkeyReader.readEssay(saved_file_path)
    for key, values in essay.items():
        st.write(f"{key}: {', '.join(values)}")
        
    # with st.status("Understanding paper..."):
        
    #     Barttokenizer = BartTokenizer.from_pretrained('facebook/bart-large-cnn')
    #     summ_model_path = 'com3dian/Bart-large-paper2slides-summarizer'
    #     summarizor = BartForConditionalGeneration.from_pretrained(summ_model_path)
    #     exp_model_path = 'com3dian/Bart-large-paper2slides-expander'
    #     expandor = BartForConditionalGeneration.from_pretrained(exp_model_path)
    #     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    #     BartSE = BARTAutoEncoder(summarizor, summarizor, device)
    #     del summarizor, expandor
            
    #     document = Document(essay, Barttokenizer)
    #     del Barttokenizer
    #     length = document.merge(25, 30, BartSE, device)

    # with st.status("Generating slides..."):
    #     summarizor = pipeline("summarization", model=summ_model_path, device = device)
    #     summ_text = summarizor(document.segmentation['text'], max_length=100, min_length=10, do_sample=False)
    #     summ_text = [text['summary_text'] for text in summ_text]
        
    #     for summ in summ_text:
    #         st.write(summ)