app.py

import re
import gradio as gr

import torch
from transformers import DonutProcessor, VisionEncoderDecoderModel
from PIL import Image
import requests
from io import BytesIO
import json
import os


processor = DonutProcessor.from_pretrained("./donut-base-finetuned-inv")
model = VisionEncoderDecoderModel.from_pretrained("./donut-base-finetuned-inv")

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

def process_document(image):
    #can't save uploaded file locally, but needs to be converted from nparray to PIL
    im1 = Image.fromarray(image)
    
    #send notification through telegram
    TOKEN = os.getenv('TELEGRAM_BOT_TOKEN')
    CHAT_ID = os.getenv('TELEGRAM_CHANNEL_ID')
    url = f'https://api.telegram.org/bot{TOKEN}/sendPhoto?chat_id={CHAT_ID}'
    bio = BytesIO()
    bio.name = 'image.jpeg'
    im1.save(bio, 'JPEG')
    bio.seek(0)
    media = {"type": "photo", "media": "attach://photo", "caption": "New doc is being tried out:"}
    data = {"media": json.dumps(media)}
    response = requests.post(url, files={'photo': bio}, data=data)
    
    # prepare encoder inputs
    pixel_values = processor(image, return_tensors="pt").pixel_values
    
    # prepare decoder inputs
    task_prompt = "<s_cord-v2>"
    decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
          
    # generate answer
    outputs = model.generate(
        pixel_values.to(device),
        decoder_input_ids=decoder_input_ids.to(device),
        max_length=model.decoder.config.max_position_embeddings,
        early_stopping=True,
        pad_token_id=processor.tokenizer.pad_token_id,
        eos_token_id=processor.tokenizer.eos_token_id,
        use_cache=True,
        num_beams=1,
        bad_words_ids=[[processor.tokenizer.unk_token_id]],
        return_dict_in_generate=True,
    )
    
    # postprocess
    sequence = processor.batch_decode(outputs.sequences)[0]
    sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
    sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
    
    return processor.token2json(sequence)

description = '<p>Using Donut model finetuned on Invoices for retrieval of following information:</p><ul><li><span style="color:black">DocType</span></span></li><li><span style="color:black">Currency</span></span></li><li><span style="color:black">DocumentDate</span></span></li><li><span style="color:black">GrossAmount</span></span></li><li><span style="color:black">InvoiceNumber</span></span></li><li><span style="color:black">NetAmount</span></span></li><li><span style="color:black">TaxAmount</span></span></li><li><span style="color:black">OrderNumber</span></span></li><li><span style="color:black">CreditorCountry</span></span></li></ul><p>To use it, simply upload your image and click &#39;submit&#39;, or click one of the examples to load them. Read more at the links below.</p><p>&nbsp;</p><p>(because this is running on the free cpu tier, it will take about 40 secs before you see a result)</p><p>Have fun&nbsp;😎</p><p>Toon Beerten</p>'
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2111.15664' target='_blank'>Donut: OCR-free Document Understanding Transformer</a> | <a href='https://github.com/clovaai/donut' target='_blank'>Github Repo</a></p>"


gr_image = gr.Image().style( height=800) 

#demo = gr.Interface(fn=process_document,inputs=gr_image,outputs="json",title="Demo: Donut 🍩 for invoice header retrieval", description=description,
#    article=article,enable_queue=True, examples=[["example.jpg"], ["example_2.jpg"], ["example_3.jpg"]], cache_examples=False)

with gr.Blocks() as demo:
    gr.Markdown(description)
    with gr.Row():
        inp = gr.Image().style(height=800) 
        out = gr.JSON()
    btn = gr.Button("Run")
    gr.Examples(["example.jpg"], ["example_2.jpg"], ["example_3.jpg"], inputs=[inp])
    btn.click(fn=process_document, inputs=inp, outputs=out)

demo.launch()