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), image

title = '<h1 style="text-align:center"><img alt="" src="circling_small.gif" />Welcome<img alt="" src="circling2_small.gif" /></h1>'
paragraph1 = '<p>Basic idea of this 🍩 model is to give it an image as input and extract indexes as text. No bounding boxes or confidences are generated.<br /> For more info, see the <a href="https://arxiv.org/abs/2111.15664">original paper</a>&nbsp;and the 🤗&nbsp;<a href="https://huggingface.co/naver-clova-ix/donut-base">model</a>.</p>'
paragraph2 = '<p><strong>Training</strong>:<br />The model was trained with a few thousand of annotated invoices and non-invoices (for those the doctype will be &#39;Other&#39;). They span across different countries and languages. They are always one page only. The dataset is proprietary unfortunately.&nbsp;Model is set to input resolution of 1280x1920 pixels. So any sample you want to try with higher dpi than 150 has no added value.<br />It was trained for about 4 hours on a&nbsp;NVIDIA RTX A4000 for 20k steps with a val_metric of&nbsp;0.03413819904382196 at the end.<br />The <u>following indexes</u> were included in the train set:</p><ul><li><span style="font-family:Calibri"><span style="color:black">DocType</span></span></li><li><span style="font-family:Calibri"><span style="color:black">Currency</span></span></li><li><span style="font-family:Calibri"><span style="color:black">DocumentDate</span></span></li><li><span style="font-family:Calibri"><span style="color:black">GrossAmount</span></span></li><li><span style="font-family:Calibri"><span style="color:black">InvoiceNumber</span></span></li><li><span style="font-family:Calibri"><span style="color:black">NetAmount</span></span></li><li><span style="font-family:Calibri"><span style="color:black">TaxAmount</span></span></li><li><span style="font-family:Calibri"><span style="color:black">OrderNumber</span></span></li><li><span style="font-family:Calibri"><span style="color:black">CreditorCountry</span></span></li></ul>'
#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)
paragraph3 = '<p><strong>Try it out:</strong><br />To use it, simply upload your image and click &#39;submit&#39;, or click one of the examples to load them.<br /><em>(because this is running on the free cpu tier, it will take about 40 secs before you see a result)</em></p><p>&nbsp;</p><p>Have fun&nbsp;😎</p><p>Toon Beerten</p>'

css = "#inp {height: auto !important; width: 100% !important;}"
# css = "@media screen and (max-width: 600px) { .output_image, .input_image {height:20rem !important; width: 100% !important;} }"
# css = ".output_image, .input_image {height: 600px !important}"

#css = ".image-preview {height: auto !important;}"


with gr.Blocks(css=css) as demo:
    
    gr.HTML(title)
    gr.HTML(paragraph1)
    gr.HTML(paragraph2)
    gr.HTML(paragraph3)
    
    with gr.Row().style():
        with gr.Column(scale=1):
            inp = gr.Image(label='Upload invoice here:')   #.style(height=400)          
        with gr.Column(scale=2):
             gr.Examples([["example.jpg"], ["example_2.jpg"], ["example_3.jpg"]], inputs=[inp],label='Or use one of these examples:')
    with gr.Row().style():       
             btn = gr.Button("↓   Extract   ↓")
    with gr.Row(css='div {margin-left: auto; margin-right: auto; width: 100%;background-image: url("background.gif"); repeat 0 0;}').style():
        with gr.Column(scale=2):
            imgout = gr.Image(label='Uploaded document:',elem_id="inp")
        with gr.Column(scale=1):
            jsonout = gr.JSON(label='Extracted information:')
            
    btn.click(fn=process_document, inputs=inp, outputs=[jsonout,imgout])

demo.launch()