to-be's picture
Update app.py
e45a37b
raw
history blame
4.05 kB
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()