File size: 3,330 Bytes
a1c32b3
 
 
 
 
 
 
 
 
 
 
 
491a9c3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a1c32b3
 
 
 
e853e36
 
 
 
 
 
 
 
 
 
 
a1c32b3
 
e853e36
a1c32b3
 
 
 
 
c52d6b0
a1c32b3
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
import re
import gradio as gr

import torch
from transformers import DonutProcessor, VisionEncoderDecoderModel

processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")

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

def process_document(image, *questions):
    output = []
    for question in questions:
        # prepare encoder inputs
        pixel_values = processor(image, return_tensors="pt").pixel_values
        
        # prepare decoder inputs
        task_prompt = "<s_docvqa><s_question>{user_input}</s_question><s_answer>"
        prompt = task_prompt.replace("{user_input}", question)
        decoder_input_ids = processor.tokenizer(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
        
        output.append(processor.token2json(sequence))
    return output

description = "Gradio Demo for Donut, an instance of `VisionEncoderDecoderModel` fine-tuned on DocVQA (document visual question answering). To use it, simply upload your image and type a question and click 'submit', or click one of the examples to load them. Read more at the links below."
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>"

vqa_questions = {}
vqa_questions["ACCOUNT/BILL NUMBER"] = "What is the account or bill number?"
vqa_questions["TOTAL"] = "What is the total amount or total price?"
vqa_questions["ITEMS"] = "What are the items?"
vqa_questions["GST AMOUNT"] = "What is the GST or tax amount?"
vqa_questions["GST NO."] = "What is the GST number?"
vqa_questions[
    "SELLER/BILLING DETAILS"
] = "What are the seller details or billing details"
vqa_questions["BILLING ADDRESS"] = "What is the billing address?"

demo = gr.Interface(
    fn=process_document,
        inputs=["image"] +  [gr.components.Textbox(value=question) for question in vqa_questions.values()],
    outputs="json",
    title="Demo: Donut 🍩 for DocVQA",
    description=description,
    article=article,
    enable_queue=True,
    # examples=[["example_3.jpg", "What is the total?"], ["example_1.png", "When is the coffee break?"], ["example_2.jpeg", "What's the population of Stoddard?"]],
    cache_examples=False)

demo.launch()