File size: 2,954 Bytes
ebd3f73
 
 
 
 
 
 
 
 
 
 
 
 
f263316
a91ce54
 
 
 
ebd3f73
 
a91ce54
ebd3f73
 
 
 
 
 
 
 
 
 
 
 
 
 
3b5e294
ebd3f73
3b5e294
 
 
ebd3f73
 
6335c31
 
 
 
ebd3f73
 
 
 
b7d24be
ebd3f73
 
 
 
b7d24be
 
f263316
 
 
 
b7d24be
 
 
ebd3f73
 
 
 
 
 
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 gradio as gr
import re
from transformers import DonutProcessor, VisionEncoderDecoderModel
import torch
from PIL import Image

def process_filename(filename, question):
    print(f"Image file: {filename}")
    print(f"Question: {question}")
    image = Image.open(filename).convert("RGB")
    return process_image(image)


def process_image(set_use_cache, set_return_dict_in_generate, set_early_stopping, set_output_scores, image, question):
    repo_id = "naver-clova-ix/donut-base-finetuned-docvqa"
    print(f"Model repo: {repo_id}")
    processor = DonutProcessor.from_pretrained(repo_id)
    model = VisionEncoderDecoderModel.from_pretrained(repo_id)

    device = "cuda" if torch.cuda.is_available() else "cpu"
    print(f"Device used: {device}")
    model.to(device)

    # prepare decoder inputs
    prompt = f"<s_docvqa><s_question>{question}</s_question><s_answer>"
    decoder_input_ids = processor.tokenizer(prompt, add_special_tokens=False, return_tensors="pt").input_ids

    pixel_values = processor(image, return_tensors="pt").pixel_values

    outputs = model.generate(
        pixel_values.to(device),
        decoder_input_ids=decoder_input_ids.to(device),
        max_length=model.decoder.config.max_position_embeddings,
        pad_token_id=processor.tokenizer.pad_token_id,
        eos_token_id=processor.tokenizer.eos_token_id,
        use_cache=set_use_cache=="True",
        bad_words_ids=[[processor.tokenizer.unk_token_id]],
        return_dict_in_generate=set_return_dict_in_generate=="True",
        early_stopping=set_early_stopping=="True",
        output_scores=set_output_scores=="True"
    )

    sequence_data = processor.batch_decode(outputs.sequences)
    print(f"Sequence data: {sequence_data}")
    sequence = sequence_data[0]
    print(f"Sequence: {sequence}")
    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
    print(processor.token2json(sequence))

    return processor.token2json(sequence)['answer']

description = "DocVQA (document visual question answering)"

demo = gr.Interface(
    fn=process_image,
    inputs=[
        gr.Radio(["True", "False"], value="True", label="Use cache", info="Define model.generate() use_cache value"), 
        gr.Radio(["True", "False"], value="True", label="Dict in generate", info="Define model.generate() return_dict_in_generate value"),
        gr.Radio(["True", "False"], value="True", label="Early stopping", info="Define model.generate() early_stopping value"),
        gr.Radio(["True", "False"], value="True", label="Output scores", info="Define model.generate() output_scores value"),
        "image", 
        gr.Textbox(label = "Question" )
    ],
    outputs=gr.Textbox(label = "Response" ),
    title="Extract data from image",
    description=description,
    cache_examples=True)

demo.launch()