app.py

import os

os.environ["TOKENIZERS_PARALLELISM"] = "false"

from PIL import Image, ImageDraw
import traceback

import gradio as gr
from gradio import processing_utils

import torch
from docquery import pipeline
from docquery.document import load_bytes, load_document, ImageDocument
from docquery.ocr_reader import get_ocr_reader


def ensure_list(x):
    if isinstance(x, list):
        return x
    else:
        return [x]


CHECKPOINTS = {
    "LayoutLMv1 for Invoices 🧾": "impira/layoutlm-invoices",
}

PIPELINES = {}


def construct_pipeline(task, model):
    global PIPELINES
    if model in PIPELINES:
        return PIPELINES[model]

    device = "cuda" if torch.cuda.is_available() else "cpu"
    ret = pipeline(task=task, model=CHECKPOINTS[model], device=device)
    PIPELINES[model] = ret
    return ret


def run_pipeline(model, question, document, top_k):
    pipeline = construct_pipeline("document-question-answering", model)
    return pipeline(question=question, **document.context, top_k=top_k)


# TODO: Move into docquery
# TODO: Support words past the first page (or window?)
def lift_word_boxes(document, page):
    return document.context["image"][page][1]


def expand_bbox(word_boxes):
    if len(word_boxes) == 0:
        return None

    min_x, min_y, max_x, max_y = zip(*[x[1] for x in word_boxes])
    min_x, min_y, max_x, max_y = [min(min_x), min(min_y), max(max_x), max(max_y)]
    return [min_x, min_y, max_x, max_y]


# LayoutLM boxes are normalized to 0, 1000
def normalize_bbox(box, width, height, padding=0.005):
    min_x, min_y, max_x, max_y = [c / 1000 for c in box]
    if padding != 0:
        min_x = max(0, min_x - padding)
        min_y = max(0, min_y - padding)
        max_x = min(max_x + padding, 1)
        max_y = min(max_y + padding, 1)
    return [min_x * width, min_y * height, max_x * width, max_y * height]


EXAMPLES = [
    [
        "invoice.png",
        "Invoice 1",
    ],
    [
        "contract.jpeg",
        "What is the purchase amount?",
    ],
    [
        "statement.png",
        "What are net sales for 2020?",
    ],
]

QUESTION_FILES = {}

FIELDS = {
    "Vendor Name": ["Vendor Name - Logo?", "Vendor Name - Address?"],
    "Vendor Address": ["Vendor Address?"],
    "Invoice Number": ["Invoice Number?"],
    "Invoice Date": ["Invoice Date?"],
    "Due Date": ["Due Date?"],
    "Subtotal": ["Subtotal?"],
    "Total Tax": ["Total Tax?"],
    "Invoice Total": ["Invoice Total?"],
    "Amount Due": ["Amount Due?"],
    "Payment Terms": ["Payment Terms?"],
}


def empty_table(fields):
    return {"value": [[name, None] for name in fields.keys()], "interactive": False}


def process_document(document, fields, model, error=None):
    if document is not None and error is None:
        preview, json_output, table = process_fields(document, fields, model)
        return (
            document,
            fields,
            preview,
            gr.update(visible=True),
            gr.update(visible=False, value=None),
            json_output,
            table,
        )
    else:
        return (
            None,
            fields,
            None,
            gr.update(visible=False),
            gr.update(visible=True, value=error) if error is not None else None,
            None,
            gr.update(**empty_table(fields)),
        )


def process_path(path, fields, model):
    error = None
    document = None
    if path:
        try:
            document = load_document(path)
        except Exception as e:
            traceback.print_exc()
            error = str(e)

    return process_document(document, fields, model, error)


def process_upload(file, fields, model):
    return process_path(file.name if file else None, fields, model)


colors = ["#64A087", "green", "black"]


def annotate_page(prediction, pages, document):
    if "word_ids" in prediction:
        image = pages[prediction["page"]]
        draw = ImageDraw.Draw(image, "RGBA")
        word_boxes = lift_word_boxes(document, prediction["page"])
        x1, y1, x2, y2 = normalize_bbox(
            expand_bbox([word_boxes[i] for i in prediction["word_ids"]]),
            image.width,
            image.height,
        )
        draw.rectangle(((x1, y1), (x2, y2)), fill=(0, 255, 0, int(0.4 * 255)))


def process_question(
    question, document, img_gallery, model, fields, output, output_table
):
    if not question or document is None:
        return None, None, None

    text_value = None
    pages = [processing_utils.decode_base64_to_image(p) for p in img_gallery]
    prediction = run_pipeline(model, question, document, 1)
    annotate_page(prediction, pages, document)

    field_name = question.rstrip("?")
    fields = {field_name: [question], **fields}
    output = {field_name: prediction, **output}
    table = [[field_name, prediction.get("answer")]] + output_table.values.tolist()
    return (
        gr.update(visible=True, value=pages),
        fields,
        output,
        gr.update(value=table, interactive=False),
    )


def process_fields(document, fields, model=list(CHECKPOINTS.keys())[0]):
    pages = [x.copy().convert("RGB") for x in document.preview]

    ret = {}
    table = []

    for (field_name, questions) in fields.items():
        answers = [run_pipeline(model, q, document, top_k=1) for q in questions]
        answers.sort(key=lambda x: -x.get("score", 0) if x else 0)
        top = answers[0]
        annotate_page(top, pages, document)
        ret[field_name] = top
        table.append([field_name, top.get("answer") if top is not None else None])

    return (
        gr.update(visible=True, value=pages),
        gr.update(visible=True, value=ret),
        gr.update(visible=True, value=table),
    )


def load_example_document(img, title, fields, model):
    document = None
    if img is not None:
        if title in QUESTION_FILES:
            document = load_document(QUESTION_FILES[title])
        else:
            document = ImageDocument(Image.fromarray(img), ocr_reader=get_ocr_reader())

    return process_document(document, fields, model)


CSS = """
#question input {
    font-size: 16px;
}
#url-textbox, #question-textbox {
    padding: 0 !important;
}
#short-upload-box .w-full {
    min-height: 10rem !important;
}
/* I think something like this can be used to re-shape
 * the table
 */
/*
.gr-samples-table tr {
    display: inline;
}
.gr-samples-table .p-2 {
    width: 100px;
}
*/
#select-a-file {
    width: 100%;
}
#file-clear {
    padding-top: 2px !important;
    padding-bottom: 2px !important;
    padding-left: 8px !important;
    padding-right: 8px !important;
	margin-top: 10px;
}
.gradio-container .gr-button-primary {
    background: linear-gradient(180deg, #CDF9BE 0%, #AFF497 100%);
    border: 1px solid #B0DCCC;
    border-radius: 8px;
    color: #1B8700;
}
.gradio-container.dark button#submit-button {
    background: linear-gradient(180deg, #CDF9BE 0%, #AFF497 100%);
    border: 1px solid #B0DCCC;
    border-radius: 8px;
    color: #1B8700
}

table.gr-samples-table tr td {
    border: none;
    outline: none;
}

table.gr-samples-table tr td:first-of-type {
    width: 0%;
}

div#short-upload-box div.absolute {
    display: none !important;
}

gradio-app > div > div > div > div.w-full > div, .gradio-app > div > div > div > div.w-full > div {
    gap: 0px 2%;
}

gradio-app div div div div.w-full, .gradio-app div div div div.w-full {
    gap: 0px;
}

gradio-app h2, .gradio-app h2 {
    padding-top: 10px;
}

#answer {
    overflow-y: scroll;
    color: white;
    background: #666;
    border-color: #666;
    font-size: 20px;
    font-weight: bold;
}

#answer span {
    color: white;
}

#answer textarea {
    color:white;
    background: #777;
    border-color: #777;
    font-size: 18px;
}

#url-error input {
    color: red;
}
"""

with gr.Blocks(css=CSS) as demo:
    gr.Markdown("# DocQuery: Document Query Engine")
    gr.Markdown(
        "DocQuery (created by [Impira](https://impira.com)) uses LayoutLMv1 fine-tuned on an invoice dataset"
        " as well as DocVQA and SQuAD, which boot its general comprehension skills. The model is an enhanced"
        " QA architecture that supports selecting blocks of text which may be non-consecutive, which is a major"
        " issue when dealing with invoice documents (e.g. addresses)."
        " To use it, simply upload an image or PDF invoice and the model will predict values for several fields."
        " You can also create additional fields by simply typing in a question."
        " DocQuery is available on [Github](https://github.com/impira/docquery)."
    )

    document = gr.Variable()
    fields = gr.Variable(value={**FIELDS})
    example_question = gr.Textbox(visible=False)
    example_image = gr.Image(visible=False)

    with gr.Row(equal_height=True):
        with gr.Column():
            with gr.Row():
                gr.Markdown("## 1. Select an invoice", elem_id="select-a-file")
                img_clear_button = gr.Button(
                    "Clear", variant="secondary", elem_id="file-clear", visible=False
                )
            image = gr.Gallery(visible=False)
            with gr.Row(equal_height=True):
                with gr.Column():
                    with gr.Row():
                        url = gr.Textbox(
                            show_label=False,
                            placeholder="URL",
                            lines=1,
                            max_lines=1,
                            elem_id="url-textbox",
                        )
                        submit = gr.Button("Get")
                    url_error = gr.Textbox(
                        visible=False,
                        elem_id="url-error",
                        max_lines=1,
                        interactive=False,
                        label="Error",
                    )
            gr.Markdown("— or —")
            upload = gr.File(label=None, interactive=True, elem_id="short-upload-box")
            gr.Examples(
                examples=EXAMPLES,
                inputs=[example_image, example_question],
            )

        with gr.Column() as col:
            with gr.Tabs():
                with gr.TabItem("Table"):
                    output_table = gr.Dataframe(
                        headers=["Field", "Value"], **empty_table(fields.value)
                    )

                with gr.TabItem("JSON"):
                    output = gr.JSON(label="Output", visible=False)

            gr.Markdown("## 2. Ask a question")
            model = gr.Radio(
                choices=list(CHECKPOINTS.keys()),
                value=list(CHECKPOINTS.keys())[0],
                label="Model",
                visible=False,
            )

            with gr.Row():
                question = gr.Textbox(
                    label="Question",
                    show_label=False,
                    placeholder="e.g. What is the invoice number?",
                    lines=1,
                    max_lines=1,
                    elem_id="question-textbox",
                )
                clear_button = gr.Button("Clear", variant="secondary", visible=False)
                submit_button = gr.Button(
                    "Add", variant="primary", elem_id="submit-button"
                )

    for cb in [img_clear_button, clear_button]:
        cb.click(
            lambda _: (
                gr.update(visible=False, value=None),  # image
                None,  # document
                # {**FIELDS},  # fields
                gr.update(visible=False, value=None),  # output
                # gr.update(**empty_table(FIELDS)),  # output_table
                # gr.update(visible=False),
                None,
                None,
                None,
                gr.update(visible=False, value=None),
                None,
            ),
            inputs=clear_button,
            outputs=[
                image,
                document,
                # fields,
                output,
                # output_table,
                # img_clear_button,
                example_image,
                upload,
                url,
                url_error,
                question,
            ],
        )

    submit_outputs = [
        document,
        fields,
        image,
        img_clear_button,
        url_error,
        output,
        output_table,
    ]

    upload.change(
        fn=process_upload,
        inputs=[upload, fields, model],
        outputs=submit_outputs,
    )

    submit.click(
        fn=process_path,
        inputs=[url, fields, model],
        outputs=submit_outputs,
    )

    question.submit(
        fn=process_question,
        inputs=[question, document, image, model, fields, output, output_table],
        outputs=[image, fields, output, output_table],
    )

    submit_button.click(
        process_question,
        inputs=[question, document, model],
        outputs=[image, output, output_table],
    )

    model.change(
        process_question,
        inputs=[question, document, model],
        outputs=[image, output, output_table],
    )

    example_image.change(
        fn=load_example_document,
        inputs=[example_image, example_question, fields, model],
        outputs=submit_outputs,
    )

if __name__ == "__main__":
    demo.launch(enable_queue=False)