paddle-NLP 文档分析

发票抽取问答、海报抽取问答、网页抽取问答、表格抽取问答、试卷抽取问答。

from pprint import pprint
from paddlenlp import Taskflow
import gradio as gr
from paddlenlp import Taskflow
import numpy as np
from PIL import Image
import uuid

docprompt = Taskflow("document_intelligence")

def model_inference(image, prompt):
    prompt = prompt.split("?")[:-1]

    # TODO:非得要个网络或本地地址，ndarray不行
    img = Image.fromarray(np.uint8(image))
    filename = "./image/" + str(uuid.uuid4()) + ".png"
    img.save(filename)

    res = docprompt([{"doc": filename, "prompt": prompt}])
    json_out = {"result": res}
    return image,json_out


def clear_all():
    return None, None, None


with gr.Blocks() as demo:
    gr.Markdown("ERNIE-Layout")
    with gr.Column(scale=1, min_width=100):
        img_in = gr.Image(value="https://bj.bcebos.com/paddlenlp/taskflow/document_intelligence/images/invoice.jpg",
                          label="Input")

        text = gr.Textbox(
            value="发票号码是多少?校验码是多少?",
            label="输入问题:",
            lines=2)

        with gr.Row():
            btn1 = gr.Button("Clear")
            btn2 = gr.Button("Submit")
        json_out = gr.JSON(label="Information Extraction Output")

    img_out = gr.Image(label="Output").style(height=400)

    btn1.click(fn=clear_all, inputs=None, outputs=[img_in, img_out, json_out])
    btn2.click(fn=model_inference, inputs=[img_in, text], outputs=[img_out,json_out])
    gr.Button.style(1)

demo.launch(server_port=7007)