deprem-ocr-test / app.py
Goodsea's picture
add paddleocr optimized inference pipeline
863f1e9
raw history blame
No virus
4.74 kB
from deprem_ocr.ocr import DepremOCR
import numpy as np
import gradio as gr
import openai
import ast
import os
from openai_api import OpenAI_API
import utils
openai.api_key = os.getenv("API_KEY")
depremOCR = DepremOCR()
def get_text(input_img):
result = depremOCR.apply_ocr(np.array(input_img))
print(result)
return " ".join(result)
# Submit button
def get_parsed_address(input_img):
address_full_text = get_text(input_img)
return openai_response(address_full_text)
# Open API on change
def text_dict(input):
eval_result = ast.literal_eval(input)
utils.write_db(eval_result)
return (
str(eval_result["city"]),
str(eval_result["distinct"]),
str(eval_result["neighbourhood"]),
str(eval_result["street"]),
str(eval_result["address"]),
str(eval_result["tel"]),
str(eval_result["name_surname"]),
str(eval_result["no"]),
)
def openai_response(ocr_input):
prompt = f"""Tabular Data Extraction You are a highly intelligent and accurate tabular data extractor from
plain text input and especially from emergency text that carries address information, your inputs can be text
of arbitrary size, but the output should be in [{{'tabular': {{'entity_type': 'entity'}} }}] JSON format Force it
to only extract keys that are shared as an example in the examples section, if a key value is not found in the
text input, then it should be ignored. Have only city, distinct, neighbourhood,
street, no, tel, name_surname, address Examples: Input: Deprem sırasında evimizde yer alan adresimiz: İstanbul,
Beşiktaş, Yıldız Mahallesi, Cumhuriyet Caddesi No: 35, cep telefonu numaram 5551231256, adim Ahmet Yilmaz
Output: {{'city': 'İstanbul', 'distinct': 'Beşiktaş', 'neighbourhood': 'Yıldız Mahallesi', 'street': 'Cumhuriyet Caddesi', 'no': '35', 'tel': '5551231256', 'name_surname': 'Ahmet Yılmaz', 'address': 'İstanbul, Beşiktaş, Yıldız Mahallesi, Cumhuriyet Caddesi No: 35'}}
Input: {ocr_input}
Output:
"""
openai_client = OpenAI_API()
response = openai_client.single_request(prompt)
resp = response["choices"][0]["text"]
print(resp)
resp = eval(resp.replace("'{", "{").replace("}'", "}"))
resp["input"] = ocr_input
dict_keys = [
"city",
"distinct",
"neighbourhood",
"street",
"no",
"tel",
"name_surname",
"address",
"input",
]
for key in dict_keys:
if key not in resp.keys():
resp[key] = ""
return resp
def ner_response(ocr_input):
API_URL = "https://api-inference.huggingface.co/models/deprem-ml/deprem-ner"
headers = {"Authorization": "Bearer xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"}
def query(payload):
response = requests.post(API_URL, headers=headers, json=payload)
return response.json()
output = query({
"inputs": ocr_input,
})
return output
# User Interface
with gr.Blocks() as demo:
gr.Markdown(
"""
# Enkaz Bildirme Uygulaması
"""
)
gr.Markdown(
"Bu uygulamada ekran görüntüsü sürükleyip bırakarak AFAD'a enkaz bildirimi yapabilirsiniz. Mesajı metin olarak da girebilirsiniz, tam adresi ayrıştırıp döndürür. API olarak kullanmak isterseniz sayfanın en altında use via api'ya tıklayın."
)
with gr.Row():
img_area = gr.Image(label="Ekran Görüntüsü yükleyin 👇")
ocr_result = gr.Textbox(label="Metin yükleyin 👇 ")
open_api_text = gr.Textbox(label="Tam Adres")
submit_button = gr.Button(label="Yükle")
with gr.Column():
with gr.Row():
city = gr.Textbox(label="İl")
distinct = gr.Textbox(label="İlçe")
with gr.Row():
neighbourhood = gr.Textbox(label="Mahalle")
street = gr.Textbox(label="Sokak/Cadde/Bulvar")
with gr.Row():
tel = gr.Textbox(label="Telefon")
with gr.Row():
name_surname = gr.Textbox(label="İsim Soyisim")
address = gr.Textbox(label="Adres")
with gr.Row():
no = gr.Textbox(label="Kapı No")
submit_button.click(
get_parsed_address,
inputs=img_area,
outputs=open_api_text,
api_name="upload_image",
)
ocr_result.change(
openai_response, ocr_result, open_api_text, api_name="upload-text"
)
open_api_text.change(
text_dict,
open_api_text,
[city, distinct, neighbourhood, street, address, tel, name_surname, no],
)
if __name__ == "__main__":
demo.launch()