Metadata-Version: 2.1
Name: icare-nlp
Version: 0.0.7
Summary: From CV detection to answer questions
Author-email: 23037086r@connect.polyu.hk
Keywords: icare,language
Description-Content-Type: text/markdown

#### 1. 安装
##### (1). 方法1：使用pip安装
~~~
pip install icare-nlp
Pypi链接: https://pypi.org/project/icare-nlp/
~~~
##### (2). 方法2： 使用Source Codes安装
~~~
git clone https://github.com/YiyiyiZhao/icare_nlp_tools.git
cd icare_nlp_tools
pip install -e .
pip install -r requirements.txt
~~~

#### 2. 使用
##### 2.0 Task_Disp: 输入User query, 输出对应下列四种的Task类型
###### (1). Commands
~~~
from icare_nlp.task_disp import TaskDisp

task_disp=TaskDisp()
task_disp.disp_start()
~~~
~~~
#Intent classification
task_disp=TaskDisp()
user_query="可以確認下有冇糖不甩？"
task_disp.intent_classify(user_query)
#SYSTEM： 我會幫你完成呢個收據問答嘅任務。
~~~
###### (2). Demo
***Object Description and QA***
![](./examples/figs/object_desc_qa_0513.gif)
***Receipt Description and QA***
![](./examples/figs/receipt_desc_qa_0513.gif)
##### 2.1 Object_Desc: 输入object detection list, 输出场景播报
~~~
import json
from icare_nlp.object_desc import ObjectDesc
obj_desc=ObjectDesc()

with open("./obj_detect_files/59.json", "r") as f:
    obj_detect=json.load(f)
print(obj_detect)
#[{'position': [1149, 580, 258, 270], 'text': 'chair'}, {'position': [958, 186, 235, 171], 'text': 'tv'}, {'position': [1130, 399, 211, 132], 'text': 'chair'}, {'position': [198, 388, 153, 52], 'text': 'chair'}, {'position': [664, 609, 259, 211], 'text': 'chair'}, {'position': [869, 384, 123, 164], 'text': 'chair'}, {'position': [162, 508, 94, 163], 'text': 'bottle'}, {'position': [785, 309, 56, 36], 'text': 'chair'}, {'position': [620, 341, 152, 177], 'text': 'suitcase'}, {'position': [577, 608, 436, 210], 'text': 'chair'}, {'position': [357, 336, 83, 72], 'text': 'chair'}, {'position': [417, 508, 830, 404], 'text': 'dining table'}, {'position': [862, 545, 121, 178], 'text': 'handbag'}, {'position': [862, 545, 122, 177], 'text': 'backpack'}, {'position': [791, 389, 91, 184], 'text': 'chair'}]

obj_desc_res=obj_desc.form_response(obj_detect)
print(obj_desc_res)

#而家眼前嘅景象有9把椅, 1台電視, 1瓶樽, 1個行李箱, 1張飯枱, 1個手袋, 1個書包.視線左上角嘅場景入面有1個行李箱.視線左上角嘅場景入面有1把椅.視線右上角嘅場景入面有1台電視.視線右上角嘅場景入面有1把椅.視線左下角嘅場景入面有2把椅.視線左下角嘅場景入面有1瓶樽.視線左下角嘅場景入面有1張飯枱.視線右下角嘅場景入面有5把椅.視線右下角嘅場景入面有1個手袋.視線右下角嘅場景入面有1個書包.
~~~

##### 2.2 Object_QA: 输入object detection list 和 Question, 输出场景有关的Answer
~~~
import json
from icare_nlp.object_qa import ObjectQA
obj_qa=ObjectQA()
with open("./obj_detect_files/24.json", "r") as f:
    obj_detect=json.load(f)
print(obj_detect)
#[{'position': [999, 349, 213, 254], 'text': 'chair'}, {'position': [221, 64, 427, 125], 'text': 'tv'}, {'position': [72, 325, 144, 58], 'text': 'chair'}, {'position': [873, 221, 78, 49], 'text': 'chair'}, {'position': [101, 535, 177, 267], 'text': 'cup'}, {'position': [1013, 589, 211, 253], 'text': 'handbag'}, {'position': [663, 289, 242, 324], 'text': 'suitcase'}, {'position': [1231, 535, 96, 303], 'text': 'chair'}]


question="我點樣可以攞到椅子？"
obj_desc_res=obj_qa.form_response(question,obj_detect)
print(obj_desc_res)
#chair喺中心點嘅右上方. 椅子有一個堅硬嘅框架，通常由木頭或金屬製成，座位和背部軟。

question="手袋喺我手嘅邊個方向？"
obj_desc_res=obj_qa.form_response(question,obj_detect)
print(obj_desc_res)
#handbag喺中心點嘅右上方. 手袋嘅大小各異，由軟皮革至硬合成材料製成。

question="椅子附近有冇其他物體？"
obj_desc_res=obj_qa.form_response(question,obj_detect)
print(obj_desc_res)
#chair最近嘅物件系 chair,handbag,chair.chair喺chair嘅左上方.handbag喺chair嘅右下方.chair喺chair嘅右下方.. 椅子有一個堅硬嘅框架，通常由木頭或金屬製成，座位和背部軟。
~~~

##### 2.3 Receipt_Desc: 输入Receipt的OCR识别文本, 输出Receipt描述
~~~
import json
from icare_nlp.receipt_desc import ReceiptDesc
receipt_desc=ReceiptDesc()


with open("./ocr_detect_files/ocr_azure.json", "r") as f:
        ocr_data = json.load(f)
ocr_text = ""
for item in ocr_data:
    ocr_text += item["text"] + '\n'


rec_desc_res=receipt_desc.form_response(ocr_text)
print(rec_desc_res)
#呢张收据主要嘅信息包括外賣, 點餐時間, 收據號碼, 新加坡海南雞, 點餐號碼, 優惠碼, 当久於, 腸仔猪扒雞扒飯, 秘製燒汁, 白飯, 少鹽, 概沟, 小計, 折扣, 總金額, 付款資料, 付款方式, 扣除金額, 餘額, 卡號, 機號, 發票號碼, 交易時間, 绿联, 深圳市绿联科技股份有限公司, 地址深圳市龙华区龙观西路龙城工业区绿联办公大楼, 电话, 官网, 执行标准
~~~

##### 2.4 Receipt_QA: 输入Receipt的OCR识别文本 和 Question, 输出Answer [GPT-3.5 assisted]
~~~
#!/bin/bash
export OPEN_API_KEY="your_api_key_here"
~~~
~~~
import json
from icare_nlp.receipt_qa import ReceiptQA
receipt_qa=ReceiptQA()


with open("./ocr_detect_files/ocr_azure.json", "r") as f:
        ocr_data = json.load(f)
ocr_text = ""
for item in ocr_data:
    ocr_text += item["text"] + '\n'

question="我總共花咗幾多錢？"
rec_desc_res=receipt_qa.form_response(ocr_text, question)
print(rec_desc_res)
#51.0.
~~~
#### 3. Structure and Other Information
The core structure is: 
~~~
├── __init__.py
├── object_desc.py
├── object_qa.py
├── receipt_desc.py
├── receipt_qa.py
├── resources
├── task_disp.py
└── utils.py
~~~
There are some files for you to have a try:
If you want to try the object description and question answering functions:
~~~
./examples/obj_detect_files
~~~
If you want to try the receipt description and question answering functions:
~~~
./examples/ocr_detect_files
~~~
