ビジョンとOracle Analytics Cloud (OAC)の統合

schema_version: 20180708
name: object-detection
version: 0.0.1
runtime: python
build_image: fnproject/python:3.8-dev
run_image: fnproject/python:3.8
entrypoint: /python/bin/fdk /function/func.py handler
memory: 256
timeout: 300

fdk>=0.1.40
oci
https://objectstorage.us-ashburn-1.oraclecloud.com/n/axhheqi2ofpb/b/vision-oac/o/vision_service_python_client-0.3.9-py2.py3-none-any.whl
pandas
requests

import io
import json
import logging
import pandas
import requests
import base64
from io import StringIO
from fdk import response
 
import oci
from vision_service_python_client.ai_service_vision_client import AIServiceVisionClient
from vision_service_python_client.models.analyze_image_details import AnalyzeImageDetails
from vision_service_python_client.models.image_object_detection_feature import ImageObjectDetectionFeature
from vision_service_python_client.models.inline_image_details import InlineImageDetails
 
def handler(ctx, data: io.BytesIO=None):
    signer = oci.auth.signers.get_resource_principals_signer()
    resp = do(signer,data)
    return response.Response(
        ctx, response_data=resp,
        headers={"Content-Type": "application/json"}
    )
 
def vision(dip, txt):
    encoded_string = base64.b64encode(requests.get(txt).content)
 
    image_object_detection_feature = ImageObjectDetectionFeature()
    image_object_detection_feature.max_results = 5
    features = [image_object_detection_feature]
    analyze_image_details = AnalyzeImageDetails()
    inline_image_details = InlineImageDetails()
    inline_image_details.data = encoded_string.decode('utf-8')
    analyze_image_details.image = inline_image_details
    analyze_image_details.features = features
    try:
        le = dip.analyze_image(analyze_image_details=analyze_image_details)
    except Exception as e:
        print(e)
        return ""
    if le.data.image_objects is not None:
      return json.loads(le.data.image_objects.__repr__())
    return ""
 
 
def do(signer, data):
    dip = AIServiceVisionClient(config={}, signer=signer)
 
    body = json.loads(data.getvalue())
    input_parameters = body.get("parameters")
    col = input_parameters.get("column")
    input_data = base64.b64decode(body.get("data")).decode()
    df = pandas.read_json(StringIO(input_data), lines=True)
    df['enr'] = df.apply(lambda row : vision(dip,row[col]), axis = 1)
    #Explode the array of aspects into row per entity
    dfe = df.explode('enr',True)
    #Add a column for each property we want to return from imageObjects struct
    ret=pandas.concat([dfe,pandas.DataFrame((d for idx, d in dfe['enr'].iteritems()))], axis=1)
 
 
    #Drop array of aspects column
    ret = ret.drop(['enr'],axis=1)
    #Drop the input text column we don't need to return that (there may be other columns there)
    ret = ret.drop([col],axis=1)
    if 'name' not in ret.columns:
        return pandas.DataFrame(columns=['id','name','confidence','x0','y0','x1','y1','x2','y2','x3','y3']).to_json(orient='records')
    for i in range(4):
        ret['x' + str(i)] = ret.apply(lambda row: row['bounding_polygon']['normalized_vertices'][i]['x'], axis=1)
        ret['y' + str(i)] = ret.apply(lambda row: row['bounding_polygon']['normalized_vertices'][i]['y'], axis=1)
    ret = ret.drop(['bounding_polygon'],axis=1)
 
    rstr=ret.to_json(orient='records')
    return rstr