serverless 实战 -通过serverless framework部署 Ams Lambda 人脸识别

今天要整的是免费使用ams 提供的serverless 实现 lambda ，通过lambda 实现人脸识别（依赖face_recognition实现），我们来看看，以下四个人，他们的相识度会怎么样呢？

 

Jay

 

Jay1

 

GZ

CXK

 

两种实现方式：

1. 本地运行测试没问题，通过serverless 部署到 ams lambda ,通过http 触发调用（ 这篇讲实现这个简单的FaaS)。

2. 本地打好依赖层包，lambda 控制台把依赖加上，然后控制台边写代码，边调整验证。（下期）

直接来，预选准备:

1.  aws账号，注册免费用。（https://aws.amazon.com/cn/）

2.  安装serverless 框架。（https://cloud.tencent.com/document/product/1154/42990）

3. 本地python 环境，3.8 

 

第一种实现： 本地就绪，直接上传。

1. 本地安装python  库face_recognition

#我本地已安装，所以会提示已安装，如果你安装有问题，可以留言我
>pip3.8 install cmake dlib face_recognition
Requirement already satisfied: cmake in /usr/local/lib/python3.8/site-packages (3.18.2.post1)
Requirement already satisfied: dlib in /usr/local/lib/python3.8/site-packages (19.21.0)
Requirement already satisfied: face_recognition in /usr/local/lib/python3.8/site-packages (1.3.0)
Requirement already satisfied: face-recognition-models>=0.3.0 in /usr/local/lib/python3.8/site-packages (from face_recognition) (0.3.0)
Requirement already satisfied: Click>=6.0 in /usr/local/lib/python3.8/site-packages (from face_recognition) (7.1.2)
Requirement already satisfied: numpy in /usr/local/lib/python3.8/site-packages (from face_recognition) (1.19.2)
Requirement already satisfied: Pillow in /usr/local/lib/python3.8/site-packages (from face_recognition) (7.2.0)

2. 利益face_recognition API，比较程序如下：

# coding=utf-8
import face_recognition


def match_handler(event, context):
    # 通过face_distance 函数来判断两张脸是否相识，相识度是多少                          
    # 此模型训练在0.6 以内的距离值，就认为是同一个人，如果想要更严格，可以把距离值设置的更新，比如设置值0.55
    jay = face_recognition.load_image_file("jay.png")
    jay1 = face_recognition.load_image_file("jay1.png")
    GZ = face_recognition.load_image_file("GZ.png")
    CXK = face_recognition.load_image_file("CXK.png")

# Get the face encodings for the known images
    jay_face_encoding = face_recognition.face_encodings(jay)[0]
    jay1_face_encoding = face_recognition.face_encodings(jay1)[0]
    GZ_face_encoding = face_recognition.face_encodings(GZ)[0]
    CXK_face_encoding = face_recognition.face_encodings(CXK)[0]

    known_encodings = [
       jay_face_encoding,
       jay1_face_encoding,
       GZ_face_encoding,
       CXK_face_encoding
    ]
    dictTag = {"0": "jay", "1": "jay1", "2": "gz", "3": "cxk"}
    result = "=====================CXK跟JAY，GZ比========================\r\n"
    print("=====================CXK跟JAY，GZ比========================")
    image_to_test = face_recognition.load_image_file("CXK.png")
    image_to_test_encoding = face_recognition.face_encodings(image_to_test)[0]

   # See how far apart the test image is from the known faces
    face_distances = face_recognition.face_distance(known_encodings, image_to_test_encoding)
    for i, face_distance in enumerate(face_distances):
        print("CXK 与{},相识度差距 {:.2}  #{}".format(dictTag[str(i)], face_distance, i))
        print("-差距值为普通的 0.6, 他们为同一人的对比结果是： {}".format(face_distance < 0.6))
        print("- 更严格的差距设置为0.5 ,  他们为同一人的对比结果是： {}".format(face_distance < 0.5))
        result+= "CXK 与{},相识度差距 {:.2}  #{}".format(dictTag[str(i)], face_distance, i) + "\r\n"
        result+="-差距值为普通的 0.6, 他们为同一人的对比结果是： {}".format(face_distance < 0.6)+"\r\n"
        result+="- 更严格的差距设置为0.5 ,  他们为同一人的对比结果是： {}".format(face_distance < 0.5)+"\r\n"


    print("=====================GZ跟JAY 相似度========================")
    result+="=====================GZ跟JAY 相似度========================\r\n"
    GZ_to_test = face_recognition.load_image_file("GZ.png")
    GZ_to_test_encoding = face_recognition.face_encodings(GZ_to_test)[0]

   # See how far apart the test image is from the known faces
    face_distances = face_recognition.face_distance(known_encodings, GZ_to_test_encoding)

    for i, face_distance in enumerate(face_distances):
        print("GZ 与{},相识度差距 {:.2}  #{}".format(dictTag[str(i)], face_distance, i))
        print("-差距值为普通的 0.6, 他们为同一人的对比结果是： {}".format(face_distance < 0.6))
        print("- 更严格的差距设置为0.5 ,  他们为同一人的对比结果是： {}".format(face_distance < 0.5))
        result +="GZ 与{},相识度差距 {:.2}  #{}".format(dictTag[str(i)], face_distance, i)+"\r\n"
        result +="-差距值为普通的 0.6, 他们为同一人的对比结果是： {}".format(face_distance < 0.6)+"\r\n"
        result += "- 更严格的差距设置为0.5 ,  他们为同一人的对比结果是： {}".format(face_distance < 0.5)+"\r\n"

    response = {
        "statusCode": 200,
        "body": result
    }

    return response


if __name__ == "__main__":
    print(match_handler('', ''))

3. 本地跑一跑

#python3.8 为你本地python3 的版本，可能你的是3.x
>python3.8 handler.py 

运行结果：

4. 上面的步骤都比较简单，怎么把他部署到amazon上去呢？前面讲到需要借助serverless框架。

#安装python serverless依赖
>npm install --save serverless-python-requirements
npm WARN face-demo No description
npm WARN face-demo No repository field.
npm WARN face-demo No license field.

+ serverless-python-requirements@5.1.0
updated 1 package in 5.656s

   在项目中增加依赖文件requirements.txt,内容为你跑的python依赖库。

face_recognition==1.3.0

   serverless.yml ，中声明你的运行环境，函数，API

service: face-demo

frameworkVersion: '2'

provider:
  name: aws
  runtime: python3.8
  timeout: 1000
  memorySize: 1024

functions:
  match_handler:
    handler: handler.match_handler
    events:
      - http:
          path: face_demo
          method: get

plugins:
  - serverless-python-requirements

custom:
  pythonRequirements:
    dockerizePip: true

   provider为你的服务厂商，运行时环境时python3.8 ,运行超时时间为1000，内存限制为1024M。

   函数名为match_handler, handler 为handler.match_handler。

   event为定义的事件，http trigger，通过get 方法的API ，可以触发函数的调用。

  部署到云：serverless deploy 命令

5. 部署成功，试试验证下

Serverless: Stack update finished...
Service Information
service: face-demo
stage: dev
region: us-east-1
stack: face-demo-dev
resources: 11
api keys:
  None
endpoints:
  GET - https://s9gj3grvig.execute-api.us-east-1.amazonaws.com/dev/face_demo
functions:
  match_handler: face-demo-dev-match_handler
layers:
  None
Serverless: Removing old service artifacts from S3...

**************************************************************************************************************************************
Serverless: Announcing Metrics, CI/CD, Secrets and more built into Serverless Framework. Run "serverless login" to activate for free..
**************************************************************************************************************************************

https://s9gj3grvig.execute-api.us-east-1.amazonaws.com/dev/face_demo 这个地址为刚才serverless.yml 定义的http trigger

访问下：

服务器日志：（GZ跟杰伦小于比较严格的0.5，系统认为是同一人，CXK当然跟杰伦，广智都不像，脱口秀大会说他像杰伦，站在程序员角度，那还真是）

 

看看ams的控制台：