深度学习模型下接口编写及Docker下生产环境部署（万能篇）
深度学习模型下接口编写及Docker下生产环境部署（万能篇）
- 一、实现目标
- 二、操作背景
- 三、编写接口
-
- 1.detect_image.py源代码
- 2.detect_image.py增加接口代码
- 四、服务端代码
- 五、客户端代码
- 六、docker生产环境部署
-
- 1、docker常用命令
- 2、部署步骤
- 3、相关命令
一、实现目标
通过阅读本篇文章，大家将会了解以下内容：
- 如何编写深度学习模型的接口。
- 如何撰写服务端和客户端。
- 如何利用docker进行生产环境部署。
- 在部署中了解docker相关知识。
二、操作背景
- 最最基础的就是大家要有一个跑的通的代码。本文以口罩检测为例。
- 已安装好docker.
三、编写接口
本项目实现的是口罩检测，运行主文件为detect_image.py，该文件功能为检测一张图片是否戴口罩，框出位置，并显示标签和置信度得分。
1.detect_image.py源代码
# USAGE
# python detect_mask_image.py --image images/pic1.jpeg
# import the necessary packages
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input
from tensorflow.keras.preprocessing.image import img_to_array
from tensorflow.keras.models import load_model
import numpy as np
import argparse
import cv2
import os
def mask_image():
# construct the argument parser and parse the arguments
ap = argparse.ArgumentParser()
ap.add_argument("-i", "--image", required=True,
help="path to input image")
ap.add_argument("-f", "--face", type=str,
default="face_detector",
help="path to face detector model directory")
ap.add_argument("-m", "--model", type=str,
default="mask_detector.model",
help="path to trained face mask detector model")
ap.add_argument("-c", "--confidence", type=float, default=0.5,
help="minimum probability to filter weak detections")
args = vars(ap.parse_args())
# load our serialized face detector model from disk
print("[INFO] loading face detector model...")
prototxtPath = os.path.sep.join([args["face"], "deploy.prototxt"])
weightsPath = os.path.sep.join([args["face"],
"res10_300x300_ssd_iter_140000.caffemodel"])
net = cv2.dnn.readNet(prototxtPath, weightsPath)
# load the face mask detector model from disk
print("[INFO] loading face mask detector model...")
model = load_model(args["model"])
# load the input image from disk, clone it, and grab the image spatial
# dimensions
image = cv2.imread(args["image"])
orig = image.copy()
(h, w) = image.shape[:2]
# construct a blob from the image
blob = cv2.dnn.blobFromImage(image, 1.0, (300, 300),
(104.0, 177.0, 123.0))
# pass the blob through the network and obtain the face detections
print("[INFO] computing face detections...")
net.setInput(blob)
detections = net.forward()
# loop over the detections
for i in range(0, detections.shape[2]):
# extract the confidence (i.e., probability) associated with
# the detection
confidence = detections[0, 0, i, 2]
# filter out weak detections by ensuring the confidence is
# greater than the minimum confidence
if confidence > args["confidence"]:
# compute the (x, y)-coordinates of the bounding box for
# the object
box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
(startX, startY, endX, endY) = box.astype("int")
# ensure the bounding boxes fall within the dimensions of
# the frame
(startX, startY) = (max(0, startX), max(0, startY))
(endX, endY) = (min(w - 1, endX), min(h - 1, endY))
# extract the face ROI, convert it from BGR to RGB channel
# ordering, resize it to 224x224, and preprocess it
face = image[startY:endY, startX:endX]
face = cv2.cvtColor(face, cv2.COLOR_BGR2RGB)
face = cv2.resize(face, (224, 224))
face = img_to_array(face)
face = preprocess_input(face)
face = np.expand_dims(face, axis=0)
# pass the face through the model to determine if the face
# has a mask or not
(mask, withoutMask) = model.predict(face)[0]
# determine the class label and color we'll use to draw
# the bounding box and text
label = "Mask" if mask > withoutMask else "No Mask"
color = (0, 255, 0) if label == "Mask" else (0, 0, 255)
# include the probability in the label
label = "{}: {:.2f}%".format(label, max(mask, withoutMask) * 100)
# display the label and bounding box rectangle on the output
# frame
cv2.putText(image, label, (startX, startY - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.45, color, 2)
cv2.rectangle(image, (startX, startY), (endX, endY), color, 2)
# show the output image
cv2.imshow("Output", image)
cv2.waitKey(0)
if __name__ == "__main__":
mask_image()
2.detect_image.py增加接口代码
功能：通过接口返回3个参数，检测结果（口罩/非口罩）、口罩坐标、置信度得分。
分析该代码可知，主要用到以下代码中的变量为：
（1）位置坐标：startX、startY, endX, endY
# frame
cv2.putText(image, label, (startX, startY - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.45, color, 2)
cv2.rectangle(image, (startX, startY), (endX, endY), color, 2)
（2） 标签：label
（3） 置信度：max(mask, withoutMask)
label = "Mask" if mask > withoutMask else "No Mask"
color = (0, 255, 0) if label == "Mask" else (0, 0, 255)
下面的主要工作就是如何变量提取出来进行返回。
1、获取标签和置信度。
mask_label = label
mask_confidence = max(mask, withoutMask)
2、将坐标、标签、置信度增加至列表
result = []
result.append([startX.item(), startY.item(), endX.item(), endY.item()])
result.append(mask_confidence)
result.append(mask_label)
3、由于一张图像可能检测到多个口罩，因此需要在最外层定义一个总的变量:
例如 results=[ ],并将第二步的列表增加至本列表
results.append(result)
增加接口后的完整代码如下：
# USAGE
# python detect_mask_image.py --image images/pic1.jpeg
# import the necessary packages
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input
from tensorflow.keras.preprocessing.image import img_to_array
from tensorflow.keras.models import load_model
import numpy as np
import argparse
import cv2
import os
def mask_image(image):
# construct the argument parser and parse the arguments
ap = argparse.ArgumentParser()
#ap.add_argument("-i", "--image", required=False,
#help="path to input image")
ap.add_argument("-f", "--face", type=str,default="face_detector",help="path to face detector model directory")
ap.add_argument("-m", "--model", type=str,default="mask_detector.model",help="path to trained face mask detector model")
ap.add_argument("-c", "--confidence", type=float, default=0.5,help="minimum probability to filter weak detections")
args = vars(ap.parse_args())
# load our serialized face detector model from disk
print("[INFO] loading face detector model...")
prototxtPath = os.path.sep.join([args["face"], "deploy.prototxt"])
weightsPath = os.path.sep.join([args["face"],
"res10_300x300_ssd_iter_140000.caffemodel"])
net = cv2.dnn.readNet(prototxtPath, weightsPath)
# load the face mask detector model from disk
print("[INFO] loading face mask detector model...")
model = load_model(args["model"])
# load the input image from disk, clone it, and grab the image spatial
# dimensions
(h, w) = image.shape[:2]
# construct a blob from the image
blob = cv2.dnn.blobFromImage(image, 1.0, (300, 300),
(104.0, 177.0, 123.0))
# pass the blob through the network and obtain the face detections
print("[INFO] computing face detections...")
net.setInput(blob)
detections = net.forward()
results = []
# loop over the detections
for i in range(0, detections.shape[2]):
# extract the confidence (i.e., probability) associated with
# the detection
confidence = detections[0, 0, i, 2]
# filter out weak detections by ensuring the confidence is
# greater than the minimum confidence
if confidence > args["confidence"]:
# compute the (x, y)-coordinates of the bounding box for
# the object
box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
(startX, startY, endX, endY) = box.astype("int")
# ensure the bounding boxes fall within the dimensions of
# the frame
(startX, startY) = (max(0, startX), max(0, startY))
(endX, endY) = (min(w - 1, endX), min(h - 1, endY))
# extract the face ROI, convert it from BGR to RGB channel
# ordering, resize it to 224x224, and preprocess it
face = image[startY:endY, startX:endX]
face = cv2.cvtColor(face, cv2.COLOR_BGR2RGB)
face = cv2.resize(face, (224, 224))
face = img_to_array(face)
face = preprocess_input(face)
face = np.expand_dims(face, axis=0)
# pass the face through the model to determine if the face
# has a mask or not
(mask, withoutMask) = model.predict(face)[0]
# determine the class label and color we'll use to draw
# the bounding box and text
label = "Mask" if mask > withoutMask else "No Mask"
color = (0, 255, 0) if label == "Mask" else (0, 0, 255)
# save label and confidence for output
mask_label = label
mask_confidence = max(mask, withoutMask)
# include the probability in the label
label = "{}: {:.2f}%".format(label, max(mask, withoutMask) * 100)
# display the label and bounding box rectangle on the output
# frame
cv2.putText(image, label, (startX, startY - 10),
cv2.FONT_HERSHEY_SIMPLEX, 0.45, color, 2)
cv2.rectangle(image, (startX, startY), (endX, endY), color, 2)
result = []
result.append([startX.item(), startY.item(), endX.item(), endY.item()])
result.append(mask_confidence)
result.append(mask_label)
results.append(result)
return results
if __name__ == "__main__":
path = 'images/pic1.jpeg'
im0s = cv2.imread(path)
res = mask_image(im0s)
print(res)
四、服务端代码
功能：接受客户端请求，返回检测结果相关参数。本代码可通用，针对不同的需求，只需修改2处：
1、导入并调用函数，如本文的detect_image.py文件中的mask_image()。
from detect_mask_image import *
其他功能代码
result = mask_image(image)
2、增加结果解析部分即可
for i, x in enumerate(result):
dictinter = {'res': str(x[2]), 'confidence': str(x[1]),
'vertexes_location': (x[0])}
data_dict['item' + str(i + 1)] = dictinter
output_dict['data'] = data_dict
3、服务端完整代码
（没有安装的模块自行安装哟，如 flasK, flask_cors）
server.py
import base64
import io
import os
import urllib.request
import logging
import time
from PIL import Image
from flask import Flask, request
from flask_cors import CORS
from detect_mask_image import *
os.environ['MXNET_CUDNN_AUTOTUNE_DEFAULT'] = '0'
app = Flask(__name__)
CORS(app, suppors_credentials=True, resources={r'/*'})
# Initialize and load the face detector.
gpuid = 0
app.logger.info('INITIALIZING MASK SERVER... USING {}'.format('GPU' if gpuid>=0 else 'CPU'))
# Logger Initialization.
#logging.basicConfig(filename="./Logs/access.log", level=logging.DEBUG, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
app.logger.info('------------------------START UP------------------------')
@app.route('/facemask', methods = ['GET', 'POST'])
def server():
# Get client's ip address.
ip = request.remote_addr
output_dict = dict()
data_dict = dict()
start_time = time.time()
req = eval(str(request.data.decode('utf-8')))
# See whether variable 'req' is of type dict.
if type(req) is not dict:
app.logger.warning('From user ip: {}, DATA FORMAT FAILURE.'.format(ip))
output_dict['code'] = '-1'
output_dict['message'] = 'ERROR: DATA FORMAT FAILURE.'
return output_dict
# Tab up the potentially missing keys.
if 'url' not in req.keys():
req["url"] = ''
if 'img' not in req.keys():
req["img"] = ''
# Get the corresponding variables.
req_img = req["img"]
url = req["url"]
if len(req_img) == 0:
if len(url) == 0:
app.logger.warning('From user ip: {}, NO INPUT DATA DETECTED.'.format(ip))
output_dict['code'] = '-2'
output_dict['message'] = 'ERROR: NO INPUT DATA DETECTED'
return output_dict
# Return the error code when there's no response from the url taken in.
try:
response = urllib.request.urlopen(url)
except:
app.logger.warning('From user ip: {}, INVALID URL.'.format(ip))
output_dict['code'] = '-3'
output_dict['message'] = 'ERROR: INVALID URL.'
return output_dict
img_array = np.array(bytearray(response.read()), dtype=np.uint8)
image = cv2.imdecode(img_array, -1)
if image is None:
app.logger.warning('From user ip: {}, None TYPE IMAGE FROM URL.'.format(ip))
output_dict['code'] = '-4'
output_dict['message'] = 'ERROR: None TYPE IMAGE FROM URL.'
return output_dict
else:
# Return the error code whenever the base64 code is not decoded successfully
try:
image_bytes = base64.b64decode(req_img)
except:
app.logger.warning('From user ip: {}, IMAGE DECODING FAILURE.'.format(ip))
output_dict['code'] = '-5'
output_dict['message'] = 'ERROR: IMAGE DECODING FAILURE.'
return output_dict
image_array = np.frombuffer(image_bytes, dtype=np.uint8)
image = cv2.imdecode(image_array, 1)
print("image convert spend time: ", time.time() - start_time)
result = mask_image(image)
if len(result) == 0:
app.logger.warning('From user ip: {}, MODLE RECOGNIZING 0 FACE MASK.'.format(ip))
output_dict['code'] = '-6'
output_dict['message'] = 'ERROR: MODLE RECOGNIZING 0 FACE MASK.'
return output_dict
for i, x in enumerate(result):
dictinter = {'res': str(x[2]), 'confidence': str(x[1]),
'vertexes_location': (x[0])}
data_dict['item' + str(i + 1)] = dictinter
output_dict['data'] = data_dict
output_dict['time'] = time.time() - start_time
output_dict['code'] = '1'
output_dict['message'] = 'SUCCESS: {} FACE MASK FOUND.'.format(len(result))
# Calculate the duration for a single image
duration = (time.time() - start_time)
app.logger.info('From user ip: {}, duration: {}s, detection successful.'.format(ip, duration))
app.logger.info('RECOGNIZING RESULT IS {}'.format(output_dict))
print('RECOGNIZING RESULT IS {}'.format(output_dict))
return output_dict
if __name__ == '__main__':
# app.run(port=8864)#8865
app.run(host='0.0.0.0', port=8864)#8865
五、客户端代码
client.py
图片路径、ip可根据真实环境修改
import requests
import base64
from requests.adapters import HTTPAdapter
from requests.packages.urllib3.util.retry import Retry
#url = "http://部署后真实IP/facemask"
url = "http://0.0.0.0:8864/facemask"
image = 'images/5.jpg'
image_url = 'http://p7.qhmsg.com/dr/270_500_/t01439835ff6c40e12e.jpg'
local_image = True
s = requests.Session()
retry = Retry(connect =5, backoff_factor =1)
adapter = HTTPAdapter(max_retries = retry)
s.keep_alive =False
if local_image:
with open(image, 'rb') as file:
base64_data = base64.b64encode(file.read())
img = base64_data.decode()
post_info = {'img': img}
else:
post_info = {'url': image_url}
r = s.post(url, json=post_info)
r.encoding = 'gb2312'
print(r.content.decode("unicode_escape"))
六、docker生产环境部署
第一种方案：编写dockfile文件，利用docker命令一箭部署。缺点，容易缺少安装某个包，编译出错后，不容易定位问。
第二种方案：一步一步部署。新手建议按照此方案。
1、docker常用命令
可参考https://www.runoob.com/docker/ubuntu-docker-install.html，docker安装，容器创建，镜像创建等命令。
2、部署步骤
首先我们要拉取一个镜像A1，如果没有可以在docker-hub上面拉取，比如我的模型是pytorch框架下的，就可以选择nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04，然后利用这个镜像A1实例化一个容器B1，然后再在这个容器B1中进环境配置，导入项目，测试服务等一列操作。最后再将该容器B1导出，然后再将导出的包C1导入成新的镜像A2，再创建新的容器B2，然后就可以在新的容器B2直接运行服、服务即可。
那麽为什麽我们要创建导出再创建再导出呢？因为第一次我们创建的容器，环境配置是一步一步操作的，如果要把这个项目部署到其他的服务器上，还需要再次配置环境。而第二次创建容器，是利用导出的容器创建的，环境已经配置好了，后续再部署，我们只需要利用这个导出包创建新的镜像就好了，就无需再配置环境，代码可一键运行。
好啦，下面我们就开始一步一步部署吧。注：本项目为centos系统
- docker安装
本地主机上运行：
curl -fsSL https://get.docker.com | bash -s docker --mirror Aliyun
- docker版本检查
本地主机上运行：
docker version
出现相应版本信息，安装即为成功。
- 拉取基础镜像
docker pull nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04
- 查看所有镜像
我们可以使用 docker images 来列出本地主机上的镜像
sudo docker images
各个选项说明:
REPOSITORY：表示镜像的仓库源
TAG：镜像的标签
IMAGE ID：镜像ID
CREATED：镜像创建时间
SIZE：镜像大小
可以看到本镜像的ID为：a4bdb1190443
- 利用镜像创建容器，并做端口映射
sudo docker run -it --runtime=nvidia -p 8860:8860 a4bdb1190443 /bin/bash
- 查看创建的容器
sudo docker ps -a
该图为本人之前创建好的容器。
假设我们的 contariner id = 9345dde9dafb
如果要删除容器，可用命令
docker rm -f 容器ID
- 项目拷贝到容器里
主机运行：
例如： docker cp 本地项目路径 contariner id :容器路径
docker cp /home/zhangjl396/code/Face-Mask-Detection-master 9345dde9dafb:/root
注：容器的最后不能使/root/,不能加“/”
- 进入容器
主机运行：
sudo docker attach 9345dde9dafb
- 容器中配置项目环境
容器中运行：
逐行运行dockfile文件代码时，不能加RUN ,前提是dockerfile已经写好。只需运行下面文件带RUN的命令行即可。安装到requirements.txt时，如果报错，可逐行安装requirements.txt所需安装包。
本人使用的dockerfile文件为：
FROM nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04
COPY . /faceMask
WORKDIR /faceMask
RUN sed -i 's/archive.ubuntu.com/mirrors.aliyun.com/g' /etc/apt/sources.list && \
sed -i 's/security.ubuntu.com/mirrors.aliyun.com/g' /etc/apt/sources.list
RUN mv /etc/apt/sources.list.d/nvidia-ml.list /etc/apt/sources.list.d/nvidia-ml.list.bak && \
mv /etc/apt/sources.list.d/cuda.list /etc/apt/sources.list.d/cuda.list.bak
RUN apt-get clean && apt-get update --fix-missing
RUN apt-get update -y && apt-get upgrade -y
RUN apt-get install -y libsm6 libxext6 libxrender-dev && \
apt-get install -y python3 python3-pip
RUN pip3 install --upgrade pip
# opencv-contrib-python 依赖
RUN apt-get install -y libglib2.0-0 libgl1-mesa-glx
RUN pip3 install --user -i https://mirrors.aliyun.com/pypi/simple -r requirements.txt
CMD python3 server.py
EXPOSE 8860
- 导出容器
如果到此运行server.py的文件测试无误，即可导出容器。
如果要导出本地某个容器，可以使用 docker export 命令。注意导出路径
主机运行：
docker export 9345dde9dafb > facemask.tar
- 导入容器快照
可以使用 docker import 从容器快照文件中再导入为镜像，以下实例将快照文件 facemask.tar 导入到镜像 test/facemask:v1.0:
格式：cat 导出的容器.tar | sudo docker import - 镜像名：版本
主机运行：
cat facemask.tar | sudo docker import - test/facemask:v1.0
- 查看镜像是否导入成功，查看所有镜像
sudo docker images
可以看到新的镜像ID为：45be85df5b5e
- 再次利用镜像创建容器，并做端口映射
sudo docker run -it --runtime=nvidia -p 8860:8860 45be85df5b5e /bin/bash
- 查看创建的容器
sudo docker ps -a
新的容器ID=e200b990da9a
- 进入容器
docker attach e200b990da9a
进入新的容器就相当于进入了一个新的服务器，可以查看目录ls, 也可以cd到相应文件夹中。
- 找到项目运行程序即可，至此大功告成
3、相关命令
- 远程拷贝文件
scp local_file remote_username@remote_ip:remote_folder
scp local_file remote_username@remote_ip:remote_file
scp local_file remote_ip:remote_folder
scp local_file remote_ip:remote_file
指定用户名是需要输入密码，不指定用户名需要同时输入用户名和密码。
- 进入远程主机
ssh username@IP
- cp 命令
docker cp :用于容器与主机之间的数据拷贝。
docker cp [OPTIONS] CONTAINER:SRC_PATH DEST_PATH|-
docker cp [OPTIONS] SRC_PATH|- CONTAINER:DEST_PATH