欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页

笑脸和口罩数据集的划分,训练和测试

程序员文章站 2022-07-04 20:06:37
...

笑脸和口罩数据集的划分,训练和测试

准备工作

我使用的环境是Python3.7.8搭配VSCode,Python现在已经出了3.8版本的,但是由于版本太新,安装dlib需要编译,不能直接安,使用退而求其次选择了3.7.8版本。

首先需要安装dlib,Keras,TensorFlow,OpenCV等库。

打开Windows PowerShell,输入以下命令来安装必要的库。

python -m pip install -i https://mirrors.aliyun.com/pypi/simple/ --user numpy 
python -m pip install -i https://mirrors.aliyun.com/pypi/simple/ --user scipy
python -m pip install -i https://mirrors.aliyun.com/pypi/simple/ --user keras
python -m pip install -i https://mirrors.aliyun.com/pypi/simple/ --user opencv-python

dlib我使用的是直接下载whl文件放在硬盘中安装的方法。
文件链接:dlib-19.17.99-cp37-cp37m-win_amd64.whl
提取码:87qi
注:这个文件仅适用于Python3.7,其他版本会报错
然后在Windows PowerShell中使用如下命令安装

pip install D:\BaiduNetdiskDownload\dlib-19.17.99-cp37-cp37m-win_amd64.whl

后面的路径随你电脑中的保存路径更改。

TensorFlow的安装分两种,CPU版本和GPU版本,如果你的电脑中使用的不是英伟达的显卡,或者比较怕麻烦,那么就安装CPU版本,在Windows PowerShell中输入如下命令就可以完成安装

python -m pip install -i https://mirrors.aliyun.com/pypi/simple/ --user tensorflow

安装GPU版的TensorFlow相对来说要麻烦不少,不过程序运行速度会快非常多,同样的数据集训练,用CPU算需要900多秒,期间CPU跑满100%,而用GPU算的话只需要90秒不到。

不过安装GPU版的TensorFlow需要先安装配置下面几个

CUDA CUDA Toolkit 10.1 update2 Archive
cuDNN NVIDIA cuDNN

据我了解,目前TensorFlow好像只支持到CUDA10.1,所以好这里放的是CUDA10.1的下载链接。下载后直接安装即可。而下载cuDNN需要注册英伟达开发者,不过还是比较简单的,用英伟达账号登录后填一些信息就可以了。

按照TensorFlow官网的教程,在下载了这两个工具后还要进行一定的配置,CUDA下载后直接安装,并将下面三个路径添加到系统变量的Path中,全程默认安装的话路径是不会有变化的。

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\extras\CUPTI\lib64
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\include

cuDNN下载下来是一个压缩包,将其解压到C:\tools文件夹下,然后将这个路径也添加进系统变量的Path中

C:\tools\cuda\bin

完成这些准备工作后,在Windows PowerShell中输入如下命令就可以安装TensorFlow的GPU版了

python -m pip install -i https://mirrors.aliyun.com/pypi/simple/ --user tensorflow-gpu

笑脸数据集

首先下载笑脸数据集:
smile-detection-master.zip 提取码:2a4o
genki4k.tar 提取码:8vyt
两个数据集有相当多的重复,不过下面那个数据量要更大一些,缺点是没有将笑与不笑划分开单独的文件夹,需要自己找到两个表情的划分。所以我以第一个链接为主。
下载好之后解压保存:
笑脸和口罩数据集的划分,训练和测试
划分数据集

import keras
import os, shutil
# The path to the directory where the original
# dataset was uncompressed
original_dataset_dir = 'D:\\python\\smask\\smile-detection-master\\datasets\\train_folder'
# The directory where we will
# store our smaller dataset
base_dir = 'D:\\python\\smask\\smile'
os.mkdir(base_dir)
# Directories for our training,
# validation and test splits
train_dir = os.path.join(base_dir, 'train')
os.mkdir(train_dir)
validation_dir = os.path.join(base_dir, 'validation')
os.mkdir(validation_dir)
test_dir = os.path.join(base_dir, 'test')
os.mkdir(test_dir)
# Directory with our training smile pictures
train_smiles_dir = os.path.join(train_dir, 'smiles')
os.mkdir(train_smiles_dir)
# Directory with our training unsmile pictures
train_unsmiles_dir = os.path.join(train_dir, 'unsmiles')
os.mkdir(train_unsmiles_dir)
# Directory with our validation smile pictures
validation_smiles_dir = os.path.join(validation_dir, 'smiles')
os.mkdir(validation_smiles_dir)
# Directory with our validation unsmile pictures
validation_unsmiles_dir = os.path.join(validation_dir, 'unsmiles')
os.mkdir(validation_unsmiles_dir)
# Directory with our validation smile pictures
test_smiles_dir = os.path.join(test_dir, 'smiles')
os.mkdir(test_smiles_dir)
# Directory with our validation unsmile pictures
test_unsmiles_dir = os.path.join(test_dir, 'unsmiles')
os.mkdir(test_unsmiles_dir)

然后把datasets文件夹里面的图片放入相应的文件夹
笑脸和口罩数据集的划分,训练和测试
构建卷积神经网络并训练

import os, shutil
from keras import layers
from keras import models
from keras import optimizers
from keras.preprocessing.image import ImageDataGenerator
base_dir = 'D:\\python\\smask\\smile'
train_dir = 'D:\\python\\smask\\smile\\train'
validation_dir = 'D:\\python\\smask\\smile\\validation'
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))
model.summary()
model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
              metrics=['acc'])
# All images will be rescaled by 1./255
train_datagen = ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
        # This is the target directory
        train_dir,
        # All images will be resized to 150x150
        target_size=(150, 150),
        batch_size=20,
        # Since we use binary_crossentropy loss, we need binary labels
        class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')
for data_batch, labels_batch in train_generator:
    print('data batch shape:', data_batch.shape)
    print('labels batch shape:', labels_batch.shape)
    break

history = model.fit_generator(
      train_generator,
      steps_per_epoch=100,
      epochs=30,
      validation_data=validation_generator,
      validation_steps=50
      )
model.save('D:\\python\\smask\\smile\\smiles_and_unsmiles_small_1.h5')

笑脸和口罩数据集的划分,训练和测试
笑脸和口罩数据集的划分,训练和测试
笑脸和口罩数据集的划分,训练和测试
由于这样子训练出来的模型过拟合比较严重,所以还需要进一步处理:数据增强并增加一个Dropout层

datagen = ImageDataGenerator(
      rotation_range=40,
      width_shift_range=0.2,
      height_shift_range=0.2,
      shear_range=0.2,
      zoom_range=0.2,
      horizontal_flip=True,
      fill_mode='nearest')

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
              metrics=['acc'])

train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,)
# Note that the validation data should not be augmented!
test_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
        # This is the target directory
        train_dir,
        # All images will be resized to 150x150
        target_size=(150, 150),
        batch_size=32,
        # Since we use binary_crossentropy loss, we need binary labels
        class_mode='binary')
validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=32,
        class_mode='binary')
history = model.fit_generator(
      train_generator,
      steps_per_epoch=100,
      epochs=30,
      validation_data=validation_generator,
      validation_steps=50)

model.save('D:\\python\\smask\\smile\\smiles_and_unsmiles_small_2.h5')

摄像头判别笑脸

import cv2
from keras.preprocessing import image
from keras.models import load_model
import numpy as np
import dlib
from PIL import Image
model = load_model('D:\\python\\smask\\smile\\smiles_and_unsmiles_small_2.h5')
detector = dlib.get_frontal_face_detector()
video=cv2.VideoCapture(0)
font = cv2.FONT_HERSHEY_SIMPLEX
def rec(img):
    gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
    dets=detector(gray,1)
    if dets is not None:
        for face in dets:
            left=face.left()
            top=face.top()
            right=face.right()
            bottom=face.bottom()
            cv2.rectangle(img,(left,top),(right,bottom),(0,255,0),2)
            img1=cv2.resize(img[top:bottom,left:right],dsize=(150,150))
            img1=cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)
            img1 = np.array(img1)/255.
            img_tensor = img1.reshape(-1,150,150,3)
            prediction =model.predict(img_tensor)    
            if prediction[0][0]>0.5:
                result='unsmile'
            else:
                result='smile'
            cv2.putText(img, result, (left,top), font, 2, (0, 255, 0), 2, cv2.LINE_AA)
        cv2.imshow('Video', img)
while video.isOpened():
    res, img_rd = video.read()
    if not res:
        break
    rec(img_rd)
    if cv2.waitKey(5) & 0xFF == ord('q'):
        break
video.release()
cv2.destroyAllWindows()

这里就请我们敬爱的局座出面客串一下演示效果

笑脸和口罩数据集的划分,训练和测试
笑脸和口罩数据集的划分,训练和测试

口罩数据集

首先下载口罩数据集
这个数据集虽然非常全面,但也有个缺点,就是里面分了多个文件夹,不好进行归档,所以在这里我也把我自己归档重命名过的一个数据集分享出来,图片均为上面链接中的RWMFD_part_1中的图片
mask.zip提取码:kbzz

然后就是划分,操作和上面是一模一样的

import keras
import os, shutil
# The path to the directory where the original
# dataset was uncompressed
original_dataset_dir = 'D:\\python\\smask\\mask\\train'
# The directory where we will
# store our smaller dataset
base_dir = 'D:\\python\\smask\\mask1'
os.mkdir(base_dir)
# Directories for our training,
# validation and test splits
train_dir = os.path.join(base_dir, 'train')
os.mkdir(train_dir)
validation_dir = os.path.join(base_dir, 'validation')
os.mkdir(validation_dir)
test_dir = os.path.join(base_dir, 'test')
os.mkdir(test_dir)
# Directory with our training masks pictures
train_masks_dir = os.path.join(train_dir, 'masks')
os.mkdir(train_masks_dir)
# Directory with our training unmasks pictures
train_unmasks_dir = os.path.join(train_dir, 'unmasks')
os.mkdir(train_unmasks_dir)
# Directory with our validation masks pictures
validation_masks_dir = os.path.join(validation_dir, 'masks')
os.mkdir(validation_masks_dir)
# Directory with our validation unmasks pictures
validation_unmasks_dir = os.path.join(validation_dir, 'unmasks')
os.mkdir(validation_unmasks_dir)
# Directory with our validation masks pictures
test_masks_dir = os.path.join(test_dir, 'masks')
os.mkdir(test_masks_dir)
# Directory with our validation unmasks pictures
test_unmasks_dir = os.path.join(test_dir, 'unmasks')
os.mkdir(test_unmasks_dir)

构建卷积神经网络并训练,数据增强并增加一个Dropout层后再进行一次训练

from keras import layers
from keras import models
from keras import optimizers
from keras.preprocessing.image import ImageDataGenerator

base_dir = 'D:\\python\\smask\\mask1'
train_dir = 'D:\\python\\smask\\mask1\\train'
validation_dir = 'D:\\python\\smask\\mask1\\validation'
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))
model.summary()
model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
              metrics=['acc'])
# All images will be rescaled by 1./255
train_datagen = ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
        # This is the target directory
        train_dir,
        # All images will be resized to 150x150
        target_size=(150, 150),
        batch_size=20,
        # Since we use binary_crossentropy loss, we need binary labels
        class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')
for data_batch, labels_batch in train_generator:
    print('data batch shape:', data_batch.shape)
    print('labels batch shape:', labels_batch.shape)
    break
history = model.fit_generator(
      train_generator,
      steps_per_epoch=100,
      epochs=30,
      validation_data=validation_generator,
      validation_steps=50)
model.save('D:/python/smask/mask1/masks_and_unmasks_small_1.h5')

datagen = ImageDataGenerator(
      rotation_range=40,
      width_shift_range=0.2,
      height_shift_range=0.2,
      shear_range=0.2,
      zoom_range=0.2,
      horizontal_flip=True,
      fill_mode='nearest')
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
              metrics=['acc'])
train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,)
# Note that the validation data should not be augmented!
test_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
        # This is the target directory
        train_dir,
        # All images will be resized to 150x150
        target_size=(150, 150),
        batch_size=32,
        # Since we use binary_crossentropy loss, we need binary labels
        class_mode='binary')
validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=32,
        class_mode='binary')
history = model.fit_generator(
      train_generator,
      steps_per_epoch=100,
      epochs=50,
      validation_data=validation_generator,
      validation_steps=50)
model.save('D:/python/smask/mask1/masks_and_unmasks_small_2.h5')

笑脸和口罩数据集的划分,训练和测试
笑脸和口罩数据集的划分,训练和测试
笑脸和口罩数据集的划分,训练和测试

摄像头判别是否佩戴口罩

import cv2
from keras.preprocessing import image
from keras.models import load_model
import numpy as np
import dlib
from PIL import Image
model = load_model('D:/python/smask/mask1/masks_and_unmasks_small_2.h5')
detector = dlib.get_frontal_face_detector()
video=cv2.VideoCapture(0)
font = cv2.FONT_HERSHEY_SIMPLEX
def rec(img):
    gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
    dets=detector(gray,1)
    if dets is not None:
        for face in dets:
            left=face.left()
            top=face.top()
            right=face.right()
            bottom=face.bottom()
            cv2.rectangle(img,(left,top),(right,bottom),(0,255,0),2)
            img1=cv2.resize(img[top:bottom,left:right],dsize=(150,150))
            img1=cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)
            img1 = np.array(img1)/255.
            img_tensor = img1.reshape(-1,150,150,3)
            prediction =model.predict(img_tensor)    
            if prediction[0][0]>0.5:
                result='unmask'
            else:
                result='mask'
            cv2.putText(img, result, (left,top), font, 2, (0, 255, 0), 2, cv2.LINE_AA)
        cv2.imshow('Video', img)
while video.isOpened():
    res, img_rd = video.read()
    if not res:
        break
    rec(img_rd)
    if cv2.waitKey(5) & 0xFF == ord('q'):
        break
video.release()
cv2.destroyAllWindows()

运行效果
笑脸和口罩数据集的划分,训练和测试
笑脸和口罩数据集的划分,训练和测试

相关标签: tensorflow cuda

上一篇: POJ 3176 Cow Bowling

下一篇: MySql引擎