I tried to make a face diagnosis AI for a female professional golfer ②
1.First of all
By previous time, we have collected images, acquired face parts, and inflated data, which are part of data preprocessing.
This time, I actually created a model and checked the accuracy. ・ Model built by myself ・ Transfer learning (VGG16) I will write about these two.
2. Creating a file divided into training data and test data
preparation.py
from PIL import Image
import os, glob
import numpy as np
from PIL import ImageFile
import cv2
from keras.utils.np_utils import to_categorical
# IOError:image file is truncated to avoid
ImageFile.LOAD_TRUNCATED_IMAGES = True
#Creation of training data
#Create an empty list for each
img_shibuno = []
img_koiwai = []
img_hara = []
img_lists = [img_shibuno, img_koiwai, img_hara]
#Get the path below face
in_dir = './face/*'
in_file = glob.glob(in_dir)
#Each folder (player)Perform processing for each
for num in range(len(in_file)):
#Get the path of each image in the player's file
in_file_name = glob.glob(in_file[num]+'/*')
#Process each image
for i in range(len(in_file_name)):
#Open image
img = Image.open(in_file_name[i])
img = img.convert("RGB")
#Adjust size
img = img.resize((64,64))
#Convert to ndarray
data = np.asarray(img)
# img_Add to lists
img_lists[num].append(data)
#Combine lists with images
X_train = np.array(img_shibuno+img_koiwai+img_hara)
#0 each~Enter a value up to 2
y_train = np.array([0]*len(img_shibuno) + [1]*len(img_koiwai) + [2]*len(img_hara))
#Creating test data
#Create an empty list for each
img_shibuno = []
img_koiwai = []
img_hara = []
img_lists = [img_shibuno, img_koiwai, img_hara]
in_dir = './valid/*'
in_file = glob.glob(in_dir)
for num in range(len(in_file)):
#Get the path of each image in the player's file
in_file_name = glob.glob(in_file[num]+'/*')
#Process each image
for i in range(len(in_file_name)):
#Open image
img = Image.open(in_file_name[i])
img = img.convert("RGB")
#Adjust size
img = img.resize((64,64))
#Convert to ndarray
data = np.asarray(img)
# img_Add to lists
img_lists[num].append(data)
#Combine lists with images
X_test = np.array(img_shibuno+img_koiwai+img_hara)
#0 each~Enter a value up to 2
y_test = np.array([0]*len(img_shibuno) + [1]*len(img_koiwai) + [2]*len(img_hara))
# one-hot-Process vector
y_train = to_categorical(y_train)
y_test = to_categorical(y_test)
#Save training and validation data to a file
xy = (X_train, X_test, y_train, y_test)
np.save('./golfer.npy', xy)
This completes data preprocessing. Next, evaluate the model
3. Model creation and evaluation
Create your own model and transfer learning (vgg16) model respectively
- The following codes are written separately for explanation, but please combine them into one.
3-1. Import module
from keras.models import Model, Sequential
from keras.layers import Conv2D, MaxPooling2D
from keras.layers import Activation, Dropout, Flatten, Dense, Input
from keras.applications.vgg16 import VGG16
from keras.utils import np_utils
import keras
from keras import optimizers, models, layers
import numpy as np
import matplotlib.pyplot as plt
classes = ['shibuno', 'koiwai', 'hara']
num_classes = len(classes)
image_size = 64
3-2. Function to read data
def load_data():
X_train, X_test, y_train, y_test = np.load('./golfer.npy', allow_pickle=True)
#Set each pixel value of the input data to 0-Normalize in the range of 1
X_train = X_train / 255
X_test = X_test / 255
return X_train, y_train, X_test, y_test
3-3. Function to learn the model
__ * Please describe only one of ① and ② below. __
① Model created by yourself
def train(X_train, y_train, X_test, y_test):
model = Sequential()
#X is(296, 64, 64, 3): X.shepe[1:]so(64, 64, 3)
model.add(Conv2D(32, (3, 3), padding='same', input_shape=X_train.shape[1:]))
model.add(Activation('relu'))
model.add(Conv2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.1))
model.add(Conv2D(64, (3, 3), padding='same'))
model.add(Activation('relu'))
model.add(Conv2D(64, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.45))
model.add(Dense(3))
model.add(Activation('softmax'))
model.summary()
#Optimization algorithm RMSprop
opt = keras.optimizers.rmsprop(lr=0.00005, decay=1e-6)
model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy'])
return model
Visualize the created model
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) (None, 64, 64, 3) 0
_________________________________________________________________
block1_conv1 (Conv2D) (None, 64, 64, 64) 1792
_________________________________________________________________
block1_conv2 (Conv2D) (None, 64, 64, 64) 36928
_________________________________________________________________
block1_pool (MaxPooling2D) (None, 32, 32, 64) 0
_________________________________________________________________
block2_conv1 (Conv2D) (None, 32, 32, 128) 73856
_________________________________________________________________
block2_conv2 (Conv2D) (None, 32, 32, 128) 147584
_________________________________________________________________
block2_pool (MaxPooling2D) (None, 16, 16, 128) 0
_________________________________________________________________
block3_conv1 (Conv2D) (None, 16, 16, 256) 295168
_________________________________________________________________
block3_conv2 (Conv2D) (None, 16, 16, 256) 590080
_________________________________________________________________
block3_conv3 (Conv2D) (None, 16, 16, 256) 590080
_________________________________________________________________
block3_pool (MaxPooling2D) (None, 8, 8, 256) 0
_________________________________________________________________
block4_conv1 (Conv2D) (None, 8, 8, 512) 1180160
_________________________________________________________________
block4_conv2 (Conv2D) (None, 8, 8, 512) 2359808
_________________________________________________________________
block4_conv3 (Conv2D) (None, 8, 8, 512) 2359808
_________________________________________________________________
block4_pool (MaxPooling2D) (None, 4, 4, 512) 0
_________________________________________________________________
block5_conv1 (Conv2D) (None, 4, 4, 512) 2359808
_________________________________________________________________
block5_conv2 (Conv2D) (None, 4, 4, 512) 2359808
_________________________________________________________________
block5_conv3 (Conv2D) (None, 4, 4, 512) 2359808
_________________________________________________________________
block5_pool (MaxPooling2D) (None, 2, 2, 512) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 2048) 0
_________________________________________________________________
dense_1 (Dense) (None, 256) 524544
_________________________________________________________________
dropout_1 (Dropout) (None, 256) 0
_________________________________________________________________
dense_2 (Dense) (None, 3) 771
=================================================================
② Transfer learning
def train(X, y, X_test, y_test):
input_tensor = Input(shape=(64, 64, 3))
vgg16 = VGG16(include_top=False, weights='imagenet', input_tensor=input_tensor)
#I am creating a model of the feature extraction part
top_model = vgg16.output
top_model = Flatten(input_shape=vgg16.output_shape[1:])(top_model)
top_model = Dense(256, activation='sigmoid')(top_model)
top_model = Dropout(0.5)(top_model)
top_model = Dense(3, activation='softmax')(top_model)
#vgg16 and top_Please concatenate models
model = Model(inputs=vgg16.input, outputs=top_model)
#Complete the following for statement and fix the weights up to the 15th layer
for layer in model.layers[:15]:
layer.trainable = False
#Check the model structure before training
model.summary()
model.compile(loss='categorical_crossentropy',
optimizer=optimizers.SGD(lr=1e-4, momentum=0.9),
metrics=['accuracy'])
return model
Visualize the created model
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) (None, 64, 64, 3) 0
_________________________________________________________________
block1_conv1 (Conv2D) (None, 64, 64, 64) 1792
_________________________________________________________________
block1_conv2 (Conv2D) (None, 64, 64, 64) 36928
_________________________________________________________________
block1_pool (MaxPooling2D) (None, 32, 32, 64) 0
_________________________________________________________________
block2_conv1 (Conv2D) (None, 32, 32, 128) 73856
_________________________________________________________________
block2_conv2 (Conv2D) (None, 32, 32, 128) 147584
_________________________________________________________________
block2_pool (MaxPooling2D) (None, 16, 16, 128) 0
_________________________________________________________________
block3_conv1 (Conv2D) (None, 16, 16, 256) 295168
_________________________________________________________________
block3_conv2 (Conv2D) (None, 16, 16, 256) 590080
_________________________________________________________________
block3_conv3 (Conv2D) (None, 16, 16, 256) 590080
_________________________________________________________________
block3_pool (MaxPooling2D) (None, 8, 8, 256) 0
_________________________________________________________________
block4_conv1 (Conv2D) (None, 8, 8, 512) 1180160
_________________________________________________________________
block4_conv2 (Conv2D) (None, 8, 8, 512) 2359808
_________________________________________________________________
block4_conv3 (Conv2D) (None, 8, 8, 512) 2359808
_________________________________________________________________
block4_pool (MaxPooling2D) (None, 4, 4, 512) 0
_________________________________________________________________
block5_conv1 (Conv2D) (None, 4, 4, 512) 2359808
_________________________________________________________________
block5_conv2 (Conv2D) (None, 4, 4, 512) 2359808
_________________________________________________________________
block5_conv3 (Conv2D) (None, 4, 4, 512) 2359808
_________________________________________________________________
block5_pool (MaxPooling2D) (None, 2, 2, 512) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 2048) 0
_________________________________________________________________
dense_1 (Dense) (None, 256) 524544
_________________________________________________________________
dropout_1 (Dropout) (None, 256) 0
_________________________________________________________________
dense_2 (Dense) (None, 6) 1542
=================================================================
3-4. Graphing function of correct answer rate and loss function
def compare_TV(history):
#Set parameters
#Correct answer rate of classification for learning data
acc = history.history['accuracy']
#Correct answer rate of classification for validation data
val_acc = history.history['val_accuracy']
#Loss function value for training data
loss = history.history['loss']
val_loss = history.history['val_loss']
#Loss function value for validation data
epochs = range(len(acc))
# 1)Graph of correct answer rate
plt.plot(epochs, acc, 'bo' ,label = 'training acc')
plt.plot(epochs, val_acc, 'b' , label= 'validation acc')
plt.title('Training and Validation acc')
plt.legend()
plt.figure()
# 2)Loss function graph
plt.plot(epochs, loss, 'bo' ,label = 'training loss')
plt.plot(epochs, val_loss, 'b' , label= 'validation loss')
plt.title('Training and Validation loss')
plt.legend()
plt.show()
3-5. Functions that read data and train models
def main():
#Data reading
X_train, y_train, X_test, y_test = load_data()
#Model learning
model = train(X_train, y_train, X_test, y_test)
history = model.fit(X_train, y_train, batch_size=32, epochs=70, verbose=1, validation_data=(X_test, y_test))
#Evaluation and display of generalization system
score = model.evaluate(X_test, y_test, batch_size=32, verbose=0)
print('validation loss:{0[0]}\nvalidation accuracy:{0[1]}'.format(score))
compare_TV(history)
model.save('./golfer.h5')
Finally, write the main () function and execute it.
#This trains the model and generates a trained model
main()
4. Learning results
Check the accuracy rate and loss function of the learning result on the graph
① Model created by yourself
__ Correct answer rate: 0.93__
② Transfer learning
__ Correct answer rate: 0.95__
Since transfer learning was slightly better, we will adopt the transfer learning model.
The model has been created. Next time, I will write an application and publish it on heroku.
reference
I tried to classify Nogizaka46 by machine learning
Recommended Posts