Scribd
Upload
60 views

NMJ406 - VGG16 Python Code

This document provides code to perform face recognition on images of three students (Hui, Kris, and Niva) using a VGG16 deep learning model. The code loads pretrained VGG16 weights, adds classification layers, and trains the model on folders of images for 20 epochs. Evaluation metrics like accuracy are calculated and graphs of loss and accuracy over epochs are generated. The trained model is saved and can then be used to classify images in a test folder.

Uploaded by

NURUL ANIS EMILLIA BINTI NAZRI STUDENT
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
60 views

NMJ406 - VGG16 Python Code

This document provides code to perform face recognition on images of three students (Hui, Kris, and Niva) using a VGG16 deep learning model. The code loads pretrained VGG16 weights, adds classification layers, and trains the model on folders of images for 20 epochs. Evaluation metrics like accuracy are calculated and graphs of loss and accuracy over epochs are generated. The trained model is saved and can then be used to classify images in a test folder.

Uploaded by

NURUL ANIS EMILLIA BINTI NAZRI STUDENT
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 5

NMJ40603 – Artificial Intelligence

Deep learning example code (VGG16)

Case study: Face recognition for three students.

Subject 1 Subject 2 Subject 3


Hui Kris Niva

Before starting run this code, make sure your images are ready in your google drive. Images must divide into
three types of folders; training, validation and testing.

Example: VGG 16 training using Google Colab

Line Code Note


1 import os
from pydrive.auth import GoogleAuth
from pydrive.drive import GoogleDrive
from google.colab import auth
from oauth2client.client import GoogleCredentials
2 !pip install Augmentor
3 from google.colab import drive
drive.mount('/content/drive')
4 from keras.layers import Input, Lambda, Conv2D, MaxPool
2D ,Dense, Flatten, Dropout
from keras.models import Model
from keras.applications.vgg16 import VGG16
from keras.applications.vgg16 import preprocess_input
from keras.preprocessing import image
from keras.preprocessing.image import ImageDataGenerato
r
from keras.models import Sequential
import numpy as np
from glob import glob
import matplotlib.pyplot as plt
5 # re-size all the images to this Make sure to
IMAGE_SIZE = [224, 224] use your own
folder path
train_path = '/content/drive/MyDrive/
DeeplearningVGGandResnet/dataset/training'
valid_path = '/content/drive/MyDrive/
DeeplearningVGGandResnet/dataset/validation'

6 vgg = VGG16(input_shape=IMAGE_SIZE + [3], weights='imag For variable


enet', include_top=False) prediction,
value 3
# don't train existing weights represents
for layer in vgg.layers: number of
    layer.trainable = False classes. In
case uses three
# our layers - you can add more if you want faces, hence, 3
x = Flatten()(vgg.output) is used.
prediction = Dense(3, activation='softmax')(x)
# create a model object
model = Model(inputs=vgg.input, outputs=prediction)

# view the structure of the model
model.summary()

# tell the model what cost and optimization method to u
se
model.compile(
    loss='categorical_crossentropy',
    optimizer='adam',
    metrics=['accuracy']
)

7 #Using ImageDataGenerator to read images from directori
es

#Rescale all images by 1/255

from keras.preprocessing.image import ImageDataGenerato
r

train_datagen = ImageDataGenerator(rescale=1./255,
                                   shear_range=0.2,
                                   zoom_range=0.2,
                                   horizontal_flip=True
)

test_datagen = ImageDataGenerator(rescale=1./255)

training_set = train_datagen.flow_from_directory(train_
path,target_size=(224, 224), batch_size=32,
class_mode='categorical')

validation_set = test_datagen.flow_from_directory(valid
_path, target_size=(224, 224), batch_size=32,
class_mode='categorical')
8 from os import listdir

label = listdir(train_path)
numClass = len(label)
print (label)

9 r = model.fit(training_set, epochs=20, validation_data=
validation_set, validation_steps=1)

10 model.save("/content/drive/MyDrive/Deeplearning- Name and save deep


VGGandResnet/modelVGG16_ep20.h5") learning model into
your own folder
11 import matplotlib as mpl
mpl.rcParams.update(mpl.rcParamsDefault)

# loss
plt.figure(1)
plt.plot(r.history['loss'], label='train loss')
plt.plot(r.history['val_loss'], label='val loss')
plt.legend()
plt.title("train-val loss graph")
plt.xlabel("Epoch")
plt.ylabel("Loss")
plt.show()
# plt.savefig("/content/drive/MyDrive/FYP/8to2ep20/
LossVal_loss.png")

# accuracies
plt.figure(2)
plt.plot(r.history['accuracy'], label='train acc')
plt.plot(r.history['val_accuracy'], label='val acc')
plt.legend()
plt.title("train-val accuracy graph")
plt.xlabel("Epoch")
plt.ylabel("Accuracy")
plt.show()
# plt.savefig("/content/drive/MyDrive/FYP/8to2ep20/
AccVal_acc.png")

12 #traning evaluation - accuracy, specificity, sensitivit
y and etc
#find the code by yourselves
Example: VGG 16 testing using Google Colab

Line Code Note


Similar with training codes from line 1 to 4
5 # re-size all the images to this Make sure to
IMAGE_SIZE = [224, 224] use your own
folder path
train_path = '/content/drive/MyDrive/
DeeplearningVGGandResnet/dataset/training'
test_path = '/content/drive/MyDrive/
DeeplearningVGGandResnet/dataset/testing'

6 vgg = VGG16(input_shape=IMAGE_SIZE + [3], weights='imag
enet', include_top=False)

# don't train existing weights
for layer in vgg.layers:
    layer.trainable = False

# our layers - you can add more if you want
x = Flatten()(vgg.output)
prediction = Dense(3, activation='softmax')(x)
# create a model object
model = Model(inputs=vgg.input, outputs=prediction)

# view the structure of the model. No need to view for


testing
# model.summary()

# tell the model what cost and optimization method to u
se
model.compile(
    loss='categorical_crossentropy',
    optimizer='adam',
    metrics=['accuracy']
)

7 #Using ImageDataGenerator to read images from directori Look similar


es with training
code line 7,
#Rescale all images by 1/255 but there has
difference at
from keras.preprocessing.image import ImageDataGenerato testing set
r variable.

train_datagen = ImageDataGenerator(rescale=1./255,
                                   shear_range=0.2,
                                   zoom_range=0.2,
                                   horizontal_flip=True
)
valid_datagen = ImageDataGenerator(rescale=1./255)

training_set = train_datagen.flow_from_directory(train_
path, target_size=(224, 224), batch_size=32,
class_mode='categorical')

testing_set = valid_datagen.flow_from_directory(test_pa
th, target_size=(224, 224), batch_size=32,
class_mode='categorical')

from os import listdir

label = listdir(train_path)
numClass = len(label)
print (label)

from keras.callbacks import ModelCheckpoint, EarlyStopp Use function


ing callbacks to
call saved
 #checkpoint model. Make
checkpoint = ModelCheckpoint("/content/drive/MyDrive/ sure to have
Deeplearning-VGGandResnet/modelVGG16_ep20.h5",verbose=  testing set as
1, save_best_only = True) validation
data.
# # don't train existing weights
# for layer in vgg.layers:
#     layer.trainable = False

r = model.fit(training_set, epochs=20, validation_data=
testing_set, validation_steps=1,
          shuffle=True,
          callbacks=checkpoint)

#confusion matrix
#find yourselves

You might also like