So basically I've been trying to use CNN for face recognition. And I think that my model is suffering from overfitting since the validation loss is not decreasing yet the training is doing well. For this code I used the Pins Face Recognition dataset and here is the code I used below:
output:
Data preparing
import glob
import os
import numpy as np
import matplotlib.pyplot as plt
def main():
train_data_dir = r'C:\Users\USER\Downloads\cnn-identity-classification-master\data\train'
validation_data_dir = r'C:\Users\USER\Downloads\cnn-identity-classification-master\data\val'
preped_train_data_dir = r'C:\Users\USER\Downloads\cnn-identity-classification-master\data\data_preped\train'
preped_validation_data_dir = r'C:\Users\USER\Downloads\cnn-identity-classification-master\data\data_preped\val'
if not os.path.exists(preped_train_data_dir) and not os.path.exists(preped_validation_data_dir):
os.makedirs(preped_train_data_dir)
os.makedirs(preped_validation_data_dir)
for filename in glob.iglob(train_data_dir+"\**", recursive=True):
#print(filename)
if os.path.isfile(filename): # filter dirs
file_ = filename.split("\\")[8]
file_class = filename.split("\\")[7]
img = cv2.imread(filename, cv2.IMREAD_COLOR)
face, found_face = detect_faces("data\haarcascade_frontalface_alt.xml", img)
if (found_face is True):
face = cv2.resize(face,(200, 200))
if not os.path.exists(preped_train_data_dir+"/"+file_class):
os.mkdir(preped_train_data_dir+"/"+file_class)
#print(preped_train_data_dir+"/"+file_class+"/"+file_)
cv2.imwrite(preped_train_data_dir+"/"+file_class+"/"+file_,face)
else:
print('Did not find any faces!')
for filename in glob.iglob(validation_data_dir+"\**", recursive=True):
print(filename)
if os.path.isfile(filename): # filter dirs
file_ = filename.split('\\')[8]
file_class = filename.split('\\')[7]
img = cv2.imread(filename, cv2.IMREAD_COLOR)
face, found_face = detect_faces("data\haarcascade_frontalface_alt.xml", img)
if found_face:
face = cv2.resize(face,(200, 200))
if not os.path.exists(preped_validation_data_dir+"/"+file_class):
os.mkdir(preped_validation_data_dir+"/"+file_class)
#print(preped_validation_data_dir+"/"+file_class+"/"+file_)
cv2.imwrite(preped_validation_data_dir+"/"+file_class+"/"+file_, face)
else:
print('Did not find any faces!')
def detect_faces(cascPath, img):
faces_data = []
faces_img_data = []
faceCascade = cv2.CascadeClassifier(cascPath)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = faceCascade.detectMultiScale(
gray,
scaleFactor=1.1,
minNeighbors=5,
minSize=(30, 30),
flags = cv2.CASCADE_SCALE_IMAGE
)
face = []
found_face = False
for (x, y, w, h) in faces:
cropped = gray[y-200:y+h+400, x-200:x+w+400]
faces_img_data.append(cropped)
faces_data.append([x-200, y-200, w+400, h+400])
# only returning first face found
face = faces_img_data[0]
found_face = True
if len(faces_data) > 1:
print("found multiple faces!")
return np.array(face), found_face
if __name__ == "__main__":
main()
Training :
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D
from keras.layers import Activation, Dropout, Flatten, Dense
from keras import backend as K
import keras
import uuid
import numpy as np
import os
import json
from PIL import Image as pil
def main():
model, class_dictionary, model_uuid = initialTrain(img_width = 200, img_height = 200, train_data_dir = r'C:\Users\USER\Downloads\cnn-identity-classification-master\data\data_preped\train', validation_data_dir = r'C:\Users\USER\Downloads\cnn-identity-classification-master\data\data_preped\val', model_directory_path = 'data/trainedModels',
epochs = 200, batch_size = 3)
def initialTrain(img_width, img_height, train_data_dir, validation_data_dir, model_directory_path,
epochs, batch_size):
class_dictionary = None
if K.image_data_format() == 'channels_first':
input_shape = (3, img_width, img_height)
else:
input_shape = (img_width, img_height, 3)
# this is the augmentation configuration we will use for training
train_datagen = ImageDataGenerator(
rescale=1. / 255,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True)
# this is the augmentation configuration we will use for testing:
# only rescaling
test_datagen = ImageDataGenerator(rescale=1. / 255)
train_generator = train_datagen.flow_from_directory(
train_data_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='binary')
class_dictionary = train_generator.class_indices
model = Sequential()
model.add(Conv2D(32, (5, 5), input_shape=input_shape))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(170, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(128))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(len(class_dictionary.keys())))
model.add(Activation('softmax'))
model.compile(loss='sparse_categorical_crossentropy',
optimizer="SGD",
metrics=['accuracy'])
validation_generator = test_datagen.flow_from_directory(
validation_data_dir,
target_size=(img_width, img_height),
batch_size=batch_size,
class_mode='binary')
model.fit(
train_generator,
epochs=epochs,
validation_data=validation_generator)
if not os.path.exists(model_directory_path):
os.makedirs(model_directory_path)
model_uuid = str(uuid.uuid1())
print("Model id: " + model_uuid)
model.save(model_directory_path+'/model_'+model_uuid+'.h5')
class_indices_file = open(model_directory_path+'/class_indices_file.txt','w')
class_indices_file.write(str(class_dictionary))
class_indices_file.close()
return model, class_dictionary, model_uuid
if __name__ == "__main__":
main()
