사람 얼굴 이진분류 모델 생성

목표

• 재우, 감중 얼굴을 분류하는 이진 분류 신경망 모델을 만들어보자

import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

from PIL import Image   # PIL : 이미지를 불러오는 라이브러리

 

# convert -> L : 흑백, RGB : 컬러

img = Image.open("/content/drive/MyDrive/Colab Notebooks/빅데이터 13차(딥러닝)/Data/Class 1-samples/0.jpg").convert('L')

img

 

# numpy 배열로 이미지 데이터를 변환(기계가 인식할 수 있또록 데이터를 수치화 시켜줌)

# 이미지의 크기에 맞게 가로, 세로 픽셀수로 수치화가 됨

img_array = np.array(img)

img_array

 

array([[123, 125, 129, ..., 253, 254, 254],
       [127, 127, 130, ..., 254, 254, 254],
       [129, 129, 130, ..., 254, 253, 254],
       ...,
       [123, 122, 122, ..., 155, 154, 152],
       [123, 122, 121, ..., 155, 154, 152],
       [122, 120, 119, ..., 154, 153, 150]], dtype=uint8)

 

img_array.shape

(224, 224)

# 반복문 실행시 어느정도 실행되고 있는지 %로 알려주는 라이브러리

from tqdm import tqdm

 

# 학습 데이터셋 구성하기

class1_list = []   #재우 데이터가 들어갈 자리

class2_list = []   #감중 데이터가 들어갈 자리

for i in tqdm(range(0,200,1)):

  # 1. 재우 데이터 작업

  # 경로와 이름에 맞게 이미지를 가져와서 흑백으로 변환 후 img1 변수에 담아주기

  img1 = Image.open(f"/content/drive/MyDrive/Colab Notebooks/빅데이터 13차(딥러닝)/Data/Class 1-samples/{str(i)}.jpg").convert('L')

  img_array1 = np.array(img1)

  # numpy배열 데이터를 빈 리스트에 하나씩 추가해주기

  class1_list.append(img_array1)

  # 2. 감중 데이터 작업

  img2 = Image.open(f"/content/drive/MyDrive/Colab Notebooks/빅데이터 13차(딥러닝)/Data/Class 2-samples/{str(i)}.jpg").convert('L')

  img_array2 = np.array(img2)

  # numpy배열 데이터를 빈 리스트에 하나씩 추가해주기

  class2_list.append(img_array2)

 

100%|██████████| 200/200 [00:34<00:00, 5.75it/s]

 

 

class1_list

        ...............................................
        [126, 124, 125, ..., 153, 154, 154],
        [125, 123, 124, ..., 153, 153, 153],
        [124, 123, 123, ..., 153, 153, 152]], dtype=uint8),
 array([[127, 130, 129, ..., 254, 253, 253],
        [127, 130, 129, ..., 254, 253, 253],
        [128, 130, 129, ..., 253, 254, 253],
        ...,
        [126, 126, 125, ..., 153, 154, 153],
        [125, 125, 124, ..., 153, 154, 152],
        [123, 124, 124, ..., 153, 154, 151]], dtype=uint8)]

# 리스트 자체도 numpy배열로 변환

# 사진 하나하나도 numpy배열로, 이를 담은 리스트도 numpy배열로 변환시켜줘야 한다 (신경망에 들어갈 수 있는 형태로 변경시켜주기 위함)

class1_numpy = np.array(class1_list)

class2_numpy = np.array(class2_list)

 

class1_numpy 

array([[[123, 125, 129, ..., 253, 254, 254],
        [127, 127, 130, ..., 254, 254, 254],
        [129, 129, 130, ..., 254, 253, 254],
        ...,
        [123, 122, 122, ..., 155, 154, 152],
        [123, 122, 121, ..., 155, 154, 152],
        [122, 120, 119, ..., 154, 153, 150]],

       [[127, 128, 130, ..., 254, 254, 254],
        [127, 127, 130, ..., 254, 254, 253],
        [129, 128, 130, ..., 254, 254, 253],
        
        ....................................

# concatenate : 두 배열을 순서대로 붙여주는 명령

data = np.concatenate((class1_numpy,class2_numpy))

 

class1_numpy.shape, data.shape

((200, 224, 224), (400, 224, 224))

# 정답 데이터 만들기(문제와 정답의 순서를 일정하게 맞춰줄 것)

# 0을 재우 데이터의 정답으로, 1을 감중 데이터의 정답으로 해보자

target = np.array([0]*200 + [1]*200)

target

array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
       1, 1, 1, 1])

data.shape, target.shape

((400, 224, 224), (400,))

from sklearn.model_selection import train_test_split

# train_test_split으로 데이터를 랜덤하게 섞었다

X_train, X_test, y_train, y_test = train_test_split(data, target,

                                                    test_size=0.2,

                                                    random_state=1

                                                    )

 

print(X_train.shape)

print(X_test.shape)

print(y_train.shape)

print(y_test.shape)

(320, 224, 224)
(80, 224, 224)
(320,)
(80,)

신경망 모델링

• 재우, 감중 이미지를 분류하는 신경망 모델 설계

from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense, Flatten
 
model = Sequential()

model.add(Flatten(input_shape=(224,224)))

model.add(Dense(500, activation="relu"))
model.add(Dense(300, activation="relu"))
model.add(Dense(100, activation="relu"))

model.add(Dense(1, activation="sigmoid"))
 
model.compile(loss="binary_crossentropy",
              optimizer="Adam",
              metrics=['acc'])
 
h = model.fit(X_train,y_train,
              validation_split=0.2,
              epochs=100,
              batch_size=128)

# validation_split : 자동으로 train 데이터에서 검증 데이터를 분리시켜주는 명령
# 주의점 : 분리시켜줄 때 뒤에서부터 20%를 잘라줌
#          일정한 값으로 정렬돼 있는 데이터에는 사용할 수 없음(데이터가 섞여있어야 한다)

Epoch 1/100
2/2 [==============================] - 1s 311ms/step - loss: 0.0000e+00 - acc: 1.0000 - val_loss: 0.0000e+00 - val_acc: 1.0000
Epoch 2/100
2/2 [==============================] - 0s 39ms/step - loss: 0.0000e+00 - acc: 1.0000 - val_loss: 0.0000e+00 - val_acc: 1.0000
Epoch 3/100
2/2 [==============================] - 0s 39ms/step - loss: 0.0000e+00 - acc: 1.0000 - val_loss: 0.0000e+00 - val_acc: 1.0000
Epoch 4/100
2/2 [==============================] - 0s 38ms/step - loss: 0.0000e+00 - acc: 1.0000 - val_loss: 0.0000e+00 - val_acc: 1.0000
Epoch 5/100
2/2 [==============================] - 0s 36ms/step - loss: 0.0000e+00 - acc: 1.0000 - val_loss: 0.0000e+00 - val_acc: 1.0000
...........................................................................................................................
Epoch 100/100
2/2 [==============================] - 0s 36ms/step - loss: 0.0000e+00 - acc: 1.0000 - val_loss: 0.0000e+00 - val_acc: 1.0000

plt.figure(figsize=(15,5))

# train 데이터

plt.plot(h.history['acc'],

         label='acc',

         c = 'blue',

         marker='.'

         )

# val 데이터

plt.plot(h.history['val_acc'],

         label='val_acc',

         c = 'red',

         marker='.'

         )

plt.xlabel("epochs")

plt.ylabel("accuracy")

plt.legend()

plt.show()

 

• 같은배경, 사람의 위치도 크게 변하지 않아서 학습하기 쉬운 데이터라 정확도 100%가 나왔다.

model.evaluate(X_test, y_test)

3/3 [==============================] - 0s 24ms/step - loss: 0.0000e+00 - acc: 1.0000
[0.0, 1.0]