Explolation1 가위바위보 분류하기

1-1.MNIST 숫자 손글씨 Dataset 불러들이기

In [7]:

import tensorflow as tf
from tensorflow import keras

import numpy as np
import matplotlib.pyplot as plt

print(tf.__version__)   # Tensorflow의 버전을 출력

mnist = keras.datasets.mnist

# MNIST 데이터를 로드. 다운로드하지 않았다면 다운로드까지 자동으로 진행됩니다. 
(x_train, y_train), (x_test, y_test) = mnist.load_data()   

print(len(x_train))  # x_train 배열의 크기를 출력

2.2.0
60000

In [8]:

# index에 0에서 59999 사이 숫자를 지정해 보세요.
index=10000     
plt.imshow(x_train[index],cmap=plt.cm.binary)
plt.show()
print( (index+1), '번째 이미지의 숫자는 바로 ',  y_train[index], '입니다.')

10001 번째 이미지의 숫자는 바로  3 입니다.

참고: Matplotlib 이란?

파이썬에서 제공하는 시각화(Visualization) 패키지인 Matplotlib은 차트(chart), 플롯(plot) 등 다양한 형태로 데이터를 시각화할 수 있는 강력한 기능을 제공합니다.

1-2.데이터 전처리 하기

In [9]:

x_train_norm, x_test_norm = x_train / 255.0, x_test / 255.0
print('최소값:',np.min(x_train_norm), ' 최대값:',np.max(x_train_norm))

최소값: 0.0  최대값: 1.0

1-3.딥러닝 네트워크 설계하기

In [10]:

model=keras.models.Sequential()
model.add(keras.layers.Conv2D(16, (3,3), activation='relu', input_shape=(28,28,1)))
model.add(keras.layers.MaxPool2D(2,2))
model.add(keras.layers.Conv2D(32, (3,3), activation='relu'))
model.add(keras.layers.MaxPooling2D((2,2)))
model.add(keras.layers.Flatten())
model.add(keras.layers.Dense(32, activation='relu'))
model.add(keras.layers.Dense(10, activation='softmax'))

print('Model에 추가된 Layer 개수: ', len(model.layers))

Model에 추가된 Layer 개수:  7

In [11]:

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 26, 26, 16)        160       
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 13, 13, 16)        0         
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 11, 11, 32)        4640      
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 5, 5, 32)          0         
_________________________________________________________________
flatten (Flatten)            (None, 800)               0         
_________________________________________________________________
dense (Dense)                (None, 32)                25632     
_________________________________________________________________
dense_1 (Dense)              (None, 10)                330       
=================================================================
Total params: 30,762
Trainable params: 30,762
Non-trainable params: 0
_________________________________________________________________

1-4. 딥러닝 네트워크 학습시키기

print(x_train.shape) 을 해보면,(60000, 28, 28) 로 채널수에 대한 정보가 없습니다. 따라서 (60000, 28, 28, 1) 로 만들어 주어야 합니다

In [12]:

print("Before Reshape - x_train_norm shape: {}".format(x_train_norm.shape))
print("Before Reshape - x_test_norm shape: {}".format(x_test_norm.shape))

x_train_reshaped=x_train_norm.reshape( -1, 28, 28, 1)  # 데이터갯수에 -1을 쓰면 reshape시 자동계산됩니다.
x_test_reshaped=x_test_norm.reshape( -1, 28, 28, 1)

print("After Reshape - x_train_reshaped shape: {}".format(x_train_reshaped.shape))
print("After Reshape - x_test_reshaped shape: {}".format(x_test_reshaped.shape))

Before Reshape - x_train_norm shape: (60000, 28, 28)
Before Reshape - x_test_norm shape: (10000, 28, 28)
After Reshape - x_train_reshaped shape: (60000, 28, 28, 1)
After Reshape - x_test_reshaped shape: (10000, 28, 28, 1)

In [13]:

model.compile(optimizer='adam',
             loss='sparse_categorical_crossentropy',
             metrics=['accuracy'])

model.fit(x_train_reshaped, y_train, epochs=10)

Epoch 1/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.1962 - accuracy: 0.9405
Epoch 2/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0628 - accuracy: 0.9803
Epoch 3/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0460 - accuracy: 0.9858
Epoch 4/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0353 - accuracy: 0.9888
Epoch 5/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0292 - accuracy: 0.9906
Epoch 6/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0244 - accuracy: 0.9923
Epoch 7/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0196 - accuracy: 0.9937
Epoch 8/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0163 - accuracy: 0.9948
Epoch 9/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0137 - accuracy: 0.9955
Epoch 10/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0120 - accuracy: 0.9958

Out[13]:

<tensorflow.python.keras.callbacks.History at 0x7f3d1c175290>

1-5. 테스트 데이터로 성능을 확인하기

In [14]:

test_loss, test_accuracy = model.evaluate(x_test_reshaped,y_test, verbose=2)
print("test_loss: {} ".format(test_loss))
print("test_accuracy: {}".format(test_accuracy))

313/313 - 2s - loss: 0.0407 - accuracy: 0.9892
test_loss: 0.0406544953584671 
test_accuracy: 0.9891999959945679

1-6 모델링, 학습, 테스트

In [15]:

#바꿔 볼 수 있는 하이퍼파라미터들
n_channel_1=16
n_channel_2=32
n_dense=32
n_train_epoch=10

model=keras.models.Sequential()
model.add(keras.layers.Conv2D(n_channel_1, (3,3), activation='relu', input_shape=(28,28,1)))
model.add(keras.layers.MaxPool2D(2,2))
model.add(keras.layers.Conv2D(n_channel_2, (3,3), activation='relu'))
model.add(keras.layers.MaxPooling2D((2,2)))
model.add(keras.layers.Flatten())
model.add(keras.layers.Dense(n_dense, activation='relu'))
model.add(keras.layers.Dense(10, activation='softmax'))

model.summary()
model.compile(optimizer='adam',
             loss='sparse_categorical_crossentropy',
             metrics=['accuracy'])

# 모델 훈련
model.fit(x_train_reshaped, y_train, epochs=n_train_epoch)

# 모델 시험
test_loss, test_accuracy = model.evaluate(x_test_reshaped, y_test, verbose=2)
print("test_loss: {} ".format(test_loss))
print("test_accuracy: {}".format(test_accuracy))

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_2 (Conv2D)            (None, 26, 26, 16)        160       
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 13, 13, 16)        0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 11, 11, 32)        4640      
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 5, 5, 32)          0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 800)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 32)                25632     
_________________________________________________________________
dense_3 (Dense)              (None, 10)                330       
=================================================================
Total params: 30,762
Trainable params: 30,762
Non-trainable params: 0
_________________________________________________________________
Epoch 1/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.1991 - accuracy: 0.9413
Epoch 2/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0645 - accuracy: 0.9803
Epoch 3/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0455 - accuracy: 0.9863
Epoch 4/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0358 - accuracy: 0.9890
Epoch 5/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0293 - accuracy: 0.9909
Epoch 6/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0252 - accuracy: 0.9919
Epoch 7/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.0195 - accuracy: 0.9935
Epoch 8/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0172 - accuracy: 0.9942
Epoch 9/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0137 - accuracy: 0.9956
Epoch 10/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.0133 - accuracy: 0.9955
313/313 - 0s - loss: 0.0382 - accuracy: 0.9893
test_loss: 0.03815147280693054 
test_accuracy: 0.989300012588501

2-1 가위바위보 이미지 수집후 resize하기

In [17]:

# PIL 라이브러리가 설치되어 있지 않다면 설치
!pip install pillow   

from PIL import Image
import os, glob

print("PIL 라이브러리 import 완료!")

Requirement already satisfied: pillow in ./anaconda3/envs/aiffel/lib/python3.7/site-packages (8.0.1)
PIL 라이브러리 import 완료!

In [20]:

import os
# 가위 이미지가 저장된 디렉토리 아래의 모든 jpg 파일을 읽어들여서
image_dir_path = os.getenv("HOME") + "/aiffel/rock_scissor_paper/scissor"
print("이미지 디렉토리 경로: ", image_dir_path)

images=glob.glob(image_dir_path + "/*.jpg")  

# 파일마다 모두 28x28 사이즈로 바꾸어 저장합니다.
target_size=(28,28)
for img in images:
    old_img=Image.open(img)
    new_img=old_img.resize(target_size,Image.ANTIALIAS)
    new_img.save(img,"JPEG")

print("가위 이미지 resize 완료!")

# 바위 이미지가 저장된 디렉토리 아래의 모든 jpg 파일을 읽어들여서
image_dir_path = os.getenv("HOME") + "/aiffel/rock_scissor_paper/rock"
print("이미지 디렉토리 경로: ", image_dir_path)

images=glob.glob(image_dir_path + "/*.jpg")  

# 파일마다 모두 28x28 사이즈로 바꾸어 저장합니다.
target_size=(28,28)
for img in images:
    old_img=Image.open(img)
    new_img=old_img.resize(target_size,Image.ANTIALIAS)
    new_img.save(img,"JPEG")

print("바위 이미지 resize 완료!")

# 보 이미지가 저장된 디렉토리 아래의 모든 jpg 파일을 읽어들여서
image_dir_path = os.getenv("HOME") + "/aiffel/rock_scissor_paper/paper"
print("이미지 디렉토리 경로: ", image_dir_path)

images=glob.glob(image_dir_path + "/*.jpg")  

# 파일마다 모두 28x28 사이즈로 바꾸어 저장합니다.
target_size=(28,28)
for img in images:
    old_img=Image.open(img)
    new_img=old_img.resize(target_size,Image.ANTIALIAS)
    new_img.save(img,"JPEG")

print("보 이미지 resize 완료!")

이미지 디렉토리 경로:  /home/ssac24/aiffel/rock_scissor_paper/scissor
가위 이미지 resize 완료!
이미지 디렉토리 경로:  /home/ssac24/aiffel/rock_scissor_paper/rock
바위 이미지 resize 완료!
이미지 디렉토리 경로:  /home/ssac24/aiffel/rock_scissor_paper/paper
보 이미지 resize 완료!

2-2 이미지 라벨링 하기

In [41]:

def load_data(img_path):
    # 가위 : 0, 바위 : 1, 보 : 2
    number_of_data=3300   # 가위바위보 이미지 개수 총합에 주의하세요.
    img_size=28
    color=3
    #이미지 데이터와 라벨(가위 : 0, 바위 : 1, 보 : 2) 데이터를 담을 행렬(matrix) 영역을 생성합니다.
    imgs=np.zeros(number_of_data*img_size*img_size*color,dtype=np.int32).reshape(number_of_data,img_size,img_size,color)
    labels=np.zeros(number_of_data,dtype=np.int32)

    idx=0
    for file in glob.iglob(img_path+'/scissor/*.jpg'):
        img = np.array(Image.open(file),dtype=np.int32)
        imgs[idx,:,:,:]=img    # 데이터 영역에 이미지 행렬을 복사
        labels[idx]=0   # 가위 : 0
        idx=idx+1

    for file in glob.iglob(img_path+'/rock/*.jpg'):
        img = np.array(Image.open(file),dtype=np.int32)
        imgs[idx,:,:,:]=img    # 데이터 영역에 이미지 행렬을 복사
        labels[idx]=1   # 바위 : 1
        idx=idx+1       
    
    for file in glob.iglob(img_path+'/paper/*.jpg'):
        img = np.array(Image.open(file),dtype=np.int32)
        imgs[idx,:,:,:]=img    # 데이터 영역에 이미지 행렬을 복사
        labels[idx]=2   # 보 : 2
        idx=idx+1
        
    print("학습데이터(x_train)의 이미지 개수는",idx,"입니다.")
    return imgs, labels

image_dir_path = os.getenv("HOME") + "/aiffel/rock_scissor_paper"
(x_train, y_train)=load_data(image_dir_path)
x_train_norm = x_train/255.0   # 입력은 0~1 사이의 값으로 정규화

print("x_train shape: {}".format(x_train.shape))
print("y_train shape: {}".format(y_train.shape))

학습데이터(x_train)의 이미지 개수는 3300 입니다.
x_train shape: (3300, 28, 28, 3)
y_train shape: (3300,)

2-3 딥러닝 설계

In [48]:

import tensorflow as tf
from tensorflow import keras
import numpy as np

n_channel_1=32
n_channel_2=64
n_dense=128
n_train_epoch=5

model=keras.models.Sequential()
model.add(keras.layers.Conv2D(n_channel_1, (3,3), activation='relu', input_shape=(28,28,3)))
model.add(keras.layers.MaxPool2D(2,2))
model.add(keras.layers.Conv2D(n_channel_2, (3,3), activation='relu'))
model.add(keras.layers.MaxPooling2D((2,2)))
model.add(keras.layers.Flatten())
model.add(keras.layers.Dense(n_dense, activation='relu'))
model.add(keras.layers.Dense(10, activation='softmax'))

#model.summary()
#model.compile(optimizer='adam',
    #         loss='sparse_categorical_crossentropy',
     #        metrics=['accuracy'])

model.summary()

import matplotlib.pyplot as plt
plt.imshow(x_train[0])
print('라벨: ', y_train[0])


index=50     
plt.imshow(x_train[index],cmap=plt.cm.binary)
plt.show()
print( (index+1), '번째 이미지는 바로 ',  y_train[index], '입니다.')

Model: "sequential_7"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_14 (Conv2D)           (None, 26, 26, 32)        896       
_________________________________________________________________
max_pooling2d_14 (MaxPooling (None, 13, 13, 32)        0         
_________________________________________________________________
conv2d_15 (Conv2D)           (None, 11, 11, 64)        18496     
_________________________________________________________________
max_pooling2d_15 (MaxPooling (None, 5, 5, 64)          0         
_________________________________________________________________
flatten_7 (Flatten)          (None, 1600)              0         
_________________________________________________________________
dense_14 (Dense)             (None, 128)               204928    
_________________________________________________________________
dense_15 (Dense)             (None, 10)                1290      
=================================================================
Total params: 225,610
Trainable params: 225,610
Non-trainable params: 0
_________________________________________________________________
라벨:  0

51 번째 이미지는 바로  0 입니다.

2-4 딥러닝 네트워크 학습

In [50]:

model.compile(optimizer='adam',
             loss='sparse_categorical_crossentropy',
             metrics=['accuracy'])

model.fit(x_train, y_train, epochs=10)

Epoch 1/10
104/104 [==============================] - 4s 36ms/step - loss: 4.4448 - accuracy: 0.4770
Epoch 2/10
104/104 [==============================] - 0s 2ms/step - loss: 0.7039 - accuracy: 0.7006
Epoch 3/10
104/104 [==============================] - 0s 2ms/step - loss: 0.4769 - accuracy: 0.8039
Epoch 4/10
104/104 [==============================] - 0s 1ms/step - loss: 0.2724 - accuracy: 0.9006
Epoch 5/10
104/104 [==============================] - 0s 1ms/step - loss: 0.1938 - accuracy: 0.9291
Epoch 6/10
104/104 [==============================] - 0s 1ms/step - loss: 0.1702 - accuracy: 0.9358
Epoch 7/10
104/104 [==============================] - 0s 2ms/step - loss: 0.1095 - accuracy: 0.9606
Epoch 8/10
104/104 [==============================] - 0s 1ms/step - loss: 0.0837 - accuracy: 0.9706
Epoch 9/10
104/104 [==============================] - 0s 2ms/step - loss: 0.0859 - accuracy: 0.9715
Epoch 10/10
104/104 [==============================] - 0s 1ms/step - loss: 0.0792 - accuracy: 0.9712

Out[50]:

<tensorflow.python.keras.callbacks.History at 0x7f3cfc1cbe50>

2-5 테스트셋 만들기

In [51]:

def load_data2(img_path):
    # 가위 : 0, 바위 : 1, 보 : 2
    number_of_data=300   # 가위바위보 이미지 개수 총합에 주의하세요.
    img_size=28
    color=3
    #이미지 데이터와 라벨(가위 : 0, 바위 : 1, 보 : 2) 데이터를 담을 행렬(matrix) 영역을 생성합니다.
    imgs=np.zeros(number_of_data*img_size*img_size*color,dtype=np.int32).reshape(number_of_data,img_size,img_size,color)
    labels=np.zeros(number_of_data,dtype=np.int32)

    idx=0
    for file in glob.iglob(img_path+'/scissor_test/*.jpg'):
        img = np.array(Image.open(file),dtype=np.int32)
        imgs[idx,:,:,:]=img    # 데이터 영역에 이미지 행렬을 복사
        labels[idx]=0   # 가위 : 0
        idx=idx+1

    for file in glob.iglob(img_path+'/rock_test/*.jpg'):
        img = np.array(Image.open(file),dtype=np.int32)
        imgs[idx,:,:,:]=img    # 데이터 영역에 이미지 행렬을 복사
        labels[idx]=1   # 바위 : 1
        idx=idx+1       
    
    for file in glob.iglob(img_path+'/paper_test/*.jpg'):
        img = np.array(Image.open(file),dtype=np.int32)
        imgs[idx,:,:,:]=img    # 데이터 영역에 이미지 행렬을 복사
        labels[idx]=2   # 보 : 2
        idx=idx+1
        
    print("학습데이터(x_test)의 이미지 개수는",idx,"입니다.")
    return imgs, labels

image_dir_path = os.getenv("HOME") + "/aiffel/rock_scissor_paper"
(x_test, y_test)=load_data2(image_dir_path)
x_test_norm = x_test/255.0   # 입력은 0~1 사이의 값으로 정규화

print("x_test shape: {}".format(x_test.shape))
print("y_test shape: {}".format(y_test.shape))

학습데이터(x_test)의 이미지 개수는 300 입니다.
x_test shape: (300, 28, 28, 3)
y_test shape: (300,)

2-6 Accuracy 측정¶

In [52]:

test_loss, test_accuracy = model.evaluate(x_test, y_test, verbose=2)
print("test_loss: {} ".format(test_loss))
print("test_accuracy: {}".format(test_accuracy))

10/10 - 2s - loss: 1.4619 - accuracy: 0.7033
test_loss: 1.461875319480896 
test_accuracy: 0.70333331823349

중점적으로 체크해야할 부분

• 경로가 올바른지
• 사진을 resize했을 경우 새롭게 하는 testset도 resize 까먹지 말것
• 모델링의 경우 식을 맞출 것
• 특징을 더 상세하게 잡기위해 Conv2d값을 올릴수록 세세한 분류가 가능함