import urllib.request
import zipfile
import numpy as np
import os
from IPython.display import Image
import tensorflow as tf
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dropout, Dense
from tensorflow.keras.models import Sequential
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import matplotlib.pyplot as plt
%matplotlib inline
import cv2Layer (type) Output Shape Param #
%%javascript
IPython.OutputArea.auto_scroll_threshold = 50;_URL = 'https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip'
path_to_zip = tf.keras.utils.get_file('cats_and_dogs.zip', origin=_URL, extract=True)
PATH = os.path.join(os.path.dirname(path_to_zip), 'cats_and_dogs_filtered')Downloading data from https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip
68608000/68606236 [==============================] - 1s 0us/steptrain_path = os.path.join(PATH, 'train')
validation_path = os.path.join(PATH, 'validation')original_datagen = ImageDataGenerator(rescale=1./255)training_datagen = ImageDataGenerator(
rescale=1. / 255,
rotation_range=30,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=0.1,
zoom_range=0.1,
horizontal_flip=True,
fill_mode='nearest')original_generator = original_datagen.flow_from_directory(train_path,
batch_size=128,
target_size=(150, 150),
class_mode='binary'
)Found 2000 images belonging to 2 classes.training_generator = training_datagen.flow_from_directory(train_path,
batch_size=128,
shuffle=True,
target_size=(150, 150),
class_mode='binary'
)Found 2000 images belonging to 2 classes.validation_generator = training_datagen.flow_from_directory(validation_path,
batch_size=128,
shuffle=True,
target_size=(150, 150),
class_mode='binary'
)Found 1000 images belonging to 2 classes.class_map = {
0: 'Cats',
1: 'Dogs',
}print('오리지널 사진 파일')
for x, y in original_generator:
print(x.shape, y.shape)
print(y[0])
fig, axes = plt.subplots(2, 5)
fig.set_size_inches(15, 6)
for i in range(10):
axes[i//5, i%5].imshow(x[i])
axes[i//5, i%5].set_title(class_map[int(y[i])], fontsize=15)
axes[i//5, i%5].axis('off')
plt.show()
break
print('Augmentation 적용한 사진 파일')
for x, y in training_generator:
print(x.shape, y.shape)
print(y[0])
fig, axes = plt.subplots(2, 5)
fig.set_size_inches(15, 6)
for i in range(10):
axes[i//5, i%5].imshow(x[i])
axes[i//5, i%5].set_title(class_map[int(y[i])], fontsize=15)
axes[i//5, i%5].axis('off')
plt.show()
break오리지널 사진 파일
(128, 150, 150, 3) (128,)
0.0
Augmentation 적용한 사진 파일
(128, 150, 150, 3) (128,)
1.0
vgg_model=tf.keras.applications.VGG16(input_shape=(150,150,3),include_top=False)Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/vgg16/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5
58892288/58889256 [==============================] - 1s 0us/steptf.keras.utils.plot_model(vgg_model,show_shapes=True)
vgg_model.trainable=Falsei=vgg_model.input
out=vgg_model.get_layer(index=-2).output
out=tf.keras.layers.GlobalAveragePooling2D()(out)
out=tf.keras.layers.Dense(2, activation='softmax')(out)
train_cam_model=tf.keras.Model(inputs=[i],outputs=[out])train_cam_model.summary()Model: "model"
_________________________________________________________________
input_1 (InputLayer) [(None, 150, 150, 3)] 0
_________________________________________________________________
block1_conv1 (Conv2D) (None, 150, 150, 64) 1792
_________________________________________________________________
block1_conv2 (Conv2D) (None, 150, 150, 64) 36928
_________________________________________________________________
block1_pool (MaxPooling2D) (None, 75, 75, 64) 0
_________________________________________________________________
block2_conv1 (Conv2D) (None, 75, 75, 128) 73856
_________________________________________________________________
block2_conv2 (Conv2D) (None, 75, 75, 128) 147584
_________________________________________________________________
block2_pool (MaxPooling2D) (None, 37, 37, 128) 0
_________________________________________________________________
block3_conv1 (Conv2D) (None, 37, 37, 256) 295168
_________________________________________________________________
block3_conv2 (Conv2D) (None, 37, 37, 256) 590080
_________________________________________________________________
block3_conv3 (Conv2D) (None, 37, 37, 256) 590080
_________________________________________________________________
block3_pool (MaxPooling2D) (None, 18, 18, 256) 0
_________________________________________________________________
block4_conv1 (Conv2D) (None, 18, 18, 512) 1180160
_________________________________________________________________
block4_conv2 (Conv2D) (None, 18, 18, 512) 2359808
_________________________________________________________________
block4_conv3 (Conv2D) (None, 18, 18, 512) 2359808
_________________________________________________________________
block4_pool (MaxPooling2D) (None, 9, 9, 512) 0
_________________________________________________________________
block5_conv1 (Conv2D) (None, 9, 9, 512) 2359808
_________________________________________________________________
block5_conv2 (Conv2D) (None, 9, 9, 512) 2359808
_________________________________________________________________
block5_conv3 (Conv2D) (None, 9, 9, 512) 2359808
_________________________________________________________________
global_average_pooling2d (Gl (None, 512) 0
_________________________________________________________________
dense (Dense) (None, 2) 1026
=================================================================
Total params: 14,715,714
Trainable params: 1,026
Non-trainable params: 14,714,688
_________________________________________________________________train_cam_model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['acc'])epochs=20history = train_cam_model.fit(training_generator,
validation_data=(validation_generator),
steps_per_epoch=len(training_generator),
validation_steps=len(validation_generator),
epochs=epochs,
)Epoch 1/20
16/16 [==============================] - 21s 1s/step - loss: 0.6727 - acc: 0.5845 - val_loss: 0.6606 - val_acc: 0.6030
Epoch 2/20
16/16 [==============================] - 21s 1s/step - loss: 0.6408 - acc: 0.7025 - val_loss: 0.6260 - val_acc: 0.7300
Epoch 3/20
16/16 [==============================] - 21s 1s/step - loss: 0.6156 - acc: 0.7395 - val_loss: 0.5999 - val_acc: 0.7600
Epoch 4/20
16/16 [==============================] - 21s 1s/step - loss: 0.5901 - acc: 0.7695 - val_loss: 0.5802 - val_acc: 0.7700
Epoch 5/20
16/16 [==============================] - 21s 1s/step - loss: 0.5711 - acc: 0.7750 - val_loss: 0.5620 - val_acc: 0.7760
Epoch 6/20
16/16 [==============================] - 21s 1s/step - loss: 0.5524 - acc: 0.7910 - val_loss: 0.5472 - val_acc: 0.7710
Epoch 7/20
16/16 [==============================] - 21s 1s/step - loss: 0.5376 - acc: 0.7880 - val_loss: 0.5356 - val_acc: 0.7890
Epoch 8/20
16/16 [==============================] - 20s 1s/step - loss: 0.5290 - acc: 0.7830 - val_loss: 0.5210 - val_acc: 0.7840
Epoch 9/20
16/16 [==============================] - 21s 1s/step - loss: 0.5118 - acc: 0.8075 - val_loss: 0.5110 - val_acc: 0.8000
Epoch 10/20
16/16 [==============================] - 21s 1s/step - loss: 0.5054 - acc: 0.8050 - val_loss: 0.5038 - val_acc: 0.8060
Epoch 11/20
16/16 [==============================] - 21s 1s/step - loss: 0.4954 - acc: 0.8160 - val_loss: 0.4976 - val_acc: 0.7960
Epoch 12/20
16/16 [==============================] - 21s 1s/step - loss: 0.4849 - acc: 0.8100 - val_loss: 0.4867 - val_acc: 0.7960
Epoch 13/20
16/16 [==============================] - 21s 1s/step - loss: 0.4777 - acc: 0.8090 - val_loss: 0.4768 - val_acc: 0.8140
Epoch 14/20
16/16 [==============================] - 21s 1s/step - loss: 0.4707 - acc: 0.8170 - val_loss: 0.4644 - val_acc: 0.8170
Epoch 15/20
16/16 [==============================] - 21s 1s/step - loss: 0.4674 - acc: 0.8140 - val_loss: 0.4639 - val_acc: 0.8210
Epoch 16/20
16/16 [==============================] - 21s 1s/step - loss: 0.4619 - acc: 0.8250 - val_loss: 0.4568 - val_acc: 0.8010
Epoch 17/20
16/16 [==============================] - 21s 1s/step - loss: 0.4515 - acc: 0.8225 - val_loss: 0.4496 - val_acc: 0.8150
Epoch 18/20
16/16 [==============================] - 21s 1s/step - loss: 0.4428 - acc: 0.8295 - val_loss: 0.4508 - val_acc: 0.8140
Epoch 19/20
16/16 [==============================] - 21s 1s/step - loss: 0.4419 - acc: 0.8300 - val_loss: 0.4461 - val_acc: 0.8100
Epoch 20/20
16/16 [==============================] - 21s 1s/step - loss: 0.4326 - acc: 0.8280 - val_loss: 0.4396 - val_acc: 0.8170cam_model = tf.keras.models.Model([train_cam_model.input], [train_cam_model.get_layer('block5_conv3').output,train_cam_model.output])class_weights = train_cam_model.layers[-1].get_weights()[0]class_num=1class_weights[:,class_num].shape(512,)img = cv2.imread('aa.jpg')img = cv2.resize(img, (150, 150))img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)plt.imshow(img)
conv_outputs, predictions = cam_model.predict(np.reshape(img,(1,150,150,3)))s=[]conv_outputs[0].shape(9, 9, 512)class_weights[100,class_num]0.0656074conv_outputs[0][:,:,100]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. , 5.1452355, 0. , 0. , 0. ,
0. , 0. , 0. ]], dtype=float32)class_weights[100,class_num] * conv_outputs[0][:,:,100]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. , 0.3375655, 0. , 0. , 0. ,
0. , 0. , 0. ]], dtype=float32)i=0
for w in class_weights[:,class_num]:
s.append(w * conv_outputs[0][:,:,i])
i=i+1s=np.array(s)s=np.sum(s,axis=0)s.shape(9, 9)plt.imshow(s)
import cv2
cam = cv2.resize(s, (150, 150))
cam = np.maximum(cam, 0)
heatmap = (cam - cam.min()) / (cam.max() - cam.min())heatmaparray([[0. , 0. , 0. , ..., 0. , 0. ,
0. ],
[0. , 0. , 0. , ..., 0. , 0. ,
0. ],
[0. , 0. , 0. , ..., 0. , 0. ,
0. ],
...,
[0.05159856, 0.05159856, 0.05159856, ..., 0.09521208, 0.09521208,
0.09521208],
[0.05159856, 0.05159856, 0.05159856, ..., 0.09521208, 0.09521208,
0.09521208],
[0.05159856, 0.05159856, 0.05159856, ..., 0.09521208, 0.09521208,
0.09521208]], dtype=float32)plt.imshow(heatmap)
cam = cv2.applyColorMap(np.uint8(255*heatmap),cv2.COLOR_BGR2RGB)plt.imshow(cv2.add(cam,img))
class_num=0
s=[]
i=0
for w in class_weights[:,class_num]:
s.append(w * conv_outputs[0][:,:,i])
i=i+1
s=np.array(s)
s=np.sum(s,axis=0)
cam = cv2.resize(s, (150, 150))
cam = np.maximum(cam, 0)
heatmap = (cam - cam.min()) / (cam.max() - cam.min())
cam = cv2.applyColorMap(np.uint8(255*heatmap),cv2.COLOR_BGR2RGB)plt.imshow(heatmap)
plt.imshow(cv2.add(cam,img))
plt.imshow(img)