感悟:
機子太差，參數太多內存炸了
keras真香
參考：
https://blog.csdn.net/weixin_43624538/article/details/84563093
https://blog.csdn.net/shankezh/article/details/87814520
https://blog.csdn.net/m0_37917271/article/details/82286252
代碼：
tensorflow版本，可能有錯誤= =
import tensorflow as tf
import cv2
import numpy as np
import os
from PIL import Image
from sklearn.model_selection import train_test_split
class VGG(object):
def __init__(self):
self.n_input = 27*27*3
self.n_classes = 2
self.batch_size = 10
self.training_iters = 20
self.display_step = 20
self.learning_rate = 0.001
def conv2d(self,x,filter,k_size,stride=[1,1],padding='SAME',activation = tf.nn.relu,scope='conv2d'):
return tf.layers.conv2d(inputs=x,filters=filter,kernel_size=k_size,
strides=stride,padding=padding,name=scope,activation=activation)
def maxpool2d(self,x,pool_size=[2,2],stride=[2,2],padding='SAME',scope='maxpool2d'):
return tf.layers.max_pooling2d(inputs=x,pool_size=pool_size,strides=stride,padding=padding,name=scope)
def dropoutx(self,x,d_rate):
return tf.layers.dropout(x,rate=d_rate)
def norm(self, x, l_size, bias=1.0, alpha=0.001 / 9.0, beta=0.75, scope='norm'):
return tf.nn.lrn(x, l_size, bias=bias, alpha=alpha, beta=beta, name=scope)
def set_net(self,x,d_rate=0.8):
x = tf.reshape(x,[-1,27,27,3])
net = self.conv2d(x,filter=64,k_size=[3,3],scope='conv1_1')
net = self.conv2d(net,filter=64,k_size=[3,3],scope='conv1_2')
net = self.maxpool2d(net,scope='pool1')
net = self.conv2d(x, filter=128, k_size=[3, 3], scope='conv2_1')
net = self.conv2d(net, filter=128, k_size=[3, 3], scope='conv2_2')
net = self.maxpool2d(net, scope='pool2')
net = self.conv2d(x, filter=256, k_size=[3, 3], scope='conv3_1')
net = self.conv2d(net, filter=256, k_size=[3, 3], scope='conv3_2')
net = self.conv2d(net, filter=256, k_size=[3, 3], scope='conv3_3')
net = self.maxpool2d(net, scope='pool3')
shape = net.get_shape()
print(shape)
len = shape[1].value*shape[2].value*shape[3].value
print(len)
net = tf.reshape(net,[-1, len])
net = tf.layers.dense(net,4096,activation=tf.nn.relu,use_bias=True,name='fc1')
net = self.dropoutx(net,d_rate)
net = tf.layers.dense(net,4096,activation=tf.nn.relu,use_bias=True,name='fc2')
net = self.dropoutx(net,d_rate)
out = tf.layers.dense(net,2,activation=tf.nn.relu,use_bias=True,name='fc3')
return out
def avgg_prediction(self, X,Y, scope='vgg'):
X_train, X_vaild, y_train, y_vaild = train_test_split(X, Y, test_size=0.2)
x = tf.placeholder(tf.float32, [None, self.n_input])
y = tf.placeholder(tf.float32, [None, self.n_classes])
pred = self.set_net(x) # pred是計算完的值，此時還沒歸一化
a = tf.nn.softmax(pred) # a是歸一化後的值。
# 定義損失函數和學習步驟
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y)) # 這個是損失loss
optimizer = tf.train.AdamOptimizer(learning_rate=self.learning_rate).minimize(cost) # 最小化loss
# 測試網絡
correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
# 初始化所有的共享變量
init = tf.initialize_all_variables()
with tf.Session() as sess:
sess.run(init)
step = 1
# Keep training until reach max iterations
while step * self.batch_size < self.training_iters: # 直到達到最大迭代次數，沒考慮梯度！！！
batch_xs, batch_ys = X_train[self.batch_size*(step-1):self.batch_size*step],y_train[self.batch_size*(step-1):self.batch_size*step]
batch_xs = np.reshape(batch_xs,(-1,27*27*3))
print('~!!!!!!!!',batch_xs.shape)
print(batch_ys.shape)
# 獲取批數據
sess.run(optimizer, feed_dict={x: batch_xs, y: batch_ys})
if step % self.display_step == 0: # 每一步裏有64batch，64*20=1280
# 計算精度
acc = sess.run(accuracy, feed_dict={x: batch_xs, y: batch_ys})
# 計算損失值
loss = sess.run(cost, feed_dict={x: batch_xs, y: batch_ys})
print("Iter " + str(step * self.batch_size) + ", Minibatch Loss= " + "{:.6f}".format(
loss) + ", Training Accuracy = " + "{:.5f}".format(acc))
step += 1
print("Optimization Finished!")
# 計算測試精度
X_vaild = np.reshape(X_vaild, (-1, 27 * 27 * 3))
print("Testing Accuracy:", sess.run(accuracy,
feed_dict={x: X_vaild[:256], y: y_vaild[:256],
})) # 拿前256個來測試
print("Testing Result:", sess.run(a, feed_dict={x: X_vaild[63:64], y: y_vaild[63:64],
})) # 數組範圍，從0開始，含左不含右
print(y_vaild[63:64])
keras版本，真香
import cv2
import numpy as np
import os
from PIL import Image
valid_exts = [".jpg",".gif",".png",".tga", ".jpeg"]
n = 800
cwd = os.getcwd()
width = 128
paths = {"images/cats", "images/dogs"}
nclass = len(paths)
X = np.zeros((n,width,width,3),dtype=np.uint8)
Y = np.zeros((n,),dtype=np.uint8)
imgcnt = 0
for i,relpath in enumerate(paths):
fullpath = cwd + "/" + relpath
print(fullpath)
flist = os.listdir(fullpath)
for f in flist:
if os.path.splitext(f)[1].lower() not in valid_exts:
continue
path = os.path.join(fullpath, f)
img = Image.open(path)
img = img.convert("RGB")
img = np.array(img)
#print(img.shape)
X[imgcnt] = cv2.resize(img,(width,width))
Y[imgcnt] = i
imgcnt += 1
print(imgcnt)
import random
import matplotlib.pyplot as plt
%matplotlib inline
%config InlineBackend.figure_format = 'retina'
plt.figure(figsize=(12,10))
for i in range(12):
random_index = random.randint(0,n-1)
plt.subplot(3,4,1+i)
plt.imshow(X[random_index])
plt.title(['dog','cat'][Y[random_index]])
from sklearn.model_selection import train_test_split
X_train,X_vaild,y_train,y_vaild = train_test_split(X,Y,test_size=0.2)
from keras.layers import *
from keras.models import *
inputs = Input((width,width,3))
x = inputs
for i , layer_num in enumerate([2,3,3,3]):
for j in range(layer_num):
x = Conv2D(32*2**i,3,padding='same',activation='relu')(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(2)(x)
x = GlobalAveragePooling2D()(x)
x = Dropout(0.5)(x)
x = Dense(1,activation='sigmoid')(x)
model = Model(inputs,x)
model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
h = model.fit(X_train,y_train,batch_size=128,epochs=20,validation_data=(X_vaild,y_vaild))