1、ResNet代碼
# -*- coding: utf-8 -*-
"""
Created on Wed Feb 26 19:38:01 2020
@author: HongyongHan
"""
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers,Sequential
class BasicBlock(layers.Layer):
#初始化函數
#filter_num:理解爲卷積核通道的數目，也就是channel的通道數
#stride = 1意味着對圖片不進行採樣
def __init__(self,filter_num,strides=1):
#調用母類的初始化方法
super(BasicBlock,self).__init__()
#filter_num：卷積核通道的數目.(3,3):卷積核的size
#padding='same'如果stride等於1，那麼輸出等於輸入。
#如果stride大於等於2的話，padding=same,會自動補全，
# 如果等於2的話，輸入是32x32,可能輸出是14x14,那麼如果padding=same
#會padding輸入的大小，使得輸出是16x16
self.conv1=layers.Conv2D(filter_num,(3,3),strides=strides,padding='same')
self.bn1=layers.BatchNormalization()
#非線性激活函數
self.relu=layers.Activation('relu')
#那麼這裏設置stride=1,就始終保持一樣
self.conv2=layers.Conv2D(filter_num,(3,3),strides=1,padding='same')
self.bn2=layers.BatchNormalization()
if strides != 1:
#下采樣
self.downsample=Sequential()
self.downsample.add(layers.Conv2D(filter_num,(1,1),strides=strides))
else:
self.downsample=lambda x:x
def call(self,inputs,training=None):
#[b,h,w,c]
out=self.conv1(inputs)
out=self.bn1(out)
out=self.relu(out)
out=self.conv2(out)
out=self.bn2(out)
identify=self.downsample(inputs)
output=layers.add([out,identify])
#使用tf的函數功能
output=tf.nn.relu(output)
return output
class ResNet(keras.Model):
def __init__(self,layer_dims,num_classes=100):
#layer_dims:resnet18裏面有[2,2,2,2]，也就是四個resblock
#這裏指定了一共有多少個resblock層，每個層有多少個basicblock
#後面在設置blocks的數量的時候，就是用的這裏的層的個數
#一個resblock裏面包含了兩層basicblock
#num_classes = 100:就是我們設置的輸出的類的個數
super(ResNet, self).__init__()
#實現預處理層
self.stem=Sequential([layers.Conv2D(64,(3,3),strides=(1,1)),
layers.BatchNormalization(),
layers.Activation('relu'),
layers.MaxPool2D(pool_size=(2,2),strides=(1,1),padding='same')
])
#創建4個res_block
#這裏blocks的數量是layer_dims[0]
#這裏創建的四個res_block與前面的layer_dims:[2,2,2,2]對應
#將stride設置爲2是爲了讓feature_size越來越小
self.layer1=self.build_resblock(64,layer_dims[0])
self.layer2=self.build_resblock(128,layer_dims[1],strides=2)
self.layer3=self.build_resblock(256,layer_dims[2],strides=2)
self.layer4=self.build_resblock(512,layer_dims[3],strides=2)
#out:[b,512,h,w]
#經過運算之後不能得到h和w的值，
#使用自適應的方法得到h,w
#GlobalAveragePooling2D:就是不管你的長和寬是多少
#會在某個channel上面的長和寬加起來，取均值
self.avgpool=layers.GlobalAveragePooling2D()
#創建全連接層
#這裏的Dense是用來分類的,這裏輸出是之前輸出的類別，num_classes
self.fc=layers.Dense(num_classes)
def call(self,inputs,training=None):
#完成前向運算過程
x = self.stem(inputs)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
#這裏已經變成[b,c]的shape,不需要reshape了
x=self.avgpool(x)
#這裏輸出是[b,100]
x=self.fc(x)
return x
def build_resblock(self,filter_num,blocks,strides=1):
res_blocks=Sequential()
#添加第一層basicblock
#可能有下采樣的功能的
res_blocks.add(BasicBlock(filter_num,strides))
#但是對於後面的basicblock不讓有下采樣功能
#從1開始，一直到blocks個
for _ in range(1,blocks):
#這樣只會在第一個下采樣，後面的不在下采樣，保持shape不變
res_blocks.add(BasicBlock(filter_num,strides=1))
return res_blocks
def resnet18():
return ResNet([2,2,2,2])
def resnet34():
return ResNet([3,4,6,3])
2、resnet18_train
# -*- coding: utf-8 -*-
"""
Created on Wed Feb 26 19:38:40 2020
@author: HongyongHan
"""
import tensorflow as tf
from tensorflow.keras import layers,optimizers,datasets,Sequential
import os
from ResNet import resnet18
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
tf.random.set_seed(2345)
#預處理函數
def preprocess(x,y):
#【-1，1】
x=2 * tf.cast(x,dtype=tf.float32)/255.-1
y=tf.cast(y,dtype=tf.int32)
return x,y
(x,y),(x_test,y_test)=datasets.cifar100.load_data()
#因爲y的維度是(64,1)的，需要squeeze掉。
y=tf.squeeze(y,axis=1)
y_test=tf.squeeze(y_test,axis=1)
print(x.shape,y.shape,x_test.shape,y_test.shape)
train_db=tf.data.Dataset.from_tensor_slices((x,y))
train_db=train_db.shuffle(1000).map(preprocess).batch(64)
test_db=tf.data.Dataset.from_tensor_slices((x_test,y_test))
test_db=test_db.map(preprocess).batch(64)
sample=next(iter(train_db))
print('sample',sample[0].shape,sample[1].shape,
tf.reduce_min(sample[0]),tf.reduce_max(sample[1]))
def main():
#[b,32,32,3] => [b,1,1,512]
model=resnet18()
model.build(input_shape=(None,32,32,3))
optimizer=optimizers.Adam(lr=1e-3)
for epoch in range(50):
for step,(x,y) in enumerate(train_db):
#這裏做一個前向循環,將需要求解梯度放進來
with tf.GradientTape() as tape:
#[b,32,32,3] => [b,100]
logits=model(x)
#[b] => [b,100]
y_onehot=tf.one_hot(y,depth=100)
#compute loss
loss=tf.losses.categorical_crossentropy(y_onehot,logits,from_logits=True)
#計算均值，對每個batch的均值進行計算
loss=tf.reduce_mean(loss)
#計算gradient
grads=tape.gradient(loss,model.trainable_variables)
#傳給優化器兩個參數：grads和variable，完成梯度更新
optimizer.apply_gradients(zip(grads,model.trainable_variables))
if step % 100 == 0:
print(epoch,step,'losses:',float(loss))
total_num=0
total_correct=0
for x,y in test_db:
logits=model(x)
prob=tf.nn.softmax(logits,axis=1)
pred=tf.argmax(prob,axis=1)
pred=tf.cast(pred,dtype=tf.int32)
correct=tf.cast(tf.equal(pred,y),dtype=tf.int32)
correct=tf.reduce_sum(correct)
total_num += x.shape[0]
total_correct += int(correct)
acc=total_correct / total_num
print(epoch,'acc:',acc)
if __name__ == '__main__':
main()