CIFAR100实战

导入模块

import os
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets,layers,optimizers,Sequential,metrics

加载数据并处理数据

def preprocess(x,y):
    x = tf.cast(x,dtype=tf.float32)/255.
    y = tf.cast(y,dtype=tf.int32)
    return x,y

(x,y),(x_test,y_test) = datasets.cifar100.load_data()
y = tf.squeeze(y,axis=1)
y_test = tf.squeeze(y_test,axis=1)
db = tf.data.Dataset.from_tensor_slices((x,y))
db = db.map(preprocess).shuffle(10000).batch(200)
db_test = tf.data.Dataset.from_tensor_slices((x_test,y_test))
db_test = db_test.map(preprocess).shuffle(10000).batch(200)
sample = next(iter(db))
print(sample[0].shape,sample[1].shape)

创建网络和优化器

#[b,32,32,3] => [b,16,16,64] => [b,8,8,128] => [b,4,4,256] => [b,2,2,512] => [b,1,1,512]
conv_net = Sequential([
    #unit1
    layers.Conv2D(64,kernel_size=3,padding='same',activation=tf.nn.relu),
    layers.Conv2D(64,kernel_size=3,padding='same',activation='relu'),
    layers.MaxPool2D(pool_size=2,padding='same',strides=2),
    #unit2
    layers.Conv2D(128,kernel_size=3,padding='same',activation=tf.nn.relu),
    layers.Conv2D(128,kernel_size=3,padding='same',activation='relu'),
    layers.MaxPool2D(pool_size=2,padding='same',strides=2),
    #unit3
    layers.Conv2D(256,kernel_size=3,padding='same',activation=tf.nn.relu),
    layers.Conv2D(256,kernel_size=3,padding='same',activation='relu'),
    layers.MaxPool2D(pool_size=2,padding='same',strides=2),
    #unit4
    layers.Conv2D(512,kernel_size=3,padding='same',activation=tf.nn.relu),
    layers.Conv2D(512,kernel_size=3,padding='same',activation='relu'),
    layers.MaxPool2D(pool_size=2,padding='same',strides=2),
    #unit5
    layers.Conv2D(512,kernel_size=3,padding='same',activation=tf.nn.relu),
    layers.Conv2D(512,kernel_size=3,padding='same',activation='relu'),
    layers.MaxPool2D(pool_size=2,padding='same',strides=2)
    ])
#[b,512] => [b,256] => [b,128] => [b,100]
fc_net = Sequential([
    layers.Dense(256,activation='relu'),
    layers.Dense(128,activation='relu'),
    layers.Dense(100)
    ])
    
conv_net.build(input_shape=[None,32,32,3])
fc_net.build(input_shape=[None,512])
variables = conv_net.trainable_variables + fc_net.trainable_variables
optimizer = optimizers.Adam(learning_rate=0.0001)

数据更新

for epoch in range(10):
    for step,(x,y) in enumerate(db):     
        #img=x[0]
        #img = tf.reshape(img,[1,32,32,3])
        #plt.subplot(231),plt.imshow(x[0],'gray')
        #plt.show()
        #[b,32,32,3] => [b,1,1,512]
        with tf.GradientTape() as tape:
            out = conv_net(x)
            #[b,1,1,512] => [b,512]
            out = tf.reshape(out,[-1,512])
            #[b,512] => [b,100]
            logits = fc_net(out)
            #[b] => [b,100]
            y_onehot = tf.one_hot(y,depth=100)
            loss = tf.reduce_mean(tf.losses.categorical_crossentropy(y_onehot,logits,from_logits=True))
        grads = tape.gradient(loss,conv_net.trainable_variables)
        optimizer.apply_gradients(zip(grads,variables))              
        if step%100==0:
            print(epoch,step,'loss:',float(loss))

测试

    total_correct = 0
    total_num = 0
    for x,y in db_test:
        out = conv_net(x)
        out = tf.reshape(out,[-1,512])
        logits = fc_net(out)
        prod = tf.nn.softmax(logits,axis=1)#将值转为[0~1]且和为1
        pred = tf.argmax(prod,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_correct += correct
        total_num += x.shape[0]
    acc = total_correct/total_num*100
    print('准确率',int(acc),'%',sep='')