メモ4:Tensorflow 2.0 VGG 13を実現

必要なライブラリをインポート:

import os
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import datasets,layers,optimizers,Sequential,metrics
os.environ["TF_CPP_MIN_LOG_LEVEL"]='2'
tf.random.set_seed(2345)

そのうちos.EnvironセクションはTensorflow印刷の情報を減らすためにネットワーク構造を構築する.

conv_layers=[
    layers.Conv2D(64,kernel_size=[3,3],padding="same",activation=tf.nn.relu),
    layers.Conv2D(64, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2,2],strides=2,padding="same"),

    layers.Conv2D(128, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.Conv2D(128, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding="same"),

    layers.Conv2D(256, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.Conv2D(256, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding="same"),

    layers.Conv2D(512, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.Conv2D(512, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding="same"),

    layers.Conv2D(512, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.Conv2D(512, kernel_size=[3, 3], padding="same", activation=tf.nn.relu),
    layers.MaxPool2D(pool_size=[2, 2], strides=2, padding="same"),
]

オプティマイザ:

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

≪データのロード|Load Data|emdw≫:ここでは比較的一般的なCIFAR 10のデータセットを使用します.

(x_train,y_train),(x_test,y_test)=datasets.cifar10.load_data()
y_train=tf.squeeze(y_train,axis=1)
y_test=tf.squeeze(y_test,axis=1)
# print(x_train.shape,y_train.shape,x_test.shape,y_test.shape)
train_data=tf.data.Dataset.from_tensor_slices((x_train,y_train))
train_data=train_data.shuffle(1000).map(preprocess).batch(64)

test_data=tf.data.Dataset.from_tensor_slices((x_test,y_test))
test_data=test_data.map(preprocess).batch(64)

sample=next(iter(train_data))
print('sample:',sample[0].shape,sample[1].shape,
      tf.reduce_min(sample[0]),tf.reduce_max(sample[0]))

sample=next(iter(train_data))この部分は印刷train_dataの情報完備ネットワーク:

def main():
    conv_net=Sequential(conv_layers)
    # x=tf.random.normal([4,32,32,3])
    # out=conv_net(x)
    # print(out.shape)
    fc_net=Sequential([
        layers.Dense(256,activation=tf.nn.relu),
        layers.Dense(128,activation=tf.nn.relu),
        layers.Dense(10,activation=None),
    ])
    conv_net.build(input_shape=[None, 32, 32, 3])
    fc_net.build(input_shape=[None,512])
    optimizer=optimizers.Adam(lr=1e-4)

計算loss:

variables=conv_net.trainable_variables+fc_net.trainable_variables
    for epoch in range(50):
        for step,(x,y) in enumerate(train_data):
            with tf.GradientTape() as tape:
                out=conv_net(x)
                out=tf.reshape(out,[-1,512])
                logits=fc_net(out)
                y_onehot=tf.one_hot(y,depth=10)
                loss=tf.losses.categorical_crossentropy(y_onehot,logits,from_logits=True)
                loss=tf.reduce_mean(loss)
            grads=tape.gradient(loss,variables)
            optimizer.apply_gradients(zip(grads,variables))
            if step%100==0:
                print(epoch,step,'loss',float(loss))

テスト:

total_num=0
        total_correct=0
        for x,y in test_data:
            out=conv_net(x)
            out=tf.reshape(out,[-1,512])
            logits=fc_net(out)
            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()

トレーニングデータ:

0 0 loss 2.302990436553955
0 100 loss 1.9521405696868896
0 200 loss 1.9435423612594604
0 300 loss 1.6067744493484497
0 400 loss 1.5959546566009521
0 500 loss 1.734712839126587
0 600 loss 1.2384529113769531
0 700 loss 1.3307044506072998
0 acc: 0.4787
5 0 loss 0.6936513185501099
5 100 loss 0.7874761819839478
5 200 loss 0.7884306907653809
5 300 loss 0.6663026809692383
5 400 loss 0.4075947105884552
5 500 loss 0.6752095222473145
5 600 loss 0.5246847867965698
5 700 loss 0.5275574922561646
5 acc: 0.7299
10 0 loss 0.7874808311462402
10 100 loss 0.5072851181030273
10 200 loss 0.4451877772808075
10 300 loss 0.177499920129776
10 400 loss 0.13723205029964447
10 500 loss 0.2971668243408203
10 600 loss 0.25279730558395386
10 700 loss 0.36453887820243835
10 acc: 0.7355
15 0 loss 0.2800075113773346
15 100 loss 0.1841358095407486
15 200 loss 0.040746696293354034
15 300 loss 0.06615383923053741
15 400 loss 0.1183178648352623
15 500 loss 0.07481158524751663
15 600 loss 0.09398414194583893
15 700 loss 0.03665520250797272
15 acc: 0.7469
20 0 loss 0.02290465496480465
20 100 loss 0.008633529767394066
20 200 loss 0.21534058451652527
20 300 loss 0.011568240821361542
20 400 loss 0.08179830759763718
20 500 loss 0.02673691138625145
20 600 loss 0.06506452709436417
20 700 loss 0.026200752705335617
20 acc: 0.7621

トレーニングは50 epochぐらいで、ここでは20個しか展示されていませんが、検証の精度は絶えず上昇していることがわかります.後のデータは展示されていません.私もトレーニングが終わっていません.興味があれば、走ってモデルを保存して、時間があればトレーニングを続けることができます.