如何读取tensorflow 2 model里的卷积核

最新推荐文章于 2023-06-02 10:14:23 发布
tubesystem
最新推荐文章于 2023-06-02 10:14:23 发布
阅读量211
点赞数
分类专栏：工具软件文章标签： tensorflow 深度学习 python
无版权
本文链接：https://blog.csdn.net/u012065954/article/details/127889890
版权
工具软件专栏收录该内容
9 篇文章 0 订阅
订阅专栏
# https://stackoverflow.com/questions/70413900/how-to-get-conv2d-kernel-values-in-tensorflow-2
  #  在 Tensorflow 2 中获取 Conv2D 内核值 
"""
import tensorflow as tf

input_shape = (4, 28, 28, 3)
x = tf.random.normal(input_shape)
model = tf.keras.layers.Conv2D(2, 3, activation='relu', input_shape=input_shape[1:])
y = model(x)
print(model.kernel.conv1 )
print(model.kernel )"""


#todo 可以设立一个特殊的随机噪声 tensor= batch，w，h，channel，用卷积核过滤一次，高亮点代表随机噪声形状与卷积核相同。 得到卷积核的形状， 再展开为w，h 用plot打印出来

# todo   model(images, training=True) ？？    没有看到training=True 在哪里定义

# todo  显示每批次耗时，打印卷积核形状，          也许 keras 没有卷积核的输出方法？？  需要用更底层的编程方法
# 用 callbacks
# 用tensorboard
# todo 如何把数据集做成一个文件，而不是一个文件夹？

# from  https://tensorflow.google.cn/tutorials/quickstart/advanced



import tensorflow as tf
print("TensorFlow version:", tf.__version__)

from tensorflow.keras.layers import Dense, Flatten, Conv2D
from tensorflow.keras import Model

from matplotlib import pyplot as plt                                 #  <<<<<<<<<<<<<<<<<<<<<<<<<<<  画图准备



mnist = tf.keras.datasets.mnist

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

# Add a channels dimension
x_train = x_train[..., tf.newaxis].astype("float32")
x_test = x_test[..., tf.newaxis].astype("float32")


CNN_kernal=[]                                                          #  <<<<<<<<<<<<<<<<<<<<<<<<<<<  画图准备


train_ds = tf.data.Dataset.from_tensor_slices(
    (x_train, y_train)).shuffle(10000).batch(32)                                                  #  <<<<<<<<<<<<<<<<<<<<<<< batch(32)

test_ds = tf.data.Dataset.from_tensor_slices((x_test, y_test)).batch(32)                         #  <<<<<<<<<<<<<<<<<<<<<<< batch(32)




class MyModel(Model):
    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = Conv2D(16, 3, activation='relu')                                           #  <<<<<<<<<<原程序中卷积核数量  =  32个
        self.flatten = Flatten()
        self.d1 = Dense(4, activation='relu')                                                 #  <<<<<<<<<<原程序中全连接层神经元数量  =  128个
        self.d2 = Dense(10)

    def call(self, x):
        x = self.conv1(x)
        x = self.flatten(x)
        x = self.d1(x)
        # CNN_kernel=self.kernel  # 
        return self.d2(x)

# Create an instance of the model
model = MyModel()


'''

压缩卷积核，压缩全连接层神经元

--------------------------------------------
    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = Conv2D(4, 3, activation='relu')                          #  <<<<<<<<<< 卷积核数量  =  4个
        self.flatten = Flatten()
        self.d1 = Dense(32, activation='relu')                                  #  <<<<<<<<<< 全连接层神经元数量  = 32个
        self.d2 = Dense(10)
        
可以达到的准确率
Epoch 1, Loss: 0.23707012832164764, Accuracy: 93.20166778564453, Test Loss: 0.11074736714363098, Test Accuracy: 96.72000122070312
Epoch 2, Loss: 0.09355325251817703, Accuracy: 97.1500015258789, Test Loss: 0.07059688121080399, Test Accuracy: 97.66999816894531
Epoch 3, Loss: 0.06814359873533249, Accuracy: 97.97833251953125, Test Loss: 0.07042749226093292, Test Accuracy: 97.75
Epoch 4, Loss: 0.05377012863755226, Accuracy: 98.2933349609375, Test Loss: 0.07100935280323029, Test Accuracy: 97.70999908447266
Epoch 5, Loss: 0.04366404563188553, Accuracy: 98.62166595458984, Test Loss: 0.07366002351045609, Test Accuracy: 97.61000061035156

--------------------------------------------

    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = Conv2D(2, 3, activation='relu')                            #  <<<<<<<<<< 卷积核数量  =  2个
        self.flatten = Flatten()
        self.d1 = Dense(16, activation='relu')                                   #  <<<<<<<<<< 全连接层神经元数量  = 16个
        self.d2 = Dense(10)
可以达到的准确率
Epoch 1, Loss: 0.4003552496433258, Accuracy: 88.25, Test Loss: 0.2517136037349701, Test Accuracy: 93.0
Epoch 2, Loss: 0.2139425277709961, Accuracy: 93.97166442871094, Test Loss: 0.1797635555267334, Test Accuracy: 95.02999877929688
Epoch 3, Loss: 0.16504262387752533, Accuracy: 95.2316665649414, Test Loss: 0.15187139809131622, Test Accuracy: 95.69000244140625
Epoch 4, Loss: 0.13611823320388794, Accuracy: 96.02000427246094, Test Loss: 0.13357087969779968, Test Accuracy: 96.13999938964844
Epoch 5, Loss: 0.11914531141519547, Accuracy: 96.44332885742188, Test Loss: 0.1219385638833046, Test Accuracy: 96.44000244140625

--------------------------------------------

    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = Conv2D(16, 3, activation='relu')                    #  <<<<<<<<<<原程序中卷积核数量  =  16个
        self.flatten = Flatten()
        self.d1 = Dense(16, activation='relu')                              #  <<<<<<<<<<原程序中全连接层神经元数量  =  16个
        self.d2 = Dense(10)
TensorFlow version: 2.3.0
Epoch 1, Loss: 0.2446216195821762, Accuracy: 92.91166687011719, Test Loss: 0.1149502843618393, Test Accuracy: 96.36000061035156
Epoch 2, Loss: 0.09467839449644089, Accuracy: 97.23332977294922, Test Loss: 0.07694929093122482, Test Accuracy: 97.63999938964844
Epoch 3, Loss: 0.06800872832536697, Accuracy: 97.96333312988281, Test Loss: 0.07287246733903885, Test Accuracy: 97.53999328613281
Epoch 4, Loss: 0.05138058960437775, Accuracy: 98.46833038330078, Test Loss: 0.06521112471818924, Test Accuracy: 97.89999389648438
Epoch 5, Loss: 0.03978590667247772, Accuracy: 98.7316665649414, Test Loss: 0.06942927837371826, Test Accuracy: 97.80999755859375

--------------------------------------------

    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = Conv2D(8, 3, activation='relu')                      #  <<<<<<<<<<原程序中卷积核数量  =  8个
        self.flatten = Flatten()
        self.d1 = Dense(8, activation='relu')                           #  <<<<<<<<<<原程序中全连接层神经元数量  =  8个
        self.d2 = Dense(10)

Epoch 1, Loss: 0.3922784924507141, Accuracy: 88.51167297363281, Test Loss: 0.20396079123020172, Test Accuracy: 94.19999694824219
Epoch 2, Loss: 0.15926599502563477, Accuracy: 95.5, Test Loss: 0.12827329337596893, Test Accuracy: 96.30000305175781
Epoch 3, Loss: 0.11262572556734085, Accuracy: 96.74500274658203, Test Loss: 0.10609399527311325, Test Accuracy: 96.88999938964844
Epoch 4, Loss: 0.09163926541805267, Accuracy: 97.27333068847656, Test Loss: 0.09585564583539963, Test Accuracy: 97.1199951171875
Epoch 5, Loss: 0.07721196115016937, Accuracy: 97.76333618164062, Test Loss: 0.09701179713010788, Test Accuracy: 97.19000244140625





    def __init__(self):
        super(MyModel, self).__init__()
        self.conv1 = Conv2D(16, 3, activation='relu')                         #  <<<<<<<<<<原程序中卷积核数量  =  16个
        self.flatten = Flatten()
        self.d1 = Dense(4, activation='relu')                                #  <<<<<<<<<<原程序中全连接层神经元数量  =  4个
        self.d2 = Dense(10)
Epoch 1, Loss: 0.8463388085365295, Accuracy: 72.0183334350586, Test Loss: 0.49747708439826965, Test Accuracy: 86.02999877929688
Epoch 2, Loss: 0.4489363729953766, Accuracy: 87.56333923339844, Test Loss: 0.4068147838115692, Test Accuracy: 88.87000274658203
Epoch 3, Loss: 0.38576701283454895, Accuracy: 89.4000015258789, Test Loss: 0.3815392553806305, Test Accuracy: 89.7800064086914
Epoch 4, Loss: 0.34514060616493225, Accuracy: 90.56166076660156, Test Loss: 0.343354195356369, Test Accuracy: 90.90999603271484
Epoch 5, Loss: 0.3147679567337036, Accuracy: 91.45833587646484, Test Loss: 0.31742098927497864, Test Accuracy: 91.75
Epoch 6, Loss: 0.2888678312301636, Accuracy: 92.05999755859375, Test Loss: 0.29518914222717285, Test Accuracy: 92.40999603271484
Epoch 7, Loss: 0.2697764039039612, Accuracy: 92.64666748046875, Test Loss: 0.291316956281662, Test Accuracy: 92.18000030517578
Epoch 8, Loss: 0.2539854943752289, Accuracy: 93.01666259765625, Test Loss: 0.27717989683151245, Test Accuracy: 92.73999786376953
Epoch 9, Loss: 0.24011439085006714, Accuracy: 93.46500396728516, Test Loss: 0.27109673619270325, Test Accuracy: 92.65999603271484
Epoch 10, Loss: 0.22615042328834534, Accuracy: 93.69833374023438, Test Loss: 0.26570791006088257, Test Accuracy: 92.73999786376953
Epoch 11, Loss: 0.21567215025424957, Accuracy: 93.98666381835938, Test Loss: 0.26256102323532104, Test Accuracy: 93.08999633789062
Epoch 12, Loss: 0.20541338622570038, Accuracy: 94.34500122070312, Test Loss: 0.2572975754737854, Test Accuracy: 93.11000061035156
Epoch 13, Loss: 0.19656269252300262, Accuracy: 94.538330078125, Test Loss: 0.250868022441864, Test Accuracy: 93.08000183105469
Epoch 14, Loss: 0.1892404407262802, Accuracy: 94.73333740234375, Test Loss: 0.26545971632003784, Test Accuracy: 93.02999877929688
Epoch 15, Loss: 0.18099074065685272, Accuracy: 95.00833129882812, Test Loss: 0.24252015352249146, Test Accuracy: 93.5999984741211
Epoch 16, Loss: 0.17358624935150146, Accuracy: 95.17499542236328, Test Loss: 0.23873189091682434, Test Accuracy: 93.37000274658203
Epoch 17, Loss: 0.16649213433265686, Accuracy: 95.29167175292969, Test Loss: 0.2428164929151535, Test Accuracy: 93.4000015258789
Epoch 18, Loss: 0.16083897650241852, Accuracy: 95.48332977294922, Test Loss: 0.2358662188053131, Test Accuracy: 93.87999725341797
Epoch 19, Loss: 0.1541549265384674, Accuracy: 95.62666320800781, Test Loss: 0.23503205180168152, Test Accuracy: 93.55999755859375
Epoch 20, Loss: 0.14946123957633972, Accuracy: 95.76000213623047, Test Loss: 0.2402830719947815, Test Accuracy: 93.68000030517578


--------------------------------------------


'''


loss_object = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)

optimizer = tf.keras.optimizers.Adam()



train_loss = tf.keras.metrics.Mean(name='train_loss')
train_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(name='train_accuracy')

test_loss = tf.keras.metrics.Mean(name='test_loss')
test_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(name='test_accuracy')


# @tf.function
def train_step(images, labels):
    with tf.GradientTape() as tape:
        # training=True is only needed if there are layers with different
        # behavior during training versus inference (e.g. Dropout).
        predictions = model(images, training=True)
        loss = loss_object(labels, predictions)
    gradients = tape.gradient(loss, model.trainable_variables)
    optimizer.apply_gradients(zip(gradients, model.trainable_variables))

    train_loss(loss)
    train_accuracy(labels, predictions)
    
    
# @tf.function
def test_step(images, labels):
    # training=False is only needed if there are layers with different
    # behavior during training versus inference (e.g. Dropout).
    predictions = model(images, training=False)
    t_loss = loss_object(labels, predictions)

    test_loss(t_loss)
    test_accuracy(labels, predictions)
    # print("self.conv1", CNN_kernal)

    # print("model.kernel.shape=========", model.kernel.shape)  #<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<,
    
    
EPOCHS = 1                                                          # epoch <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< 

for epoch in range(EPOCHS):
    # Reset the metrics at the start of the next epoch
    train_loss.reset_states()
    train_accuracy.reset_states()
    test_loss.reset_states()
    test_accuracy.reset_states()

    for images, labels in train_ds:
        train_step(images, labels)

    for test_images, test_labels in test_ds:
        test_step(test_images, test_labels)

    print(
        f'Epoch {epoch + 1}, '
        f'Loss: {train_loss.result()}, '
        f'Accuracy: {train_accuracy.result() * 100}, '
        f'Test Loss: {test_loss.result()}, '
        f'Test Accuracy: {test_accuracy.result() * 100}'
        )   
    
    print("model.conv1.kernel=========", model.conv1.kernel)
    print("model.conv1.kernel.shape=========", model.conv1.kernel.shape)
如何读取tensorflow 2 model里的卷积核