调整 Keras 变分自动编码器以对图像进行降噪

Question

我正在尝试为 VAE https://blog.keras.io/building-autoencoders-in-keras.html

调整 Keras 示例

我修改了代码，使用有噪声的 mnist 图像作为自动编码器的输入，并使用原始的、无噪声的 mnist 图像作为输出。

import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm

from keras.layers import Input, Dense, Lambda, Layer
from keras.models import Model
from keras import backend as K
from keras import metrics
from keras.datasets import mnist

batch_size = 100
original_dim = 784
latent_dim = 2
intermediate_dim = 256
epochs = 1
epsilon_std = 1.0



x = Input(shape=(original_dim,))
h = Dense(intermediate_dim, activation='relu')(x)
z_mean = Dense(latent_dim)(h)
z_log_var = Dense(latent_dim)(h)

def sampling(args):
    z_mean, z_log_var = args
    epsilon = K.random_normal(shape=(K.shape(z_mean)[0], latent_dim), mean=0.,
                              stddev=epsilon_std)
    return z_mean + K.exp(z_log_var / 2) * epsilon


z = Lambda(sampling, output_shape=(latent_dim,))([z_mean, z_log_var])

# we instantiate these layers separately so as to reuse them later
decoder_h = Dense(intermediate_dim, activation='relu')
decoder_mean = Dense(original_dim, activation='sigmoid')
h_decoded = decoder_h(z)
x_decoded_mean = decoder_mean(h_decoded)


# Custom loss layer
class CustomVariationalLayer(Layer):
    def __init__(self, **kwargs):
        self.is_placeholder = True
        super(CustomVariationalLayer, self).__init__(**kwargs)

    def vae_loss(self, x, x_decoded_mean):
        xent_loss = original_dim * metrics.binary_crossentropy(x, x_decoded_mean)
        kl_loss = - 0.5 * K.sum(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
        return K.mean(xent_loss + kl_loss)

    def call(self, inputs):
        x = inputs[0]
        x_decoded_mean = inputs[1]
        loss = self.vae_loss(x, x_decoded_mean)
        self.add_loss(loss, inputs=inputs)
        # We won't actually use the output.
        return x

y = CustomVariationalLayer()([x, x_decoded_mean])
vae = Model(x, y)
vae.compile(optimizer='rmsprop', loss=None)


# train the VAE on MNIST digits
(x_train, y_train), (x_test, y_test) = mnist.load_data()

x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.
x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))
x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))

noise_factor = 0.5
x_train_noisy = x_train + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_train.shape) 
x_test_noisy = x_test + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_test.shape) 

x_train_noisy = np.clip(x_train_noisy, 0., 1.)
x_test_noisy = np.clip(x_test_noisy, 0., 1.)



vae.fit(x_train_noisy, x_train,
        shuffle=True,
        epochs=epochs,
        batch_size=batch_size,
        validation_data=( x_test_noisy,x_test))

但我收到以下错误消息：

File "ask_vae.py", line 86, in <module>
    validation_data=( x_test_noisy,x_test))
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1574, in fit
    batch_size=batch_size)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1411, in _standardize_user_data
    exception_prefix='target')
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 58, in _standardize_input_data
    'expected no data, but got:', data)
ValueError: ('Error when checking model target: expected no data, but got:', array([[ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       ...,
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.]], dtype=float32))

该模型似乎无法接收输出；当我将输出更改为 None 时它会起作用，如下所示：

vae.fit(x_train_noisy, None,
        shuffle=True,
        epochs=epochs,
        batch_size=batch_size,
        validation_data=( x_test_noisy,None))

这是因为自定义损失层的定义方式吗？我该如何进行？

谢谢:)

Answer 1

我使用了不同的方式来定义 VAE 损失，如下所示：

https://github.com/keras-team/keras/blob/keras-2/examples/variational_autoencoder.py

我更改了它以允许对数据进行去噪。它现在可以工作了，但我必须尝试使用超参数以使其能够正确地重建原始图像。

import numpy as np
import time
import sys
import os


from scipy.stats import norm

from keras.layers import Input, Dense, Lambda
from keras.models import Model
from keras import backend as K
from keras import metrics
from keras.datasets import mnist

from keras.callbacks import ModelCheckpoint

filepath_for_w='denoise_by_VAE_weights_1.h5'



###########
##########
experiment_dir= 'exp_'+str(int(time.time()))
os.mkdir(experiment_dir)
this_script=sys.argv[0]
from shutil import copyfile
copyfile(this_script, experiment_dir+'/'+this_script)
##########
###########


batch_size = 100
original_dim = 784
latent_dim = 2
intermediate_dim = 256
epochs = 10
epsilon_std = 1.0

x = Input(batch_shape=(batch_size, original_dim))
h = Dense(intermediate_dim, activation='relu')(x)
z_mean = Dense(latent_dim)(h)
z_log_var = Dense(latent_dim)(h)


def sampling(args):
    z_mean, z_log_var = args
    epsilon = K.random_normal(shape=(batch_size, latent_dim), mean=0.,
                              stddev=epsilon_std)
    return z_mean + K.exp(z_log_var / 2) * epsilon

# note that "output_shape" isn't necessary with the TensorFlow backend
z = Lambda(sampling, output_shape=(latent_dim,))([z_mean, z_log_var])

# we instantiate these layers separately so as to reuse them later
decoder_h = Dense(intermediate_dim, activation='relu')
decoder_mean = Dense(original_dim, activation='sigmoid')
h_decoded = decoder_h(z)
x_decoded_mean = decoder_mean(h_decoded)


def vae_loss(x, x_decoded_mean):
    xent_loss = original_dim * metrics.binary_crossentropy(x, x_decoded_mean)
    kl_loss = - 0.5 * K.sum(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
    return xent_loss + kl_loss

vae = Model(x, x_decoded_mean)
vae.compile(optimizer='rmsprop', loss=vae_loss)




# train the VAE on MNIST digits
(x_train, y_train), (x_test, y_test) = mnist.load_data()


#after loading the data, change to the new experiment dir
os.chdir(experiment_dir) #
##########################

x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.
x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))
x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))


noise_factor = 0.5

x_test_noisy = x_test + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_test.shape) 
x_test_noisy = np.clip(x_test_noisy, 0., 1.)


for i in range (10):

    x_train_noisy = x_train + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_train.shape) 
    x_train_noisy = np.clip(x_train_noisy, 0., 1.)

    checkpointer=ModelCheckpoint(filepath_for_w, monitor='val_loss', verbose=0, save_best_only=True, save_weights_only=True, mode='auto', period=1)
    vae.fit(x_train_noisy, x_train,
            shuffle=True,
            epochs=epochs,
            batch_size=batch_size,
            validation_data=(x_test_noisy, x_test),
            callbacks=[checkpointer])
    vae.load_weights(filepath_for_w) 

    #print (x_train.shape)
    #print (x_test.shape)

    decoded_imgs = vae.predict(x_test,batch_size=batch_size)
    np.save('decoded'+str(i)+'.npy',decoded_imgs)


np.save('tested.npy',x_test_noisy)
#np.save ('true_catagories.npy',y_test)
np.save('original.npy',x_test)

Answer 2

我认为问题出在这里：

enter code here vae.fit(x_train_noisy, None,
    shuffle=True,
    epochs=epochs,
    batch_size=batch_size,
    validation_data=( x_test_noisy,None)

VAE 需要比较输入 v/s 输出，而您正在为它提供 xtrain 噪声但没有任何东西可以比较它（X_train_noisy、None）。

调整 Keras 变分自动编码器以对图像进行降噪

Adapting the Keras variational autoencoder for denoising images

python

machine-learning

autoencoder

keras