Autoencoderを用いたGANの作成

Question

現在autoencoderに画像を入力し、autoencoderから出力された画像で教師データどうかの判定を行いautoencoderを学習させるといったGANの開発を行なっています。
標準のGANならばランダムノイズから画像生成するコードの部分の改良を行なったため、エラーが起こってるのではないかと思います。
しかし実行しようとすると以下のようなエラーが発生しています。

Traceback (most recent call last):
  File "keras_dcgan_copy.py", line 192, in <module>
    main()
  File "keras_dcgan_copy.py", line 150, in main
    discriminator_on_generator = generator_containing_discriminator(generator, discriminator)
  File "keras_dcgan_copy.py", line 107, in generator_containing_discriminator
    model.add(discriminator)
  File "/usr/local/lib/python3.6/dist-packages/keras/engine/sequential.py", line 181, in add
    output_tensor = layer(self.outputs[0])
  File "/usr/local/lib/python3.6/dist-packages/keras/engine/base_layer.py", line 457, in __call__
    output = self.call(inputs, **kwargs)
  File "/usr/local/lib/python3.6/dist-packages/keras/engine/network.py", line 564, in call
    output_tensors, _, _ = self.run_internal_graph(inputs, masks)
  File "/usr/local/lib/python3.6/dist-packages/keras/engine/network.py", line 721, in run_internal_graph
    layer.call(computed_tensor, **kwargs))
  File "/usr/local/lib/python3.6/dist-packages/keras/layers/convolutional.py", line 171, in call
    dilation_rate=self.dilation_rate)
  File "/usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py", line 3650, in conv2d
    data_format=tf_data_format)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/nn_ops.py", line 779, in convolution
    data_format=data_format)
  File "/usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/nn_ops.py", line 839, in __init__
    filter_shape[num_spatial_dims]))
ValueError: number of input channels does not match corresponding dimension of filter, 32 != 3

機械学習初心者で汚いコードかもしれませんが、ご教授お願いします。
以上がコードの全文です

from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Reshape
from keras.layers.core import Activation
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import UpSampling2D
from keras.layers.convolutional import Conv2D, MaxPooling2D
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.core import Flatten,Dropout
from keras.optimizers import Adam
import numpy as np
from PIL import Image
import os
import glob
import random

n_colors = 3

'''
def generator_model():
    model = Sequential()

    model.add(Dense(1024, input_shape=(100,)))
    model.add(Activation('tanh'))

    model.add(Dense(128 * 16 * 16))
    model.add(BatchNormalization())
    model.add(Activation('tanh'))

    model.add(Reshape((16, 16, 128)))
    model.add(UpSampling2D(size=(2, 2)))
    model.add(Conv2D(64, (5, 5), padding='same'))
    model.add(Activation('tanh'))

    model.add(UpSampling2D(size=(2, 2)))
    model.add(Conv2D(n_colors, (5, 5), padding='same'))
    model.add(Activation('tanh'))
    return model
'''

def generator_model():
    model = Sequential()
    #encode
    model.add(Conv2D(32, (5, 5),input_shape=(64, 64, n_colors),activation='relu', padding='same'))
    model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))
    
    model.add(Conv2D(16, (5, 5),activation='relu', padding='same'))
    model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))
    
    model.add(Conv2D(16, (5, 5),activation='relu', padding='same'))
    model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))
    #decode
    model.add(Conv2D(16,(5,5),activation='relu',padding='same'))
    model.add(UpSampling2D(size=(2, 2)))
    
    model.add(Conv2D(16,(5,5),activation='relu',padding='same'))
    model.add(UpSampling2D(size=(2, 2)))
    
    model.add(Conv2D(32,(5,5),activation='relu'))
    model.add(UpSampling2D(size=(2, 2)))
    return model


def discriminator_model():
    model = Sequential()

    model.add(Conv2D(64, (5, 5), input_shape=(64, 64, n_colors), padding='same'))
    model.add(LeakyReLU(alpha=0.2))
    model.add(MaxPooling2D(pool_size=(2, 2)))

    model.add(Conv2D(128, (5, 5)))
    model.add(LeakyReLU(alpha=0.2))
    model.add(MaxPooling2D(pool_size=(2, 2)))
    model.add(Flatten())

    model.add(Dense(1024))
    model.add(LeakyReLU(alpha=0.2))

    model.add(Dense(1))
    model.add(Activation('sigmoid'))
    return model

'''
def discriminator_model():
    model = Sequential()

    model.add(Conv2D(64, (5,5), strides=(2, 2), input_shape=(64, 64, 3), padding="same"))
    model.add(LeakyReLU(0.2))
    
    model.add(Conv2D(128, (5,5), strides=(2, 2)))
    model.add(LeakyReLU(0.2))
    
    model.add(Flatten())
    model.add(Dense(256))
    
    model.add(LeakyReLU(0.2))
    model.add(Dropout(0.5))
    
    model.add(Dense(1))
    model.add(Activation('sigmoid'))
    return model
'''
def generator_containing_discriminator(generator, discriminator):
    model = Sequential()
    model.add(generator)
    model.add(discriminator)
    return model

def image_batch(batch_size):
    files = glob.glob("./in_images/**/*.jpg", recursive=True)
    files = random.sample(files, batch_size)
    # print(files)
    res = []
    for path in files:
        img = Image.open(path)
        img = img.resize((64, 64))
        arr = np.array(img)
        arr = (arr - 127.5) / 127.5
        arr.resize((64, 64, n_colors))
        res.append(arr)
    return np.array(res)

def combine_images(generated_images, cols=5, rows=5):
    shape = generated_images.shape
    h = shape[1]
    w = shape[2]
    image = np.zeros((rows * h,  cols * w, n_colors))
    for index, img in enumerate(generated_images):
        if index >= cols * rows:
            break
        i = index // cols
        j = index % cols
        image[i*h:(i+1)*h, j*w:(j+1)*w, :] = img[:, :, :]
    image = image * 127.5 + 127.5
    image = Image.fromarray(image.astype(np.uint8))
    return image

def set_trainable(model, trainable):
    model.trainable = trainable
    for layer in model.layers:
        layer.trainable = trainable

def main():
    batch_size = 64
    
    discriminator = discriminator_model()
    generator = generator_model()

    discriminator_on_generator = generator_containing_discriminator(generator, discriminator)
    
    set_trainable(discriminator, False)
    discriminator_on_generator.compile(loss='binary_crossentropy', optimizer=Adam(lr=0.0002, beta_1=0.5))
    
    print(generator.summary())
    print(discriminator_on_generator.summary())

    set_trainable(discriminator, True)
    discriminator.compile(loss='binary_crossentropy', optimizer=Adam(lr=0.0002, beta_1=0.5))
    
    print(discriminator.summary())

    for i in range(30 * 1000):
        batch_images = image_batch(batch_size)
        
        # generator更新
        noise = np.random.uniform(size=[batch_size, batch_images], low=-1.0, high=1.0)
        
        generated_images = generator.predict(image_batch)
        
        # discriminatorを更新
        X = np.concatenate((batch_images, generated_images))
        # 訓練データのラベルが1、生成画像のラベルが0になるよう学習する
        y = [1] * batch_size + [0] * batch_size
        d_loss = discriminator.train_on_batch(X, y)
        
        # generator更新
        noise = np.random.uniform(size=[batch_size, batch_images], low=-1.0, high=1.0)
        
        # 生成画像をdiscriminatorにいれたときに
        # 出力が1に近くなる(訓練画像と識別される確率が高くなる)ように学習する
        g_loss = discriminator_on_generator.train_on_batch(noise, [1] * batch_size)
        
        if i % 100 == 0:
            print("step %d d_loss, g_loss : %g %g" % (i, d_loss, g_loss))
            image = combine_images(generated_images)
            os.system('mkdir -p ./gen_images')
            image.save("./gen_images/gen%05d.jpg" % i)
            generator.save_weights('generator.h5', True)
            discriminator.save_weights('discriminator.h5', True)

main()

Answer 1

generator_containing_discriminatorの中でモデルを結合してますけどgeneratorの最終層とdiscriminatorの入力層で構造が異なってませんか？

Comments

[oriharas]

回答ありがとうございます！！
実査に、構造を同じにするため、
generatorの部分のコードを以下のように変更しましたら、動きました！！
def generator_model():
model = Sequential()
#encode
model.add(Conv2D(64, (5, 5),input_shape=(64, 64, n_colors),activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))

model.add(Conv2D(32, (5, 5),activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))

model.add(Conv2D(16, (5, 5),activation='relu', padding='same'))
model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))
#decode
model.add(Conv2D(16,(5,5),activation='relu',padding='same'))
model.add(UpSampling2D(size=(2, 2)))

model.add(Conv2D(32,(5,5),activation='relu',padding='same'))
model.add(UpSampling2D(size=(2, 2)))

model.add(Conv2D(64,(5,5),activation='relu'))
model.add(UpSampling2D(size=(2, 2)))

model.add(Conv2D(n_colors,(5,5),activation='sigmoid', border_mode='same'))
return model

ほんとうにありがとうございます

[oriharas]

なんども質問知るようで申し訳ないのですが’、main()の部分
generatorの値を更新する際にエラーが起こってします。
もしよければご教授お願いします。
for i in range(30 * 1000):
batch_images = image_batch(batch_size)

# generator更新
#noise = np.random.uniform(size=[batch_size, batch_images], low=-1.0, high=1.0)
noise = np.array(size=[batch_size, batch_images], low=-1.0, high=1.0)

generated_images = generator.predict(image_batch)

# discriminatorを更新
X = np.concatenate((batch_images, generated_images))
# 訓練データのラベルが1、生成画像のラベルが0になるよう学習する
y = [1] * batch_size + [0] * batch_size
d_loss = discriminator.train_on_batch(X, y)

# generator更新
noise = np.array(size=[batch_size, batch_images], low=-1.0, high=1.0)

# 生成画像をdiscriminatorにいれたときに
# 出力が1に近くなる(訓練画像と識別される確率が高くなる)ように学習する
g_loss = discriminator_on_generator.train_on_batch(noise, [1] * batch_size)

if i % 100 == 0:
print("step %d d_loss, g_loss : %g %g" % (i, d_loss, g_loss))
image = combine_images(generated_images)
os.system('mkdir -p ./generate_images')
image.save("./gen_images/gen%05d.jpg" % i)
generator.save_weights('generator.h5', True)
discriminator.save_weights('discriminator.h5', True)
また実行を行うと以下のようなエラーが起こってします
Traceback (most recent call last):
File "keras_dcgan_copy.py", line 213, in <module>
main()
File "keras_dcgan_copy.py", line 188, in main
noise = np.array(size=[batch_size, batch_images], low=-1.0, high=1.0)
TypeError: Required argument 'object' (pos 1) not found
何卒よろしくお願いします

[iwanote]

pythonほとんど使わないのでよくわかりませんが、size=等が不要なのでは？