在制作GAN模型时鉴别器模型层错误

Question

我基于动画的样本训练图像制作了一个生成图像的GAN模型。在执行代码时，我得到了这个错误。

ValueError: Input 0 of layer "discriminator" is incompatible with the layer: expected shape=(None, 64, 64, 3), found shape=(64, 64, 3)

即使改变鉴别器的第一层形状为(None, 64, 64, 3)也没有帮助

代码:

预处理:

import numpy as np
import tensorflow as tf
from tqdm import tqdm
from tensorflow import keras
from tensorflow.keras import layers
img_h,img_w,img_c=64,64,3
batch_size=128
latent_dim=128
num_epochs=100
dir='/home/samar/Desktop/project2/anime-gan/data'
dataset = tf.keras.utils.image_dataset_from_directory(
directory=dir,
seed=123,
image_size=(img_h, img_w),
batch_size=batch_size,
shuffle=True)
xtrain, ytrain = next(iter(dataset))
xtrain=np.array(xtrain)
xtrain=np.apply_along_axis(lambda x: x/255.0,0,xtrain)

鉴别器模型:

discriminator = keras.Sequential(
[
keras.Input(shape=(64, 64, 3)),
layers.Conv2D(64, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2D(128, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2D(128, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Flatten(),
layers.Dropout(0.2),
layers.Dense(1, activation="sigmoid"),
],
name="discriminator",
)
discriminator.summary()

发电机模型:

generator = keras.Sequential(
[
keras.Input(shape=(latent_dim,)),
layers.Dense(8 * 8 * 128),
layers.Reshape((8, 8, 128)),
layers.Conv2DTranspose(128, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2DTranspose(256, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2DTranspose(512, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2D(3, kernel_size=5, padding="same", activation="sigmoid"),
],
name="generator",
)
generator.summary()

培训:

opt_gen = keras.optimizers.Adam(1e-4)
opt_disc = keras.optimizers.Adam(1e-4)
loss_fn = keras.losses.BinaryCrossentropy()
for epoch in range(10):
for idx, real in enumerate(tqdm(xtrain)):
batch_size=real.shape[0]
random_latent_vectors = tf.random.normal(shape=(batch_size, latent_dim))
with tf.GradientTape() as gen_tape:
fake = generator(random_latent_vectors)
if idx % 100 == 0:
img = keras.preprocessing.image.array_to_img(fake[0])
img.save("/home/samar/Desktop/project2/anime-gan/gen_images/generated_img_%03d_%d.png" % (epoch, idx))    
with tf.GradientTape() as disc_tape:
loss_disc_real = loss_fn(tf.ones((batch_size,1)), discriminator(real))
loss_disc_fake = loss_fn(tf.zeros((batch_size,1)), discriminator(fake))
loss_disc = (loss_disc_real + loss_disc_fake) / 2
gradients_of_discriminator = disc_tape.gradient(loss_disc, discriminator.trainable_variables)
opt_disc.apply_gradients(zip(gradients_of_discriminator, discriminator.trainable_variables))
with tf.GradientTape() as gen_tape:
fake = generator(random_latent_vectors)
output = discriminator(fake)
loss_gen = loss_fn(tf.ones(batch_size, 1), output)
grads = gen_tape.gradient(loss_gen, generator.trainable_weights)
opt_gen.apply_gradients(zip(grads, generator.trainable_weights))

还有，你能不能给我解释一下形状(None, 64, 64, 3)和(64,64,3)的区别

Answer 1

问题是，当运行xtrain, ytrain = next(iter(train_ds))时，您正在提取恰好一批，然后在训练循环中迭代此批。这就是为什么你缺少批尺寸(None)。我不确定您的数据集是什么样子，但这里是使用tf.keras.utils.image_dataset_from_directory的一个工作示例:

import numpy as np
import tensorflow as tf
from tqdm import tqdm
from tensorflow import keras
from tensorflow.keras import layers
import pathlib
dataset_url = "https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz"
data_dir = tf.keras.utils.get_file('flower_photos', origin=dataset_url, untar=True)
data_dir = pathlib.Path(data_dir)
img_h,img_w,img_c=64,64,3
batch_size=128
latent_dim=128
num_epochs=100
train_ds = tf.keras.utils.image_dataset_from_directory(
data_dir,
seed=123,
image_size=(img_h,img_w),
batch_size=batch_size)
normalization_layer = tf.keras.layers.Rescaling(1./255)
train_ds = train_ds.map(lambda x, y: normalization_layer(x))
discriminator = keras.Sequential(
[
keras.Input(shape=(64, 64, 3)),
layers.Conv2D(64, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2D(128, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2D(128, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Flatten(),
layers.Dropout(0.2),
layers.Dense(1, activation="sigmoid"),
],
name="discriminator",
)
discriminator.summary()
generator = keras.Sequential(
[
keras.Input(shape=(latent_dim,)),
layers.Dense(8 * 8 * 128),
layers.Reshape((8, 8, 128)),
layers.Conv2DTranspose(128, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2DTranspose(256, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2DTranspose(512, kernel_size=4, strides=2, padding="same"),
layers.LeakyReLU(alpha=0.2),
layers.Conv2D(3, kernel_size=5, padding="same", activation="sigmoid"),
],
name="generator",
)
generator.summary()
opt_gen = keras.optimizers.Adam(1e-4)
opt_disc = keras.optimizers.Adam(1e-4)
loss_fn = keras.losses.BinaryCrossentropy()
for epoch in range(10):
for idx, real in enumerate(tqdm(train_ds)):
batch_size=real.shape[0]
random_latent_vectors = tf.random.normal(shape=(batch_size, latent_dim))
with tf.GradientTape() as gen_tape:
fake = generator(random_latent_vectors)
if idx % 100 == 0:
img = keras.preprocessing.image.array_to_img(fake[0])
with tf.GradientTape() as disc_tape:
loss_disc_real = loss_fn(tf.ones((batch_size,1)), discriminator(real))
loss_disc_fake = loss_fn(tf.zeros((batch_size,1)), discriminator(fake))
loss_disc = (loss_disc_real + loss_disc_fake) / 2
gradients_of_discriminator = disc_tape.gradient(loss_disc, discriminator.trainable_variables)
opt_disc.apply_gradients(zip(gradients_of_discriminator, discriminator.trainable_variables))
with tf.GradientTape() as gen_tape:
fake = generator(random_latent_vectors)
output = discriminator(fake)
loss_gen = loss_fn(tf.ones(batch_size, 1), output)
grads = gen_tape.gradient(loss_gen, generator.trainable_weights)
opt_gen.apply_gradients(zip(grads, generator.trainable_weights))