层 "conv2d_transpose_4" 的输入 0 与层不兼容：预期 ndim=4，发现 ndim=2。已收到完整形状：(None, 100)
Input 0 of layer "conv2d_transpose_4" is incompatible with the layer: expected ndim=4, found ndim=2. Full shape received: (None, 100)
我正在尝试开发一个 GAN，我已经创建了生成器和鉴别器，现在我正在尝试训练它。我正在使用 Mnist 数据集，但我打算使用更多。问题是当我训练它时出现这个错误：Input 0 of layer "conv2d_transpose_4" is incompatible with the layer: expected ndim=4, found ndim=2. Full shape received: (None, 100)
我真的不知道问题出在网络上还是出在用于训练 GAN 的数据上，谁能告诉我应该如何训练它或问题出在哪里？
进口：
import tensorflow
import keras
from keras.models import Sequential, Model
from keras.layers import Dense, Dropout, Flatten, Input, BatchNormalization,
LeakyReLU, Reshape
from keras.layers import Conv2D, Conv2DTranspose, MaxPooling2D
from tensorflow.keras.optimizers import Adam
from keras import backend as K
from keras.utils import np_utils
from keras.datasets import mnist
import numpy as np
import matplotlib.pyplot as plt
import os
import cv2
发电机：
def generator():
model = Sequential()
model.add(Conv2DTranspose(32, (3,3), strides=(2,
2), activation='relu', use_bias=False,
input_shape=img_shape))
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2DTranspose(128,(3,3),strides=
(2,2), activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
model.add(Conv2DTranspose(128,(3,3),strides=
(2,2), activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.5))
model.add(BatchNormalization(momentum=0.3))
model.add(LeakyReLU(alpha=0.2))
model.add(Conv2DTranspose(128,(3,3),strides=
(2,2), activation='relu', padding='same',
use_bias=False))
model.add(BatchNormalization())
model.add(Dense(512,
activation=LeakyReLU(alpha=0.2)))
model.add(BatchNormalization(momentum=0.7))
model.build()
model.summary()
return model
判别器：
def discriminator():
model = Sequential()
model.add(Conv2D(32, (5,5), strides=(2, 2),
activation='relu', use_bias=False,
input_shape=img_shape))
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(Dropout(0.5))
model.add(BatchNormalization(momentum=0.3))
model.add(LeakyReLU(alpha=0.2))
model.add(Dense(512,
activation=LeakyReLU(alpha=0.2)))
model.add(Flatten())
model.add(BatchNormalization(momentum=0.7))
model.add(Dense(1, activation='sigmoid'))
model.build()
model.summary()
return model
训练函数：
def train(epochs, batch_size, save_interval):
(x_train, _), (_, _) = mnist.load_data()
x_train = (x_train.astype(np.float32) - 127.5) / 127.5
x_train = np.expand_dims(x_train, axis=3)
half_batch = int(batch_size / 2)
for epoch in range(epochs):
idx = np.random.randint(0, x_train.shape[0], half_batch)
imgs = x_train[idx]
noise = np.random.normal(0, 1, (half_batch, 100))
gen_imgs = generator.predict(noise)
d_loss_real = discriminator.train_on_batch(imgs, np.ones((half_batch, 1)))
d_loss_fake = discriminator.train_on_batch(gen_imgs, np.zeros((half_batch, 1)))
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
noise = np.random.normal(0, 1, (batch_size, 100))
valid_y = np.array([1] * batch_size)
g_loss = combined.train_on_batch(noise, valid_y)
print ("%d [D loss: %f, acc.: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100*d_loss[1], g_loss))
if epoch % save_interval == 0:
save_imgs(epoch)
使用的数据：
img_rows = 28
img_cols = 28
channels = 1
img_shape = (img_rows, img_cols, channels)
optimizer = Adam(0.0002, 0.5)
discriminator = discriminator()
discriminator.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
generator = generator()
generator.compile(loss='binary_crossentropy',
optimizer=optimizer)
z = Input(shape=(100,))
img = generator(z) #error
discriminator.trainable = False
valid = discriminator(img)
combined = Model(z, valid)
combined.compile(loss='binary_crossentropy',
optimizer=optimizer)
train(epochs=100000, batch_size=32,
save_interval=10)
generator.save('generator_model.h5')
问题出在 Discriminator 模型中的第一个 Flatten 层，它将您的 n-dimensional 张量转换为一维张量。由于 MaxPooling2D 层无法处理一维张量，因此您会看到该错误。如果删除它，它应该可以工作：
def discriminator():
model = Sequential()
model.add(Conv2D(32, (5,5), strides=(2, 2),
activation='relu', use_bias=False,
input_shape=img_shape))
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(Dropout(0.5))
model.add(BatchNormalization(momentum=0.3))
model.add(LeakyReLU(alpha=0.2))
model.add(Flatten())
model.add(Dense(512,
activation=LeakyReLU(alpha=0.2)))
model.add(BatchNormalization(momentum=0.7))
model.add(Dense(1, activation='sigmoid'))
model.build()
model.summary()
return model
更新 1：
尝试像这样重写您的 Generator 模型：
def generator():
model = Sequential()
model = tf.keras.Sequential()
model.add(Dense(7*7*256, use_bias=False, input_shape=(100,)))
model.add(BatchNormalization())
model.add(LeakyReLU())
model.add(Reshape((7, 7, 256)))
assert model.output_shape == (None, 7, 7, 256) # Note: None is the batch size
model.add(Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False))
assert model.output_shape == (None, 7, 7, 128)
model.add(BatchNormalization())
model.add(LeakyReLU())
model.add(Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False))
assert model.output_shape == (None, 14, 14, 64)
model.add(BatchNormalization())
model.add(LeakyReLU())
model.add(Conv2DTranspose(1, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh'))
assert model.output_shape == (None, 28, 28, 1)
model.summary()
return model
它应该可以工作，但是您一定要通过这个 tutorial 来理解所有内容。
首先，鉴别器是错误的，你期望输入噪声到具有相同维度的生成器，其中它与交叉输入相似，噪声可以在创建批处理的多个级别中生成！
当你查看每一层时，你会看到层识别需要多少！
只需测量输入/输出即可，您无需删除或编辑模型的含义。
它不是生成器 flase 当您输入错误时，请尝试查看他们正在训练批量大小和生成的输入图像的鉴别器层。
（是噪音相似度cross-entrophy）
当您使用正确的输入方式时，模型图层和形状没有变化
### name='conv2d_transpose_input'), name='conv2d_transpose_input', description="created by layer 'conv2d_transpose_input'"),
### but it was called on an input with incompatible shape (None, 100).
model = Sequential()
model.add(Conv2DTranspose(32, (3,3), strides=(2,
2), activation='relu', use_bias=False,
input_shape=(28, 28, 1)))
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2DTranspose(128,(3,3),strides=
(2,2), activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
########################
def discriminator():
model = Sequential()
### Input 0 of layer "conv2d" is incompatible with the layer: expected axis -1 of input shape to have value 1, but received input with shape (None, 114, 114, 512)
### model.add(tf.keras.layers.Reshape((1, 100), input_shape=img_shape))
model.add(Conv2D(32, (5,5), strides=(2, 2),
activation='relu', use_bias=False,
input_shape=( 28, 28, 1))) ### img_shape
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
...
我正在尝试开发一个 GAN，我已经创建了生成器和鉴别器，现在我正在尝试训练它。我正在使用 Mnist 数据集，但我打算使用更多。问题是当我训练它时出现这个错误：Input 0 of layer "conv2d_transpose_4" is incompatible with the layer: expected ndim=4, found ndim=2. Full shape received: (None, 100)
我真的不知道问题出在网络上还是出在用于训练 GAN 的数据上，谁能告诉我应该如何训练它或问题出在哪里？
进口：
import tensorflow
import keras
from keras.models import Sequential, Model
from keras.layers import Dense, Dropout, Flatten, Input, BatchNormalization,
LeakyReLU, Reshape
from keras.layers import Conv2D, Conv2DTranspose, MaxPooling2D
from tensorflow.keras.optimizers import Adam
from keras import backend as K
from keras.utils import np_utils
from keras.datasets import mnist
import numpy as np
import matplotlib.pyplot as plt
import os
import cv2
发电机：
def generator():
model = Sequential()
model.add(Conv2DTranspose(32, (3,3), strides=(2,
2), activation='relu', use_bias=False,
input_shape=img_shape))
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2DTranspose(128,(3,3),strides=
(2,2), activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
model.add(Conv2DTranspose(128,(3,3),strides=
(2,2), activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.5))
model.add(BatchNormalization(momentum=0.3))
model.add(LeakyReLU(alpha=0.2))
model.add(Conv2DTranspose(128,(3,3),strides=
(2,2), activation='relu', padding='same',
use_bias=False))
model.add(BatchNormalization())
model.add(Dense(512,
activation=LeakyReLU(alpha=0.2)))
model.add(BatchNormalization(momentum=0.7))
model.build()
model.summary()
return model
判别器：
def discriminator():
model = Sequential()
model.add(Conv2D(32, (5,5), strides=(2, 2),
activation='relu', use_bias=False,
input_shape=img_shape))
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(Dropout(0.5))
model.add(BatchNormalization(momentum=0.3))
model.add(LeakyReLU(alpha=0.2))
model.add(Dense(512,
activation=LeakyReLU(alpha=0.2)))
model.add(Flatten())
model.add(BatchNormalization(momentum=0.7))
model.add(Dense(1, activation='sigmoid'))
model.build()
model.summary()
return model
训练函数：
def train(epochs, batch_size, save_interval):
(x_train, _), (_, _) = mnist.load_data()
x_train = (x_train.astype(np.float32) - 127.5) / 127.5
x_train = np.expand_dims(x_train, axis=3)
half_batch = int(batch_size / 2)
for epoch in range(epochs):
idx = np.random.randint(0, x_train.shape[0], half_batch)
imgs = x_train[idx]
noise = np.random.normal(0, 1, (half_batch, 100))
gen_imgs = generator.predict(noise)
d_loss_real = discriminator.train_on_batch(imgs, np.ones((half_batch, 1)))
d_loss_fake = discriminator.train_on_batch(gen_imgs, np.zeros((half_batch, 1)))
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
noise = np.random.normal(0, 1, (batch_size, 100))
valid_y = np.array([1] * batch_size)
g_loss = combined.train_on_batch(noise, valid_y)
print ("%d [D loss: %f, acc.: %.2f%%] [G loss: %f]" % (epoch, d_loss[0], 100*d_loss[1], g_loss))
if epoch % save_interval == 0:
save_imgs(epoch)
使用的数据：
img_rows = 28
img_cols = 28
channels = 1
img_shape = (img_rows, img_cols, channels)
optimizer = Adam(0.0002, 0.5)
discriminator = discriminator()
discriminator.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
generator = generator()
generator.compile(loss='binary_crossentropy',
optimizer=optimizer)
z = Input(shape=(100,))
img = generator(z) #error
discriminator.trainable = False
valid = discriminator(img)
combined = Model(z, valid)
combined.compile(loss='binary_crossentropy',
optimizer=optimizer)
train(epochs=100000, batch_size=32,
save_interval=10)
generator.save('generator_model.h5')
问题出在 Discriminator 模型中的第一个 Flatten 层，它将您的 n-dimensional 张量转换为一维张量。由于 MaxPooling2D 层无法处理一维张量，因此您会看到该错误。如果删除它，它应该可以工作：
def discriminator():
model = Sequential()
model.add(Conv2D(32, (5,5), strides=(2, 2),
activation='relu', use_bias=False,
input_shape=img_shape))
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(Dropout(0.5))
model.add(BatchNormalization(momentum=0.3))
model.add(LeakyReLU(alpha=0.2))
model.add(Flatten())
model.add(Dense(512,
activation=LeakyReLU(alpha=0.2)))
model.add(BatchNormalization(momentum=0.7))
model.add(Dense(1, activation='sigmoid'))
model.build()
model.summary()
return model
更新 1：
尝试像这样重写您的 Generator 模型：
def generator():
model = Sequential()
model = tf.keras.Sequential()
model.add(Dense(7*7*256, use_bias=False, input_shape=(100,)))
model.add(BatchNormalization())
model.add(LeakyReLU())
model.add(Reshape((7, 7, 256)))
assert model.output_shape == (None, 7, 7, 256) # Note: None is the batch size
model.add(Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False))
assert model.output_shape == (None, 7, 7, 128)
model.add(BatchNormalization())
model.add(LeakyReLU())
model.add(Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False))
assert model.output_shape == (None, 14, 14, 64)
model.add(BatchNormalization())
model.add(LeakyReLU())
model.add(Conv2DTranspose(1, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh'))
assert model.output_shape == (None, 28, 28, 1)
model.summary()
return model
它应该可以工作，但是您一定要通过这个 tutorial 来理解所有内容。
首先，鉴别器是错误的，你期望输入噪声到具有相同维度的生成器，其中它与交叉输入相似，噪声可以在创建批处理的多个级别中生成！
当你查看每一层时，你会看到层识别需要多少！
只需测量输入/输出即可，您无需删除或编辑模型的含义。
它不是生成器 flase 当您输入错误时，请尝试查看他们正在训练批量大小和生成的输入图像的鉴别器层。 （是噪音相似度cross-entrophy）
当您使用正确的输入方式时，模型图层和形状没有变化
### name='conv2d_transpose_input'), name='conv2d_transpose_input', description="created by layer 'conv2d_transpose_input'"),
### but it was called on an input with incompatible shape (None, 100).
model = Sequential()
model.add(Conv2DTranspose(32, (3,3), strides=(2,
2), activation='relu', use_bias=False,
input_shape=(28, 28, 1)))
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2DTranspose(128,(3,3),strides=
(2,2), activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
########################
def discriminator():
model = Sequential()
### Input 0 of layer "conv2d" is incompatible with the layer: expected axis -1 of input shape to have value 1, but received input with shape (None, 114, 114, 512)
### model.add(tf.keras.layers.Reshape((1, 100), input_shape=img_shape))
model.add(Conv2D(32, (5,5), strides=(2, 2),
activation='relu', use_bias=False,
input_shape=( 28, 28, 1))) ### img_shape
model.add(BatchNormalization(momentum=0.3))
model.add(Conv2D(64,(5,5),strides=(2,2),
activation='relu', padding='same',
use_bias=False))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(LeakyReLU(alpha=0.2))
...