我已经实现了一个条件 WGAN-GP，它可以很好地从 0-9 采样数字，但是一旦我想采样单个数字，我就会遇到维度问题。
noise = np.random.normal(0,1,(1,100))
labels = np.array([0., 0., 1.0, 0., 0., 0., 0., 0., 0., 0.]).reshape(-1,1)
gen_imgs = self.generator.predict([noise, labels])
给我：ValueError: Error when checking : expected input_2 to have shape (10,) but got array with shape (1,)。
但是，对 10 个图像进行采样可以正常工作。
r, c = 2, 5
noise = np.random.normal(0, 1, (r * c, self.latent_dim))
labels = to_categorical(np.arange(0, 10).reshape(-1, 1))
gen_imgs = self.generator.predict([noise, labels])
我怎样才能只采样一个数字？
完整代码如下：
from __future__ import division, print_function
import sys
from functools import partial
import keras.backend as K
import matplotlib.pyplot as plt
import numpy as np
from keras.datasets import mnist
from keras.layers import (Activation, BatchNormalization, Dense, Dropout,
Flatten, Input, Reshape, ZeroPadding2D, concatenate)
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.convolutional import Conv2D, UpSampling2D
from keras.layers.merge import _Merge
from keras.models import Model, Sequential
from keras.optimizers import Adam, RMSprop
from keras.utils import plot_model, to_categorical
class RandomWeightedAverage(_Merge):
"""Provides a (random) weighted average between real and generated image samples"""
def _merge_function(self, inputs):
alpha = K.random_uniform(shape=K.shape(inputs[0]))
return (alpha * inputs[0]) + ((1 - alpha) * inputs[1])
class WGANGP():
def __init__(self):
self.img_rows = 28
self.img_cols = 28
self.channels = 1
self.img_shape = (self.img_rows*self.img_cols)
self.latent_dim = 100
self.label_dim = 10
self.layers = [256, 128]
# Following parameter and optimizer set as recommended in paper
self.n_critic = 5
optimizer = Adam(lr=0.0001, beta_1=0.5, beta_2=0.9)
optimizer = RMSprop(lr=0.0001)
# Build the generator and critic
self.generator = self.build_generator()
self.critic = self.build_critic()
# -------------------------------
# Construct Computational Graph
# for the Critic
# -------------------------------
# Freeze generator's layers while training critic
self.generator.trainable = False
# Image input (real sample)
real_img = Input(shape=(self.img_shape,))
label = Input(shape=(self.label_dim,))
# Noise input
z_disc = Input(shape=(100,))
# Generate image based of noise (fake sample)
fake_img = self.generator([z_disc, label])
# Discriminator determines validity of the real and fake images
fake = self.critic([fake_img, label])
valid = self.critic([real_img, label])
# Construct weighted average between real and fake images
interpolated_img = RandomWeightedAverage()([real_img, fake_img])
# Determine validity of weighted sample
validity_interpolated = self.critic([interpolated_img, label])
# Use Python partial to provide loss function with additional
# 'averaged_samples' argument
partial_gp_loss = partial(self.gradient_penalty_loss,
averaged_samples=interpolated_img)
partial_gp_loss.__name__ = 'gradient_penalty' # Keras requires function names
self.critic_model = Model(inputs=[real_img, z_disc, label],
outputs=[valid, fake, validity_interpolated])
self.critic_model.compile(loss=[self.wasserstein_loss,
self.wasserstein_loss,
partial_gp_loss],
optimizer=optimizer,
loss_weights=[1, 1, 10])
# -------------------------------
# Construct Computational Graph
# for Generator
# -------------------------------
# For the generator we freeze the critic's layers
self.critic.trainable = False
self.generator.trainable = True
# Sampled noise for input to generator
z_gen = Input(shape=(100,))
# Generate images based of noise
img = self.generator([z_gen, label])
# Discriminator determines validity
valid = self.critic([img, label])
# Defines generator model
self.generator_model = Model([z_gen, label], valid)
self.generator_model.compile(loss=self.wasserstein_loss, optimizer=optimizer)
def gradient_penalty_loss(self, y_true, y_pred, averaged_samples):
"""
Computes gradient penalty based on prediction and weighted real / fake samples
"""
gradients = K.gradients(y_pred, averaged_samples)[0]
# compute the euclidean norm by squaring ...
gradients_sqr = K.square(gradients)
# ... summing over the rows ...
gradients_sqr_sum = K.sum(gradients_sqr,
axis=np.arange(1, len(gradients_sqr.shape)))
# ... and sqrt
gradient_l2_norm = K.sqrt(gradients_sqr_sum)
# compute lambda * (1 - ||grad||)^2 still for each single sample
gradient_penalty = K.square(1 - gradient_l2_norm)
# return the mean as loss over all the batch samples
return K.mean(gradient_penalty)
def wasserstein_loss(self, y_true, y_pred):
return K.mean(y_true * y_pred)
def build_generator(self):
noise = Input(shape=(self.latent_dim,))
label = Input(shape=(self.label_dim,))
x = concatenate([noise, label])
for l in self.layers[::-1][1:]:
x = Dense(l)(x)
x = BatchNormalization(momentum=0.8)(x)
x = Activation('relu')(x)
out = Dense(self.img_shape, activation='tanh')(x)
return Model([noise, label], out)
def build_critic(self):
# todo: convert to functional api
model = Sequential()
nodes = int(np.median(self.layers))
activation = LeakyReLU(0.2)
model.add(Dense(nodes, input_dim=self.img_shape+self.label_dim))
model.add(activation)
if len(self.layers) > 1:
for l in self.layers[1:]:
model.add(Dense(nodes))
model.add(activation)
model.add(Dense(1, activation='linear'))
model.summary()
data = Input(shape=(self.img_shape, ))
label = Input(shape=(self.label_dim, ))
model_input = concatenate([data, label])
validity = model(model_input)
discriminator = Model([data, label], validity)
return discriminator
def train(self, epochs, batch_size, sample_interval=100):
# Load the dataset
(X_train, Y_train), (_, _) = mnist.load_data()
# Rescale -1 to 1
X_train = (X_train.astype(np.float32) - 127.5) / 127.5
X_train = X_train.reshape((len(X_train), np.prod(X_train.shape[1:])))
Y_train = to_categorical(Y_train)
# Adversarial ground truths
valid = -np.ones((batch_size, 1))
fake = np.ones((batch_size, 1))
dummy = np.zeros((batch_size, 1)) # Dummy gt for gradient penalty
for epoch in range(epochs):
for _ in range(self.n_critic):
# ---------------------
# Train Discriminator
# ---------------------
# Select a random batch of images
idx = np.random.randint(0, X_train.shape[0], batch_size)
imgs = X_train[idx]
labels = Y_train[idx]
# Sample generator input
noise = np.random.normal(0, 1, (batch_size, self.latent_dim))
# Train the critic
d_loss = self.critic_model.train_on_batch([imgs, noise, labels],
[valid, fake, dummy])
# ---------------------
# Train Generator
# ---------------------
idx = np.random.randint(0, X_train.shape[0], batch_size)
noise = np.random.normal(0, 1, (batch_size, self.latent_dim))
labels = Y_train[idx]
g_loss = self.generator_model.train_on_batch([noise, labels], valid)
# Plot the progress
print("%d [D loss: %f] [G loss: %f]" % (epoch, d_loss[0], g_loss))
# If at save interval => save generated image samples
if epoch % sample_interval == 0 or epoch == epochs:
self.sample_images(epoch)
def sample_images(self, epoch):
r, c = 2, 5
noise = np.random.normal(0, 1, (r * c, self.latent_dim))
labels = to_categorical(np.arange(0, 10).reshape(-1, 1))
noise = np.random.normal(0,1,(1,100))
labels = np.array([0., 0., 1.0, 0., 0., 0., 0., 0., 0., 0.]).reshape(-1,1)
gen_imgs = self.generator.predict([noise, labels])
# Rescale images 0 - 1
gen_imgs = 0.5 * gen_imgs + 1
fig, axs = plt.subplots(r, c)
cnt = 0
for i in range(r):
for j in range(c):
axs[i, j].imshow(gen_imgs[cnt].reshape(28, 28))
axs[i, j].axis('off')
cnt += 1
fig.savefig("tmp/mnist_%d.png" % epoch)
plt.close()
if __name__ == '__main__':
wgan = WGANGP()
wgan.train(epochs=5000, batch_size=32, sample_interval=100)