DCGAN生成漫画头像

109702008

于 2024-10-03 16:31:57 发布

阅读量634

点赞数 10

分类专栏：人工智能文章标签：人工智能学习

本文链接：https://blog.csdn.net/eidolon_foot/article/details/142690371

版权

人工智能专栏收录该内容

254 篇文章 3 订阅

订阅专栏

tutorials/application/source_zh_cn/generative/dcgan.ipynb · MindSpore/docs - Gitee.com

在下面的教程中，我们将通过示例代码说明DCGAN网络如何设置网络、优化器、如何计算损失函数以及如何初始化模型权重。在本教程中，使用的动漫头像数据集共有70,171张动漫头像图片，图片大小均为96*96。

GAN基础原理

这部分原理介绍参考GAN图像生成。

DCGAN原理

DCGAN（深度卷积对抗生成网络，Deep Convolutional Generative Adversarial Networks）是GAN的直接扩展。不同之处在于，DCGAN会分别在判别器和生成器中使用卷积和转置卷积层。

它最早由Radford等人在论文Unsupervised Representation Learning With Deep Convolutional Generative Adversarial Networks中进行描述。判别器由分层的卷积层、BatchNorm层和LeakyReLU激活层组成。输入是3x64x64的图像，输出是该图像为真图像的概率。生成器则是由转置卷积层、BatchNorm层和ReLU激活层组成。输入是标准正态分布中提取出的隐向量𝑧z，输出是3x64x64的RGB图像。

本教程将使用动漫头像数据集来训练一个生成式对抗网络，接着使用该网络生成动漫头像图片。

数据准备与处理

首先我们将数据集下载到指定目录下并解压。示例代码如下：

%%capture captured_output
# 实验环境已经预装了mindspore==2.3.0，如需更换mindspore版本，可更改下面 MINDSPORE_VERSION 变量
!pip uninstall mindspore -y
%env MINDSPORE_VERSION=2.3.0
!pip install https://ms-release.obs.cn-north-4.myhuaweicloud.com/${MINDSPORE_VERSION}/MindSpore/unified/aarch64/mindspore-${MINDSPORE_VERSION}-cp39-cp39-linux_aarch64.whl --trusted-host ms-release.obs.cn-north-4.myhuaweicloud.com -i https://pypi.mirrors.ustc.edu.cn/simple

# 查看当前 mindspore 版本
!pip show mindspore

Name: mindspore
Version: 2.3.0
Summary: MindSpore is a new open source deep learning training/inference framework that could be used for mobile, edge and cloud scenarios.
Home-page: https://www.mindspore.cn
Author: The MindSpore Authors
Author-email: contact@mindspore.cn
License: Apache 2.0
Location: /home/mindspore/miniconda/envs/jupyter/lib/python3.9/site-packages
Requires: asttokens, astunparse, numpy, packaging, pillow, protobuf, psutil, scipy
Required-by:

from download import download

url = "https://download.mindspore.cn/dataset/Faces/faces.zip"

path = download(url, "./faces", kind="zip", replace=True)

Creating data folder...
Downloading data from https://download-mindspore.osinfra.cn/dataset/Faces/faces.zip (274.6 MB)

file_sizes: 100%|█████████████████████████████| 288M/288M [00:01<00:00, 207MB/s]
Extracting zip file...
Successfully downloaded / unzipped to ./faces

下载后的数据集目录结构如下：

./faces/faces
├── 0.jpg
├── 1.jpg
├── 2.jpg
├── 3.jpg
├── 4.jpg
    ...
├── 70169.jpg
└── 70170.jpg

数据处理

首先为执行过程定义一些输入：

batch_size = 128          # 批量大小
image_size = 64           # 训练图像空间大小
nc = 3                    # 图像彩色通道数
nz = 100                  # 隐向量的长度
ngf = 64                  # 特征图在生成器中的大小
ndf = 64                  # 特征图在判别器中的大小
num_epochs = 100           # 训练周期数
lr = 0.0002               # 学习率
beta1 = 0.5               # Adam优化器的beta1超参数

定义create_dataset_imagenet函数对数据进行处理和增强操作。

import numpy as np
import mindspore.dataset as ds
import mindspore.dataset.vision as vision

def create_dataset_imagenet(dataset_path):
    """数据加载"""
    dataset = ds.ImageFolderDataset(dataset_path,
                                    num_parallel_workers=4,
                                    shuffle=True,
                                    decode=True)

    # 数据增强操作
    transforms = [
        vision.Resize(image_size),
        vision.CenterCrop(image_size),
        vision.HWC2CHW(),
        lambda x: ((x / 255).astype("float32"))
    ]

    # 数据映射操作
    dataset = dataset.project('image')
    dataset = dataset.map(transforms, 'image')

    # 批量操作
    dataset = dataset.batch(batch_size)
    return dataset

dataset = create_dataset_imagenet('./faces')

通过create_dict_iterator函数将数据转换成字典迭代器，然后使用matplotlib模块可视化部分训练数据。

import matplotlib.pyplot as plt

def plot_data(data):
    # 可视化部分训练数据
    plt.figure(figsize=(10, 3), dpi=140)
    for i, image in enumerate(data[0][:30], 1):
        plt.subplot(3, 10, i)
        plt.axis("off")
        plt.imshow(image.transpose(1, 2, 0))
    plt.show()

sample_data = next(dataset.create_tuple_iterator(output_numpy=True))
plot_data(sample_data)

构造网络

当处理完数据后，就可以来进行网络的搭建了。按照DCGAN论文中的描述，所有模型权重均应从mean为0，sigma为0.02的正态分布中随机初始化。

生成器

生成器G的功能是将隐向量z映射到数据空间。由于数据是图像，这一过程也会创建与真实图像大小相同的 RGB 图像。在实践场景中，该功能是通过一系列Conv2dTranspose转置卷积层来完成的，每个层都与BatchNorm2d层和ReLu激活层配对，输出数据会经过tanh函数，使其返回[-1,1]的数据范围内。

DCGAN论文生成图像如下所示：

dcgangenerator

图片来源：Unsupervised Representation Learning With Deep Convolutional Generative Adversarial Networks.

我们通过输入部分中设置的nz、ngf和nc来影响代码中的生成器结构。nz是隐向量z的长度，ngf与通过生成器传播的特征图的大小有关，nc是输出图像中的通道数。

以下是生成器的代码实现：

import mindspore as ms
from mindspore import nn, ops
from mindspore.common.initializer import Normal

weight_init = Normal(mean=0, sigma=0.02)
gamma_init = Normal(mean=1, sigma=0.02)

class Generator(nn.Cell):
    """DCGAN网络生成器"""

    def __init__(self):
        super(Generator, self).__init__()
        self.generator = nn.SequentialCell(
            nn.Conv2dTranspose(nz, ngf * 8, 4, 1, 'valid', weight_init=weight_init),
            nn.BatchNorm2d(ngf * 8, gamma_init=gamma_init),
            nn.ReLU(),
            nn.Conv2dTranspose(ngf * 8, ngf * 4, 4, 2, 'pad', 1, weight_init=weight_init),
            nn.BatchNorm2d(ngf * 4, gamma_init=gamma_init),
            nn.ReLU(),
            nn.Conv2dTranspose(ngf * 4, ngf * 2, 4, 2, 'pad', 1, weight_init=weight_init),
            nn.BatchNorm2d(ngf * 2, gamma_init=gamma_init),
            nn.ReLU(),
            nn.Conv2dTranspose(ngf * 2, ngf, 4, 2, 'pad', 1, weight_init=weight_init),
            nn.BatchNorm2d(ngf, gamma_init=gamma_init),
            nn.ReLU(),
            nn.Conv2dTranspose(ngf, nc, 4, 2, 'pad', 1, weight_init=weight_init),
            nn.Tanh()
            )

    def construct(self, x):
        return self.generator(x)

generator = Generator()

判别器

如前所述，判别器D是一个二分类网络模型，输出判定该图像为真实图的概率。通过一系列的Conv2d、BatchNorm2d和LeakyReLU层对其进行处理，最后通过Sigmoid激活函数得到最终概率。

DCGAN论文提到，使用卷积而不是通过池化来进行下采样是一个好方法，因为它可以让网络学习自己的池化特征。

判别器的代码实现如下：

class Discriminator(nn.Cell):
    """DCGAN网络判别器"""

    def __init__(self):
        super(Discriminator, self).__init__()
        self.discriminator = nn.SequentialCell(
            nn.Conv2d(nc, ndf, 4, 2, 'pad', 1, weight_init=weight_init),
            nn.LeakyReLU(0.2),
            nn.Conv2d(ndf, ndf * 2, 4, 2, 'pad', 1, weight_init=weight_init),
            nn.BatchNorm2d(ngf * 2, gamma_init=gamma_init),
            nn.LeakyReLU(0.2),
            nn.Conv2d(ndf * 2, ndf * 4, 4, 2, 'pad', 1, weight_init=weight_init),
            nn.BatchNorm2d(ngf * 4, gamma_init=gamma_init),
            nn.LeakyReLU(0.2),
            nn.Conv2d(ndf * 4, ndf * 8, 4, 2, 'pad', 1, weight_init=weight_init),
            nn.BatchNorm2d(ngf * 8, gamma_init=gamma_init),
            nn.LeakyReLU(0.2),
            nn.Conv2d(ndf * 8, 1, 4, 1, 'valid', weight_init=weight_init),
            )
        self.adv_layer = nn.Sigmoid()

    def construct(self, x):
        out = self.discriminator(x)
        out = out.reshape(out.shape[0], -1)
        return self.adv_layer(out)

discriminator = Discriminator()

模型训练

损失函数

当定义了D和G后，接下来将使用MindSpore中定义的二进制交叉熵损失函数BCELoss。

# 定义损失函数
adversarial_loss = nn.BCELoss(reduction='mean')

优化器

这里设置了两个单独的优化器，一个用于D，另一个用于G。这两个都是lr = 0.0002和beta1 = 0.5的Adam优化器。

# 为生成器和判别器设置优化器
optimizer_D = nn.Adam(discriminator.trainable_params(), learning_rate=lr, beta1=beta1)
optimizer_G = nn.Adam(generator.trainable_params(), learning_rate=lr, beta1=beta1)
optimizer_G.update_parameters_name('optim_g.')
optimizer_D.update_parameters_name('optim_d.')

训练模型

训练分为两个主要部分：训练判别器和训练生成器。

训练判别器

训练判别器的目的是最大程度地提高判别图像真伪的概率。按照Goodfellow的方法，是希望通过提高其随机梯度来更新判别器，所以我们要最大化𝑙𝑜𝑔𝐷(𝑥)+𝑙𝑜𝑔(1−𝐷(𝐺(𝑧))logD(x)+log(1−D(G(z))的值。
训练生成器

如DCGAN论文所述，我们希望通过最小化𝑙𝑜𝑔(1−𝐷(𝐺(𝑧)))log(1−D(G(z)))来训练生成器，以产生更好的虚假图像。

在这两个部分中，分别获取训练过程中的损失，并在每个周期结束时进行统计，将fixed_noise批量推送到生成器中，以直观地跟踪G的训练进度。

下面实现模型训练正向逻辑：

def generator_forward(real_imgs, valid):
    # 将噪声采样为发生器的输入
    z = ops.standard_normal((real_imgs.shape[0], nz, 1, 1))

    # 生成一批图像
    gen_imgs = generator(z)

    # 损失衡量发生器绕过判别器的能力
    g_loss = adversarial_loss(discriminator(gen_imgs), valid)

    return g_loss, gen_imgs

def discriminator_forward(real_imgs, gen_imgs, valid, fake):
    # 衡量鉴别器从生成的样本中对真实样本进行分类的能力
    real_loss = adversarial_loss(discriminator(real_imgs), valid)
    fake_loss = adversarial_loss(discriminator(gen_imgs), fake)
    d_loss = (real_loss + fake_loss) / 2
    return d_loss

grad_generator_fn = ms.value_and_grad(generator_forward, None,
                                      optimizer_G.parameters,
                                      has_aux=True)
grad_discriminator_fn = ms.value_and_grad(discriminator_forward, None,
                                          optimizer_D.parameters)

@ms.jit
def train_step(imgs):
    valid = ops.ones((imgs.shape[0], 1), mindspore.float32)
    fake = ops.zeros((imgs.shape[0], 1), mindspore.float32)

    (g_loss, gen_imgs), g_grads = grad_generator_fn(imgs, valid)
    optimizer_G(g_grads)
    d_loss, d_grads = grad_discriminator_fn(imgs, gen_imgs, valid, fake)
    optimizer_D(d_grads)

    return g_loss, d_loss, gen_imgs

循环训练网络，每经过50次迭代，就收集生成器和判别器的损失，以便于后面绘制训练过程中损失函数的图像。

%%time
import mindspore

G_losses = []
D_losses = []
image_list = []

total = dataset.get_dataset_size()
for epoch in range(num_epochs):
    generator.set_train()
    discriminator.set_train()
    # 为每轮训练读入数据
    for i, (imgs, ) in enumerate(dataset.create_tuple_iterator()):
        g_loss, d_loss, gen_imgs = train_step(imgs)
        if i % 100 == 0 or i == total - 1:
            # 输出训练记录
            print('[%2d/%d][%3d/%d]   Loss_D:%7.4f  Loss_G:%7.4f' % (
                epoch + 1, num_epochs, i + 1, total, d_loss.asnumpy(), g_loss.asnumpy()))
        D_losses.append(d_loss.asnumpy())
        G_losses.append(g_loss.asnumpy())

    # 每个epoch结束后，使用生成器生成一组图片
    generator.set_train(False)
    fixed_noise = ops.standard_normal((batch_size, nz, 1, 1))
    img = generator(fixed_noise)
    image_list.append(img.transpose(0, 2, 3, 1).asnumpy())

    # 保存网络模型参数为ckpt文件
    mindspore.save_checkpoint(generator, "./generator.ckpt")
    mindspore.save_checkpoint(discriminator, "./discriminator.ckpt")

为了减少打印，改为每轮打印一次：

%%time
import mindspore

G_losses = []
D_losses = []
image_list = []

total = dataset.get_dataset_size()
for epoch in range(num_epochs):
    generator.set_train()
    discriminator.set_train()
    # 为每轮训练读入数据
    for i, (imgs, ) in enumerate(dataset.create_tuple_iterator()):
        g_loss, d_loss, gen_imgs = train_step(imgs)
        if i == total - 1:
            # 输出训练记录
            print('[%2d/%d][%3d/%d]   Loss_D:%7.4f  Loss_G:%7.4f' % (
                epoch + 1, num_epochs, i + 1, total, d_loss.asnumpy(), g_loss.asnumpy()))
        D_losses.append(d_loss.asnumpy())
        G_losses.append(g_loss.asnumpy())

    # 每个epoch结束后，使用生成器生成一组图片
    generator.set_train(False)
    fixed_noise = ops.standard_normal((batch_size, nz, 1, 1))
    img = generator(fixed_noise)
    image_list.append(img.transpose(0, 2, 3, 1).asnumpy())

    # 保存网络模型参数为ckpt文件
    mindspore.save_checkpoint(generator, "./generator.ckpt")
    mindspore.save_checkpoint(discriminator, "./discriminator.ckpt")

[ 1/100][549/549]   Loss_D: 0.6106  Loss_G: 0.6764
[ 2/100][549/549]   Loss_D: 0.3092  Loss_G: 4.2549
[ 3/100][549/549]   Loss_D: 0.3885  Loss_G: 1.6692
[ 4/100][549/549]   Loss_D: 0.1445  Loss_G: 2.1410
[ 5/100][549/549]   Loss_D: 0.2237  Loss_G: 2.4766
[ 6/100][549/549]   Loss_D: 0.1699  Loss_G: 2.8469
[ 7/100][549/549]   Loss_D: 0.4511  Loss_G: 8.5175
[ 8/100][549/549]   Loss_D: 0.0959  Loss_G: 2.6130
[ 9/100][549/549]   Loss_D: 0.1494  Loss_G: 2.6417
[10/100][549/549]   Loss_D: 0.0880  Loss_G: 2.5666
[11/100][549/549]   Loss_D: 0.0870  Loss_G: 2.1918
[12/100][549/549]   Loss_D: 0.2240  Loss_G: 3.6016
[13/100][549/549]   Loss_D: 0.1102  Loss_G: 2.8687
[14/100][549/549]   Loss_D: 0.1859  Loss_G: 2.7287
[15/100][549/549]   Loss_D: 0.1399  Loss_G: 1.9229
[16/100][549/549]   Loss_D: 0.1043  Loss_G: 2.7204
[17/100][549/549]   Loss_D: 0.2446  Loss_G: 1.7036
[18/100][549/549]   Loss_D: 0.1583  Loss_G: 2.6729
[19/100][549/549]   Loss_D: 0.2957  Loss_G: 6.4399
[20/100][549/549]   Loss_D: 0.1509  Loss_G: 3.9361
[21/100][549/549]   Loss_D: 0.0795  Loss_G: 3.8061
[22/100][549/549]   Loss_D: 1.5071  Loss_G: 9.9033
[23/100][549/549]   Loss_D: 0.0741  Loss_G: 2.7125
[24/100][549/549]   Loss_D: 0.0451  Loss_G: 3.4519
[25/100][549/549]   Loss_D: 0.1035  Loss_G: 3.0197
[26/100][549/549]   Loss_D: 0.1196  Loss_G: 2.3379
[27/100][549/549]   Loss_D: 0.1113  Loss_G: 2.3422
[28/100][549/549]   Loss_D: 0.1827  Loss_G: 2.5274
[29/100][549/549]   Loss_D: 0.0607  Loss_G: 2.9610
[30/100][549/549]   Loss_D: 1.6338  Loss_G: 8.6766
[31/100][549/549]   Loss_D: 0.0839  Loss_G: 3.9443
[32/100][549/549]   Loss_D: 0.1627  Loss_G: 2.3885
[33/100][549/549]   Loss_D: 0.1081  Loss_G: 3.0734
[34/100][549/549]   Loss_D: 0.3051  Loss_G: 1.8697
[35/100][549/549]   Loss_D: 0.1608  Loss_G: 5.1093
[36/100][549/549]   Loss_D: 1.5805  Loss_G: 8.1907
[37/100][549/549]   Loss_D: 0.1765  Loss_G: 2.0164
[38/100][549/549]   Loss_D: 0.0220  Loss_G: 4.0469
[39/100][549/549]   Loss_D: 0.1006  Loss_G: 2.4763
[40/100][549/549]   Loss_D: 0.0794  Loss_G: 2.4766
[41/100][549/549]   Loss_D: 0.0986  Loss_G: 4.9125
[42/100][549/549]   Loss_D: 0.0607  Loss_G: 3.5990
[43/100][549/549]   Loss_D: 0.0304  Loss_G: 3.5067
[44/100][549/549]   Loss_D: 0.0579  Loss_G: 5.1174
[45/100][549/549]   Loss_D: 0.2503  Loss_G: 3.6695
[46/100][549/549]   Loss_D: 0.0344  Loss_G: 4.8349
[47/100][549/549]   Loss_D: 0.1953  Loss_G: 2.7192
[48/100][549/549]   Loss_D: 0.0469  Loss_G: 5.5249
[49/100][549/549]   Loss_D: 0.0960  Loss_G: 3.0558
[50/100][549/549]   Loss_D: 0.0581  Loss_G: 4.1038
[51/100][549/549]   Loss_D: 0.0299  Loss_G: 3.8405
[52/100][549/549]   Loss_D: 0.0990  Loss_G: 2.7754
[53/100][549/549]   Loss_D: 0.0939  Loss_G: 3.8828
[54/100][549/549]   Loss_D: 0.0618  Loss_G: 3.1732
[55/100][549/549]   Loss_D: 0.3889  Loss_G: 1.2231
[56/100][549/549]   Loss_D: 0.0990  Loss_G: 4.5803
[57/100][549/549]   Loss_D: 0.0286  Loss_G: 5.2037
[58/100][549/549]   Loss_D: 0.0617  Loss_G: 3.4745
[59/100][549/549]   Loss_D: 0.2001  Loss_G: 2.2047
[60/100][549/549]   Loss_D: 0.0353  Loss_G: 4.7155
[61/100][549/549]   Loss_D: 0.0253  Loss_G: 4.6451
[62/100][549/549]   Loss_D: 0.3381  Loss_G: 1.2426
[63/100][549/549]   Loss_D: 1.6174  Loss_G: 6.0268
[64/100][549/549]   Loss_D: 0.0446  Loss_G: 3.1168
[65/100][549/549]   Loss_D: 0.0776  Loss_G: 3.7572
[66/100][549/549]   Loss_D: 0.0178  Loss_G: 5.3017
[67/100][549/549]   Loss_D: 0.0550  Loss_G: 5.0807
[68/100][549/549]   Loss_D: 0.0359  Loss_G: 3.9622
[69/100][549/549]   Loss_D: 0.0046  Loss_G: 6.1130
[70/100][549/549]   Loss_D: 0.1212  Loss_G: 2.8980
[71/100][549/549]   Loss_D: 0.1356  Loss_G: 3.5367
[72/100][549/549]   Loss_D: 0.0879  Loss_G: 4.8944
[73/100][549/549]   Loss_D: 0.0243  Loss_G: 4.2409
[74/100][549/549]   Loss_D: 0.0418  Loss_G: 4.0617
[75/100][549/549]   Loss_D: 0.1328  Loss_G: 5.4291
[76/100][549/549]   Loss_D: 0.0150  Loss_G: 5.0857
[77/100][549/549]   Loss_D: 0.5582  Loss_G: 0.9203
[78/100][549/549]   Loss_D: 0.0579  Loss_G: 5.5855
[79/100][549/549]   Loss_D: 0.3879  Loss_G: 3.7747
[80/100][549/549]   Loss_D: 0.1436  Loss_G: 2.8071
[81/100][549/549]   Loss_D: 0.3205  Loss_G: 7.8155
[82/100][549/549]   Loss_D: 0.0180  Loss_G: 4.8216
[83/100][549/549]   Loss_D: 0.4509  Loss_G: 1.6024
[84/100][549/549]   Loss_D: 0.0254  Loss_G: 5.2787
[85/100][549/549]   Loss_D: 0.0109  Loss_G: 5.8957
[86/100][549/549]   Loss_D: 0.2196  Loss_G: 1.6766
[87/100][549/549]   Loss_D: 0.0948  Loss_G: 2.7880
[88/100][549/549]   Loss_D: 0.1002  Loss_G: 3.3182
[89/100][549/549]   Loss_D: 0.0323  Loss_G: 5.7229
[90/100][549/549]   Loss_D: 0.0565  Loss_G: 4.1676
[91/100][549/549]   Loss_D: 0.0265  Loss_G: 4.7548
[92/100][549/549]   Loss_D: 0.0741  Loss_G: 6.0792
[93/100][549/549]   Loss_D: 0.0253  Loss_G: 5.2812
[94/100][549/549]   Loss_D: 0.0388  Loss_G: 4.5826
[95/100][549/549]   Loss_D: 0.0229  Loss_G: 4.4684
[96/100][549/549]   Loss_D: 0.3439  Loss_G: 5.0983
[97/100][549/549]   Loss_D: 0.2110  Loss_G: 6.5986
[98/100][549/549]   Loss_D: 0.0588  Loss_G: 3.2943
[99/100][549/549]   Loss_D: 0.0432  Loss_G: 3.6790
[100/100][549/549]   Loss_D: 0.2591  Loss_G: 1.5543
CPU times: user 3h 10min 44s, sys: 1h 4s, total: 4h 10min 49s
Wall time: 25min 21s

结果展示

运行下面代码，描绘D和G损失与训练迭代的关系图：

plt.figure(figsize=(10, 5))
plt.title("Generator and Discriminator Loss During Training")
plt.plot(G_losses, label="G", color='blue')
plt.plot(D_losses, label="D", color='orange')
plt.xlabel("iterations")
plt.ylabel("Loss")
plt.legend()
plt.show()

可视化训练过程中通过隐向量fixed_noise生成的图像。

import matplotlib.pyplot as plt
import matplotlib.animation as animation

def showGif(image_list):
    show_list = []
    fig = plt.figure(figsize=(8, 3), dpi=120)
    for epoch in range(len(image_list)):
        images = []
        for i in range(3):
            row = np.concatenate((image_list[epoch][i * 8:(i + 1) * 8]), axis=1)
            images.append(row)
        img = np.clip(np.concatenate((images[:]), axis=0), 0, 1)
        plt.axis("off")
        show_list.append([plt.imshow(img)])

    ani = animation.ArtistAnimation(fig, show_list, interval=1000, repeat_delay=1000, blit=True)
    ani.save('./dcgan.gif', writer='pillow', fps=1)

showGif(image_list)

dcgan

从上面的图像可以看出，随着训练次数的增多，图像质量也越来越好。如果增大训练周期数，当num_epochs达到50以上时，生成的动漫头像图片与数据集中的较为相似，下面我们通过加载生成器网络模型参数文件来生成图像，代码如下：

# 从文件中获取模型参数并加载到网络中
mindspore.load_checkpoint("./generator.ckpt", generator)

fixed_noise = ops.standard_normal((batch_size, nz, 1, 1))
img64 = generator(fixed_noise).transpose(0, 2, 3, 1).asnumpy()

fig = plt.figure(figsize=(8, 3), dpi=120)
images = []
for i in range(3):
    images.append(np.concatenate((img64[i * 8:(i + 1) * 8]), axis=1))
img = np.clip(np.concatenate((images[:]), axis=0), 0, 1)
plt.axis("off")
plt.imshow(img)
plt.show()

当num_epochs = 200时，好像出现了模式崩溃（生成器开始生成非常相似或重复的样本）。

batch_size = 128          # 批量大小
image_size = 64           # 训练图像空间大小
nc = 3                    # 图像彩色通道数
nz = 100                  # 隐向量的长度
ngf = 64                  # 特征图在生成器中的大小
ndf = 64                  # 特征图在判别器中的大小
num_epochs = 200           # 训练周期数
lr = 0.0002               # 学习率
beta1 = 0.5               # Adam优化器的beta1超参数

%%time
import mindspore

G_losses = []
D_losses = []
image_list = []

total = dataset.get_dataset_size()
for epoch in range(num_epochs):
    generator.set_train()
    discriminator.set_train()
    # 为每轮训练读入数据
    for i, (imgs, ) in enumerate(dataset.create_tuple_iterator()):
        g_loss, d_loss, gen_imgs = train_step(imgs)
        if i % 500 == 0 or i == total - 1:
            # 输出训练记录
            print('[%2d/%d][%3d/%d]   Loss_D:%7.4f  Loss_G:%7.4f' % (
                epoch + 1, num_epochs, i + 1, total, d_loss.asnumpy(), g_loss.asnumpy()))
        D_losses.append(d_loss.asnumpy())
        G_losses.append(g_loss.asnumpy())

    # 每个epoch结束后，使用生成器生成一组图片
    generator.set_train(False)
    fixed_noise = ops.standard_normal((batch_size, nz, 1, 1))
    img = generator(fixed_noise)
    image_list.append(img.transpose(0, 2, 3, 1).asnumpy())

    # 保存网络模型参数为ckpt文件
    mindspore.save_checkpoint(generator, "./generator.ckpt")
    mindspore.save_checkpoint(discriminator, "./discriminator.ckpt")

[ 1/200][  1/549]   Loss_D: 0.8324  Loss_G: 1.2639
[ 1/200][501/549]   Loss_D: 0.2243  Loss_G: 1.4972

[ 1/200][549/549]   Loss_D: 0.1225  Loss_G: 3.7119
[ 2/200][  1/549]   Loss_D: 0.2634  Loss_G: 1.7945
[ 2/200][501/549]   Loss_D: 0.3243  Loss_G: 5.4640
[ 2/200][549/549]   Loss_D: 0.2435  Loss_G: 2.2054
[ 3/200][  1/549]   Loss_D: 0.2918  Loss_G: 1.3228
[ 3/200][501/549]   Loss_D: 0.2894  Loss_G: 3.4179
[ 3/200][549/549]   Loss_D: 0.6185  Loss_G: 0.5866
[ 4/200][  1/549]   Loss_D: 1.2035  Loss_G:10.6546
[ 4/200][501/549]   Loss_D: 0.3380  Loss_G: 3.2789
[ 4/200][549/549]   Loss_D: 0.3168  Loss_G: 3.2505
[ 5/200][  1/549]   Loss_D: 0.3357  Loss_G: 1.2956
[ 5/200][501/549]   Loss_D: 0.2061  Loss_G: 4.2091
[ 5/200][549/549]   Loss_D: 0.1292  Loss_G: 2.4829
[ 6/200][  1/549]   Loss_D: 0.1827  Loss_G: 2.1405
[ 6/200][501/549]   Loss_D: 0.2145  Loss_G: 6.3399
[ 6/200][549/549]   Loss_D: 0.1928  Loss_G: 1.6432
[ 7/200][  1/549]   Loss_D: 0.0830  Loss_G: 3.3874
[ 7/200][501/549]   Loss_D: 0.4541  Loss_G: 0.7287
[ 7/200][549/549]   Loss_D: 0.1416  Loss_G: 2.0121
[ 8/200][  1/549]   Loss_D: 0.1891  Loss_G: 2.9087
[ 8/200][501/549]   Loss_D: 0.1970  Loss_G: 4.0175
[ 8/200][549/549]   Loss_D: 0.1693  Loss_G: 2.0178
[ 9/200][  1/549]   Loss_D: 0.0937  Loss_G: 3.4365
[ 9/200][501/549]   Loss_D: 0.1757  Loss_G: 2.1768
[ 9/200][549/549]   Loss_D: 0.1696  Loss_G: 1.9579
[10/200][  1/549]   Loss_D: 0.2211  Loss_G: 2.2690
[10/200][501/549]   Loss_D: 0.1164  Loss_G: 2.1523
[10/200][549/549]   Loss_D: 0.2187  Loss_G: 2.0313
[11/200][  1/549]   Loss_D: 0.1367  Loss_G: 2.5421
[11/200][501/549]   Loss_D: 0.1174  Loss_G: 2.4074
[11/200][549/549]   Loss_D: 1.4441  Loss_G: 0.1132
[12/200][  1/549]   Loss_D: 0.8690  Loss_G: 7.9187
[12/200][501/549]   Loss_D: 0.1522  Loss_G: 2.2471
[12/200][549/549]   Loss_D: 0.1498  Loss_G: 2.2950
[13/200][  1/549]   Loss_D: 0.2269  Loss_G: 2.5328
[13/200][501/549]   Loss_D: 0.1723  Loss_G: 2.6353
[13/200][549/549]   Loss_D: 0.1196  Loss_G: 2.7592
[14/200][  1/549]   Loss_D: 0.1173  Loss_G: 2.7692
[14/200][501/549]   Loss_D: 0.1994  Loss_G: 1.8361
[14/200][549/549]   Loss_D: 0.2566  Loss_G: 1.4707
[15/200][  1/549]   Loss_D: 0.1332  Loss_G: 2.0971
[15/200][501/549]   Loss_D: 0.2121  Loss_G: 1.6003
[15/200][549/549]   Loss_D: 2.2433  Loss_G: 9.1780
[16/200][  1/549]   Loss_D: 0.1690  Loss_G: 2.0324
[16/200][501/549]   Loss_D: 0.2154  Loss_G: 3.4251
[16/200][549/549]   Loss_D: 0.6030  Loss_G: 6.0030
[17/200][  1/549]   Loss_D: 0.2687  Loss_G: 1.1912
[17/200][501/549]   Loss_D: 0.1808  Loss_G: 1.7610
[17/200][549/549]   Loss_D: 0.3762  Loss_G: 0.8627
[18/200][  1/549]   Loss_D: 0.4494  Loss_G: 5.9094
[18/200][501/549]   Loss_D: 0.4202  Loss_G: 0.9349
[18/200][549/549]   Loss_D: 0.2420  Loss_G: 1.4175
[19/200][  1/549]   Loss_D: 0.1227  Loss_G: 2.6680
[19/200][501/549]   Loss_D: 0.2651  Loss_G: 1.1539
[19/200][549/549]   Loss_D: 0.2152  Loss_G: 4.3104
[20/200][  1/549]   Loss_D: 0.3126  Loss_G: 1.0723
[20/200][501/549]   Loss_D: 0.1923  Loss_G: 1.8047
[20/200][549/549]   Loss_D: 0.1228  Loss_G: 2.9775
[21/200][  1/549]   Loss_D: 0.1811  Loss_G: 1.9004
[21/200][501/549]   Loss_D: 0.0985  Loss_G: 2.7778
[21/200][549/549]   Loss_D: 0.7840  Loss_G: 6.9031
[22/200][  1/549]   Loss_D: 0.1365  Loss_G: 2.0484
[22/200][501/549]   Loss_D: 0.2438  Loss_G: 2.9704
[22/200][549/549]   Loss_D: 0.5911  Loss_G: 5.4517
[23/200][  1/549]   Loss_D: 0.9774  Loss_G: 0.2537
[23/200][501/549]   Loss_D: 0.3037  Loss_G: 1.1571
[23/200][549/549]   Loss_D: 0.1643  Loss_G: 3.9519
[24/200][  1/549]   Loss_D: 0.2573  Loss_G: 1.2705
[24/200][501/549]   Loss_D: 0.1461  Loss_G: 4.0229
[24/200][549/549]   Loss_D: 0.1226  Loss_G: 2.3870
[25/200][  1/549]   Loss_D: 0.1045  Loss_G: 3.9232
[25/200][501/549]   Loss_D: 0.1356  Loss_G: 2.3146
[25/200][549/549]   Loss_D: 0.2543  Loss_G: 3.6102
[26/200][  1/549]   Loss_D: 0.2078  Loss_G: 1.4601
[26/200][501/549]   Loss_D: 0.1547  Loss_G: 2.7833
[26/200][549/549]   Loss_D: 0.1853  Loss_G: 1.9727
[27/200][  1/549]   Loss_D: 0.1648  Loss_G: 2.2708
[27/200][501/549]   Loss_D: 0.1452  Loss_G: 2.4902
[27/200][549/549]   Loss_D: 0.2875  Loss_G: 3.2696
[28/200][  1/549]   Loss_D: 0.4400  Loss_G: 0.7629
[28/200][501/549]   Loss_D: 0.0707  Loss_G: 3.8629
[28/200][549/549]   Loss_D: 0.1430  Loss_G: 1.8769
[29/200][  1/549]   Loss_D: 0.7581  Loss_G: 5.9448
[29/200][501/549]   Loss_D: 0.2469  Loss_G: 5.8975
[29/200][549/549]   Loss_D: 0.0966  Loss_G: 3.0256
[30/200][  1/549]   Loss_D: 0.0998  Loss_G: 2.3803
[30/200][501/549]   Loss_D: 0.1621  Loss_G: 4.5309
[30/200][549/549]   Loss_D: 0.1170  Loss_G: 2.2813
[31/200][  1/549]   Loss_D: 0.2743  Loss_G: 1.2051
[31/200][501/549]   Loss_D: 0.1116  Loss_G: 2.5379
[31/200][549/549]   Loss_D: 0.2577  Loss_G: 2.2138
[32/200][  1/549]   Loss_D: 0.1581  Loss_G: 2.1937
[32/200][501/549]   Loss_D: 1.1011  Loss_G: 0.2635
[32/200][549/549]   Loss_D: 0.1970  Loss_G: 1.7803
[33/200][  1/549]   Loss_D: 0.4104  Loss_G: 4.4137
[33/200][501/549]   Loss_D: 0.1607  Loss_G: 1.8982
[33/200][549/549]   Loss_D: 0.0772  Loss_G: 3.9291
[34/200][  1/549]   Loss_D: 0.1640  Loss_G: 1.7486
[34/200][501/549]   Loss_D: 0.1159  Loss_G: 2.3928
[34/200][549/549]   Loss_D: 0.1051  Loss_G: 3.2938
[35/200][  1/549]   Loss_D: 0.2062  Loss_G: 1.5448
[35/200][501/549]   Loss_D: 0.0764  Loss_G: 3.4685
[35/200][549/549]   Loss_D: 0.0783  Loss_G: 4.0028
[36/200][  1/549]   Loss_D: 0.0691  Loss_G: 3.1478
[36/200][501/549]   Loss_D: 0.1945  Loss_G: 1.8608
[36/200][549/549]   Loss_D: 0.0664  Loss_G: 3.2432
[37/200][  1/549]   Loss_D: 0.1332  Loss_G: 2.8768
[37/200][501/549]   Loss_D: 0.1357  Loss_G: 2.0631
[37/200][549/549]   Loss_D: 0.1194  Loss_G: 3.4386
[38/200][  1/549]   Loss_D: 0.1071  Loss_G: 2.6619
[38/200][501/549]   Loss_D: 0.0323  Loss_G: 3.8000
[38/200][549/549]   Loss_D: 0.2826  Loss_G: 5.0917
[39/200][  1/549]   Loss_D: 0.8349  Loss_G: 0.3916
[39/200][501/549]   Loss_D: 0.0603  Loss_G: 3.2233
[39/200][549/549]   Loss_D: 0.0758  Loss_G: 2.6212
[40/200][  1/549]   Loss_D: 0.0472  Loss_G: 4.0676
[40/200][501/549]   Loss_D: 0.1662  Loss_G: 4.0471
[40/200][549/549]   Loss_D: 0.0901  Loss_G: 3.0110
[41/200][  1/549]   Loss_D: 0.1109  Loss_G: 2.3747
[41/200][501/549]   Loss_D: 0.0864  Loss_G: 3.5582
[41/200][549/549]   Loss_D: 0.1556  Loss_G: 4.0981
[42/200][  1/549]   Loss_D: 0.1805  Loss_G: 1.7918
[42/200][501/549]   Loss_D: 0.1553  Loss_G: 2.5847
[42/200][549/549]   Loss_D: 0.2758  Loss_G: 3.6331
[43/200][  1/549]   Loss_D: 1.6624  Loss_G: 0.1618
[43/200][501/549]   Loss_D: 0.0344  Loss_G: 3.7180
[43/200][549/549]   Loss_D: 0.0191  Loss_G: 4.4786
[44/200][  1/549]   Loss_D: 0.0239  Loss_G: 4.5359
[44/200][501/549]   Loss_D: 0.0648  Loss_G: 3.8651
[44/200][549/549]   Loss_D: 0.1347  Loss_G: 2.8805
[45/200][  1/549]   Loss_D: 0.2039  Loss_G: 1.7305
[45/200][501/549]   Loss_D: 0.0523  Loss_G: 3.4660
[45/200][549/549]   Loss_D: 0.0300  Loss_G: 4.3123
[46/200][  1/549]   Loss_D: 0.0935  Loss_G: 4.7765
[46/200][501/549]   Loss_D: 0.1090  Loss_G: 2.7217
[46/200][549/549]   Loss_D: 0.0705  Loss_G: 4.1516
[47/200][  1/549]   Loss_D: 0.0726  Loss_G: 3.0272
[47/200][501/549]   Loss_D: 0.0537  Loss_G: 3.3243
[47/200][549/549]   Loss_D: 0.0693  Loss_G: 2.8473
[48/200][  1/549]   Loss_D: 0.1066  Loss_G: 2.2994
[48/200][501/549]   Loss_D: 0.0318  Loss_G: 4.9260
[48/200][549/549]   Loss_D: 0.0633  Loss_G: 4.0919
[49/200][  1/549]   Loss_D: 0.1076  Loss_G: 3.9757
[49/200][501/549]   Loss_D: 0.4428  Loss_G: 3.9113
[49/200][549/549]   Loss_D: 0.0768  Loss_G: 3.8784
[50/200][  1/549]   Loss_D: 0.1388  Loss_G: 2.6904
[50/200][501/549]   Loss_D: 0.5701  Loss_G: 7.2282
[50/200][549/549]   Loss_D: 0.1603  Loss_G: 2.0840
[51/200][  1/549]   Loss_D: 0.1041  Loss_G: 2.8508
[51/200][501/549]   Loss_D: 1.2263  Loss_G: 8.3074
[51/200][549/549]   Loss_D: 0.1378  Loss_G: 2.7209
[52/200][  1/549]   Loss_D: 0.1355  Loss_G: 3.6068
[52/200][501/549]   Loss_D: 0.0744  Loss_G: 3.1665
[52/200][549/549]   Loss_D: 0.1439  Loss_G: 3.1384
[53/200][  1/549]   Loss_D: 0.1215  Loss_G: 2.4754
[53/200][501/549]   Loss_D: 0.2279  Loss_G: 2.0314
[53/200][549/549]   Loss_D: 0.0733  Loss_G: 2.7644
[54/200][  1/549]   Loss_D: 0.0536  Loss_G: 3.3702
[54/200][501/549]   Loss_D: 0.0475  Loss_G: 4.4517
[54/200][549/549]   Loss_D: 0.0437  Loss_G: 3.8798
[55/200][  1/549]   Loss_D: 0.0290  Loss_G: 3.6527
[55/200][501/549]   Loss_D: 0.0985  Loss_G: 3.1062
[55/200][549/549]   Loss_D: 0.0468  Loss_G: 3.9260
[56/200][  1/549]   Loss_D: 0.0751  Loss_G: 2.8475
[56/200][501/549]   Loss_D: 0.0685  Loss_G: 3.6570
[56/200][549/549]   Loss_D: 0.0366  Loss_G: 4.6187
[57/200][  1/549]   Loss_D: 0.0788  Loss_G: 2.9598
[57/200][501/549]   Loss_D: 0.1491  Loss_G: 2.2459
[57/200][549/549]   Loss_D: 0.1917  Loss_G: 1.8331
[58/200][  1/549]   Loss_D: 0.1782  Loss_G: 2.2357
[58/200][501/549]   Loss_D: 0.1041  Loss_G: 2.5734
[58/200][549/549]   Loss_D: 0.1353  Loss_G: 3.2996
[59/200][  1/549]   Loss_D: 0.2526  Loss_G: 1.3601
[59/200][501/549]   Loss_D: 0.0873  Loss_G: 2.9913
[59/200][549/549]   Loss_D: 0.0579  Loss_G: 5.3517
[60/200][  1/549]   Loss_D: 0.0932  Loss_G: 2.3693
[60/200][501/549]   Loss_D: 0.1138  Loss_G: 2.6479
[60/200][549/549]   Loss_D: 0.0952  Loss_G: 2.6705
[61/200][  1/549]   Loss_D: 0.1107  Loss_G: 2.7407
[61/200][501/549]   Loss_D: 0.0529  Loss_G: 3.3084
[61/200][549/549]   Loss_D: 0.2721  Loss_G: 4.1151
[62/200][  1/549]   Loss_D: 0.2245  Loss_G: 1.5675
[62/200][501/549]   Loss_D: 0.0878  Loss_G: 2.9496
[62/200][549/549]   Loss_D: 0.4308  Loss_G: 6.5642
[63/200][  1/549]   Loss_D: 2.2300  Loss_G: 0.0525
[63/200][501/549]   Loss_D: 0.1626  Loss_G: 2.4035
[63/200][549/549]   Loss_D: 0.1175  Loss_G: 2.5961
[64/200][  1/549]   Loss_D: 0.1353  Loss_G: 3.4148
[64/200][501/549]   Loss_D: 0.0238  Loss_G: 4.8436
[64/200][549/549]   Loss_D: 0.0207  Loss_G: 4.8270
[65/200][  1/549]   Loss_D: 0.0280  Loss_G: 3.7038
[65/200][501/549]   Loss_D: 0.0916  Loss_G: 3.3575
[65/200][549/549]   Loss_D: 0.0687  Loss_G: 4.0699
[66/200][  1/549]   Loss_D: 0.0923  Loss_G: 3.2198
[66/200][501/549]   Loss_D: 0.0238  Loss_G: 4.5188
[66/200][549/549]   Loss_D: 0.0719  Loss_G: 2.5030
[67/200][  1/549]   Loss_D: 0.1975  Loss_G: 1.6979
[67/200][501/549]   Loss_D: 0.1652  Loss_G: 2.2843
[67/200][549/549]   Loss_D: 1.0853  Loss_G: 8.0768
[68/200][  1/549]   Loss_D: 1.5934  Loss_G: 0.1117
[68/200][501/549]   Loss_D: 0.1611  Loss_G: 2.2162
[68/200][549/549]   Loss_D: 0.0672  Loss_G: 4.4790
[69/200][  1/549]   Loss_D: 0.1048  Loss_G: 2.4682
[69/200][501/549]   Loss_D: 0.1481  Loss_G: 2.5849
[69/200][549/549]   Loss_D: 0.0604  Loss_G: 2.8652
[70/200][  1/549]   Loss_D: 0.0333  Loss_G: 4.2357
[70/200][501/549]   Loss_D: 0.0769  Loss_G: 3.6993
[70/200][549/549]   Loss_D: 0.0793  Loss_G: 4.3074
[71/200][  1/549]   Loss_D: 0.0839  Loss_G: 2.7068
[71/200][501/549]   Loss_D: 0.0468  Loss_G: 3.5372
[71/200][549/549]   Loss_D: 0.1252  Loss_G: 2.2772
[72/200][  1/549]   Loss_D: 0.1966  Loss_G: 3.0858
[72/200][501/549]   Loss_D: 1.5845  Loss_G: 0.0953
[72/200][549/549]   Loss_D: 0.1905  Loss_G: 4.1469
[73/200][  1/549]   Loss_D: 0.2518  Loss_G: 1.4223
[73/200][501/549]   Loss_D: 0.3203  Loss_G: 1.7023
[73/200][549/549]   Loss_D: 0.1659  Loss_G: 2.8936
[74/200][  1/549]   Loss_D: 0.1719  Loss_G: 2.0328
[74/200][501/549]   Loss_D: 0.0169  Loss_G: 4.6490
[74/200][549/549]   Loss_D: 0.0710  Loss_G: 3.2644
[75/200][  1/549]   Loss_D: 0.0653  Loss_G: 3.5217
[75/200][501/549]   Loss_D: 0.0319  Loss_G: 5.4261
[75/200][549/549]   Loss_D: 0.0227  Loss_G: 6.9689
[76/200][  1/549]   Loss_D: 0.0286  Loss_G: 4.0718
[76/200][501/549]   Loss_D: 0.1246  Loss_G: 2.3175
[76/200][549/549]   Loss_D: 0.1094  Loss_G: 2.7421
[77/200][  1/549]   Loss_D: 0.0917  Loss_G: 2.7597
[77/200][501/549]   Loss_D: 0.1620  Loss_G: 5.8381
[77/200][549/549]   Loss_D: 0.9196  Loss_G: 0.3436
[78/200][  1/549]   Loss_D: 0.4835  Loss_G: 5.1983
[78/200][501/549]   Loss_D: 0.1648  Loss_G: 1.8297
[78/200][549/549]   Loss_D: 2.2909  Loss_G: 8.9165
[79/200][  1/549]   Loss_D: 0.2514  Loss_G: 2.0934
[79/200][501/549]   Loss_D: 0.0312  Loss_G: 4.1159
[79/200][549/549]   Loss_D: 0.1320  Loss_G: 3.1260
[80/200][  1/549]   Loss_D: 0.2042  Loss_G: 2.3162
[80/200][501/549]   Loss_D: 0.2032  Loss_G: 2.1090
[80/200][549/549]   Loss_D: 0.0824  Loss_G: 2.9415
[81/200][  1/549]   Loss_D: 0.1062  Loss_G: 3.5188
[81/200][501/549]   Loss_D: 0.3138  Loss_G: 4.9336
[81/200][549/549]   Loss_D: 0.0459  Loss_G: 3.6886
[82/200][  1/549]   Loss_D: 0.0373  Loss_G: 3.7398
[82/200][501/549]   Loss_D: 0.0676  Loss_G: 4.7946
[82/200][549/549]   Loss_D: 0.0519  Loss_G: 3.7365
[83/200][  1/549]   Loss_D: 0.0460  Loss_G: 5.1354
[83/200][501/549]   Loss_D: 0.1640  Loss_G: 2.4931
[83/200][549/549]   Loss_D: 0.0270  Loss_G: 4.5542
[84/200][  1/549]   Loss_D: 0.0279  Loss_G: 3.9441
[84/200][501/549]   Loss_D: 0.0163  Loss_G: 4.4843
[84/200][549/549]   Loss_D: 0.0237  Loss_G: 5.2358
[85/200][  1/549]   Loss_D: 0.0419  Loss_G: 3.4550
[85/200][501/549]   Loss_D: 0.0413  Loss_G: 4.0424
[85/200][549/549]   Loss_D: 0.0626  Loss_G: 5.2893
[86/200][  1/549]   Loss_D: 0.0702  Loss_G: 2.7308
[86/200][501/549]   Loss_D: 0.0741  Loss_G: 2.8321
[86/200][549/549]   Loss_D: 0.0316  Loss_G: 4.7986
[87/200][  1/549]   Loss_D: 0.0360  Loss_G: 3.9111
[87/200][501/549]   Loss_D: 0.0237  Loss_G: 4.3538
[87/200][549/549]   Loss_D: 0.1407  Loss_G: 2.9554
[88/200][  1/549]   Loss_D: 0.0863  Loss_G: 2.8311
[88/200][501/549]   Loss_D: 0.0218  Loss_G: 5.3543
[88/200][549/549]   Loss_D: 0.0587  Loss_G: 6.2412
[89/200][  1/549]   Loss_D: 0.0476  Loss_G: 5.0605
[89/200][501/549]   Loss_D: 0.0050  Loss_G: 6.1558
[89/200][549/549]   Loss_D: 0.0051  Loss_G: 6.1796
[90/200][  1/549]   Loss_D: 0.0025  Loss_G: 6.7527
[90/200][501/549]   Loss_D: 0.0471  Loss_G: 3.5542
[90/200][549/549]   Loss_D: 0.1369  Loss_G: 3.1567
[91/200][  1/549]   Loss_D: 0.1530  Loss_G: 2.9616
[91/200][501/549]   Loss_D: 0.0785  Loss_G: 3.0815
[91/200][549/549]   Loss_D: 0.0555  Loss_G: 3.9266
[92/200][  1/549]   Loss_D: 0.0650  Loss_G: 3.2796
[92/200][501/549]   Loss_D: 0.0272  Loss_G: 4.1397
[92/200][549/549]   Loss_D: 0.1264  Loss_G: 2.5896
[93/200][  1/549]   Loss_D: 0.0603  Loss_G: 3.6887
[93/200][501/549]   Loss_D: 0.1040  Loss_G: 5.9669
[93/200][549/549]   Loss_D: 0.1973  Loss_G: 3.0978
[94/200][  1/549]   Loss_D: 0.2311  Loss_G: 1.7136
[94/200][501/549]   Loss_D: 0.0374  Loss_G: 4.1025
[94/200][549/549]   Loss_D: 0.0258  Loss_G: 4.1492
[95/200][  1/549]   Loss_D: 0.0140  Loss_G: 4.7239
[95/200][501/549]   Loss_D: 0.0559  Loss_G: 4.3814
[95/200][549/549]   Loss_D: 0.0883  Loss_G: 5.0761
[96/200][  1/549]   Loss_D: 0.1652  Loss_G: 1.9615
[96/200][501/549]   Loss_D: 0.0282  Loss_G: 5.0828
[96/200][549/549]   Loss_D: 0.1106  Loss_G: 2.4472
[97/200][  1/549]   Loss_D: 0.1388  Loss_G: 4.0282
[97/200][501/549]   Loss_D: 0.5366  Loss_G: 5.0007
[97/200][549/549]   Loss_D: 0.0202  Loss_G: 4.8210
[98/200][  1/549]   Loss_D: 0.0234  Loss_G: 4.4471
[98/200][501/549]   Loss_D: 0.1178  Loss_G: 3.1939
[98/200][549/549]   Loss_D: 0.1322  Loss_G: 4.2913
[99/200][  1/549]   Loss_D: 0.2034  Loss_G: 1.7403
[99/200][501/549]   Loss_D: 0.0475  Loss_G: 3.6451
[99/200][549/549]   Loss_D: 0.2296  Loss_G: 1.9866
[100/200][  1/549]   Loss_D: 0.2815  Loss_G: 1.5961
[100/200][501/549]   Loss_D: 0.0442  Loss_G: 4.8189
[100/200][549/549]   Loss_D: 0.1277  Loss_G: 4.8222
[101/200][  1/549]   Loss_D: 0.6803  Loss_G: 0.6855
[101/200][501/549]   Loss_D: 0.0402  Loss_G: 5.0521
[101/200][549/549]   Loss_D: 0.0686  Loss_G: 6.2515
[102/200][  1/549]   Loss_D: 0.0939  Loss_G: 2.7984
[102/200][501/549]   Loss_D: 0.1204  Loss_G: 2.6007
[102/200][549/549]   Loss_D: 0.1181  Loss_G: 5.6810
[103/200][  1/549]   Loss_D: 0.2361  Loss_G: 1.7472
[103/200][501/549]   Loss_D: 0.0427  Loss_G: 4.1614
[103/200][549/549]   Loss_D: 0.0504  Loss_G: 5.8122
[104/200][  1/549]   Loss_D: 0.0538  Loss_G: 3.1531
[104/200][501/549]   Loss_D: 2.0043  Loss_G: 0.1088
[104/200][549/549]   Loss_D: 0.0687  Loss_G: 4.7651
[105/200][  1/549]   Loss_D: 0.0604  Loss_G: 3.5815
[105/200][501/549]   Loss_D: 0.1369  Loss_G: 4.1490
[105/200][549/549]   Loss_D: 0.0890  Loss_G: 2.7475
[106/200][  1/549]   Loss_D: 0.1495  Loss_G: 2.4499
[106/200][501/549]   Loss_D: 0.0154  Loss_G: 5.3612
[106/200][549/549]   Loss_D: 0.0784  Loss_G: 3.2332
[107/200][  1/549]   Loss_D: 0.1104  Loss_G: 2.6433
[107/200][501/549]   Loss_D: 1.5659  Loss_G: 0.1345
[107/200][549/549]   Loss_D: 0.1050  Loss_G: 2.6811
[108/200][  1/549]   Loss_D: 0.0709  Loss_G: 3.9519
[108/200][501/549]   Loss_D: 0.0201  Loss_G: 4.5984
[108/200][549/549]   Loss_D: 0.0284  Loss_G: 7.5778
[109/200][  1/549]   Loss_D: 0.0402  Loss_G: 3.6348
[109/200][501/549]   Loss_D: 0.2203  Loss_G: 6.5892
[109/200][549/549]   Loss_D: 0.0646  Loss_G: 4.8339
[110/200][  1/549]   Loss_D: 0.1203  Loss_G: 2.1755
[110/200][501/549]   Loss_D: 0.2087  Loss_G: 1.8776
[110/200][549/549]   Loss_D: 0.0963  Loss_G: 4.9759
[111/200][  1/549]   Loss_D: 1.2461  Loss_G: 0.2730
[111/200][501/549]   Loss_D: 0.0719  Loss_G: 5.2483
[111/200][549/549]   Loss_D: 0.0291  Loss_G: 5.4161
[112/200][  1/549]   Loss_D: 0.0276  Loss_G: 4.1824
[112/200][501/549]   Loss_D: 0.0478  Loss_G: 4.4935
[112/200][549/549]   Loss_D: 0.0399  Loss_G: 4.4753
[113/200][  1/549]   Loss_D: 0.0474  Loss_G: 5.7254
[113/200][501/549]   Loss_D: 0.2062  Loss_G: 3.2870
[113/200][549/549]   Loss_D: 0.0296  Loss_G: 3.9117
[114/200][  1/549]   Loss_D: 0.0317  Loss_G: 4.0872
[114/200][501/549]   Loss_D: 0.0141  Loss_G: 5.2961
[114/200][549/549]   Loss_D: 0.1965  Loss_G: 2.2977
[115/200][  1/549]   Loss_D: 0.2078  Loss_G: 2.9888
[115/200][501/549]   Loss_D: 0.0288  Loss_G: 4.0424
[115/200][549/549]   Loss_D: 0.0101  Loss_G: 5.6805
[116/200][  1/549]   Loss_D: 0.0227  Loss_G: 5.2681
[116/200][501/549]   Loss_D: 0.0855  Loss_G: 3.3251
[116/200][549/549]   Loss_D: 0.0549  Loss_G: 4.5450
[117/200][  1/549]   Loss_D: 0.0361  Loss_G: 4.0229
[117/200][501/549]   Loss_D: 0.0300  Loss_G: 4.6825
[117/200][549/549]   Loss_D: 0.0535  Loss_G: 3.4068
[118/200][  1/549]   Loss_D: 0.0677  Loss_G: 3.5772
[118/200][501/549]   Loss_D: 0.0059  Loss_G: 6.3412
[118/200][549/549]   Loss_D: 0.0963  Loss_G: 4.4405
[119/200][  1/549]   Loss_D: 0.0630  Loss_G: 5.0424
[119/200][501/549]   Loss_D: 0.0352  Loss_G: 5.1691
[119/200][549/549]   Loss_D: 0.0080  Loss_G: 6.2851
[120/200][  1/549]   Loss_D: 0.0108  Loss_G: 5.3682
[120/200][501/549]   Loss_D: 0.0292  Loss_G: 5.7229
[120/200][549/549]   Loss_D: 0.0031  Loss_G: 7.5459
[121/200][  1/549]   Loss_D: 0.0018  Loss_G: 7.1893
[121/200][501/549]   Loss_D: 0.0211  Loss_G: 4.6909
[121/200][549/549]   Loss_D: 0.0255  Loss_G: 4.5894
[122/200][  1/549]   Loss_D: 0.1269  Loss_G: 2.4639
[122/200][501/549]   Loss_D: 0.0168  Loss_G: 5.4434
[122/200][549/549]   Loss_D: 0.0961  Loss_G: 3.4046
[123/200][  1/549]   Loss_D: 0.1574  Loss_G: 3.3845
[123/200][501/549]   Loss_D: 0.0390  Loss_G: 4.3531
[123/200][549/549]   Loss_D: 0.0439  Loss_G: 3.7912
[124/200][  1/549]   Loss_D: 0.0812  Loss_G: 3.1290
[124/200][501/549]   Loss_D: 0.0792  Loss_G: 3.6031
[124/200][549/549]   Loss_D: 0.2097  Loss_G: 2.7773
[125/200][  1/549]   Loss_D: 0.1206  Loss_G: 2.6864
[125/200][501/549]   Loss_D: 0.0994  Loss_G: 4.2767
[125/200][549/549]   Loss_D: 0.0862  Loss_G: 4.2990
[126/200][  1/549]   Loss_D: 0.1475  Loss_G: 4.4362
[126/200][501/549]   Loss_D: 0.0956  Loss_G: 2.9088
[126/200][549/549]   Loss_D: 0.0784  Loss_G: 3.2672
[127/200][  1/549]   Loss_D: 0.0667  Loss_G: 3.1276
[127/200][501/549]   Loss_D: 0.0593  Loss_G: 5.1181
[127/200][549/549]   Loss_D: 0.0231  Loss_G: 5.1231
[128/200][  1/549]   Loss_D: 0.0099  Loss_G: 5.8862
[128/200][501/549]   Loss_D: 0.0993  Loss_G: 3.1930
[128/200][549/549]   Loss_D: 0.0120  Loss_G: 4.6784
[129/200][  1/549]   Loss_D: 0.0294  Loss_G: 4.4950
[129/200][501/549]   Loss_D: 0.0961  Loss_G: 2.7850
[129/200][549/549]   Loss_D: 0.0488  Loss_G: 4.2303
[130/200][  1/549]   Loss_D: 0.2489  Loss_G: 1.8707
[130/200][501/549]   Loss_D: 0.0497  Loss_G: 4.6631
[130/200][549/549]   Loss_D: 0.0122  Loss_G: 5.4436
[131/200][  1/549]   Loss_D: 0.0222  Loss_G: 5.9528
[131/200][501/549]   Loss_D: 0.0081  Loss_G: 6.6881
[131/200][549/549]   Loss_D: 0.0430  Loss_G: 3.7047
[132/200][  1/549]   Loss_D: 0.3198  Loss_G: 1.5745
[132/200][501/549]   Loss_D: 0.0327  Loss_G: 4.7865
[132/200][549/549]   Loss_D: 0.0316  Loss_G: 4.7662
[133/200][  1/549]   Loss_D: 0.0672  Loss_G: 5.4749
[133/200][501/549]   Loss_D: 0.0124  Loss_G: 7.7918
[133/200][549/549]   Loss_D: 0.0541  Loss_G: 4.4108
[134/200][  1/549]   Loss_D: 0.1942  Loss_G: 1.9203
[134/200][501/549]   Loss_D: 0.3104  Loss_G: 8.1072
[134/200][549/549]   Loss_D: 0.0261  Loss_G: 7.3222
[135/200][  1/549]   Loss_D: 0.0065  Loss_G: 5.6847
[135/200][501/549]   Loss_D: 0.0219  Loss_G: 5.1363
[135/200][549/549]   Loss_D: 0.0120  Loss_G: 5.1785
[136/200][  1/549]   Loss_D: 0.0166  Loss_G: 4.6034
[136/200][501/549]   Loss_D: 0.0266  Loss_G: 4.3288
[136/200][549/549]   Loss_D: 0.0225  Loss_G: 5.6862
[137/200][  1/549]   Loss_D: 0.0261  Loss_G: 4.1803
[137/200][501/549]   Loss_D: 0.0441  Loss_G: 6.0091
[137/200][549/549]   Loss_D: 0.2349  Loss_G: 3.2408
[138/200][  1/549]   Loss_D: 0.2589  Loss_G: 1.7933
[138/200][501/549]   Loss_D: 0.1926  Loss_G: 2.3397
[138/200][549/549]   Loss_D: 0.1333  Loss_G: 2.9139
[139/200][  1/549]   Loss_D: 0.1251  Loss_G: 3.5228
[139/200][501/549]   Loss_D: 0.0115  Loss_G: 6.0259
[139/200][549/549]   Loss_D: 0.0036  Loss_G: 7.2037
[140/200][  1/549]   Loss_D: 0.0148  Loss_G: 6.5842
[140/200][501/549]   Loss_D: 0.0096  Loss_G: 5.4448
[140/200][549/549]   Loss_D: 0.0633  Loss_G: 3.3144
[141/200][  1/549]   Loss_D: 0.1099  Loss_G: 5.2266
[141/200][501/549]   Loss_D: 0.0495  Loss_G: 4.0772
[141/200][549/549]   Loss_D: 0.0496  Loss_G: 5.8312
[142/200][  1/549]   Loss_D: 0.0587  Loss_G: 3.2091
[142/200][501/549]   Loss_D: 0.0041  Loss_G: 7.1615
[142/200][549/549]   Loss_D: 0.0009  Loss_G: 7.3294
[143/200][  1/549]   Loss_D: 0.0044  Loss_G: 5.7795
[143/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[143/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[144/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[144/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[144/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[145/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[145/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[145/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[146/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[146/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[146/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[147/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[147/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[147/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[148/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[148/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[148/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[149/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[149/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[149/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[150/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[150/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[150/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[151/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[151/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[151/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[152/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[152/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[152/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[153/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[153/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[153/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[154/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[154/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[154/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[155/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[155/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[155/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[156/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[156/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[156/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[157/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[157/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[157/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[158/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[158/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[158/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[159/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[159/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[159/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[160/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[160/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[160/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[161/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[161/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[161/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[162/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[162/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[162/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[163/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[163/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[163/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[164/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[164/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[164/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[165/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[165/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[165/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[166/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[166/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[166/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[167/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[167/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[167/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[168/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[168/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[168/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[169/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[169/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[169/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[170/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[170/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[170/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[171/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[171/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[171/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[172/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[172/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[172/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[173/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[173/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[173/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[174/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[174/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[174/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[175/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[175/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[175/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[176/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[176/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[176/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[177/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[177/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[177/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[178/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[178/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[178/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[179/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[179/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[179/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[180/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[180/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[180/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[181/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[181/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[181/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[182/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[182/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[182/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[183/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[183/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[183/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[184/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[184/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[184/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[185/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[185/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[185/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[186/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[186/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[186/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[187/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[187/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[187/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[188/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[188/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[188/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[189/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[189/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[189/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[190/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[190/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[190/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[191/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[191/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[191/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[192/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[192/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[192/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[193/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[193/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[193/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[194/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[194/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[194/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[195/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[195/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[195/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[196/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[196/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[196/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[197/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[197/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[197/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[198/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[198/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[198/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[199/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[199/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[199/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
[200/200][  1/549]   Loss_D: 0.0000  Loss_G:27.6310
[200/200][501/549]   Loss_D: 0.0000  Loss_G:27.6310
[200/200][549/549]   Loss_D: 0.0000  Loss_G:27.6310
CPU times: user 6h 18min 32s, sys: 1h 56min 9s, total: 8h 14min 41s
Wall time: 49min 9s

结果展示

运行下面代码，描绘D和G损失与训练迭代的关系图：

plt.figure(figsize=(10, 5))
plt.title("Generator and Discriminator Loss During Training")
plt.plot(G_losses, label="G", color='blue')
plt.plot(D_losses, label="D", color='orange')
plt.xlabel("iterations")
plt.ylabel("Loss")
plt.legend()
plt.show()

可视化训练过程中通过隐向量fixed_noise生成的图像。

import matplotlib.pyplot as plt
import matplotlib.animation as animation

def showGif(image_list):
    show_list = []
    fig = plt.figure(figsize=(8, 3), dpi=120)
    for epoch in range(len(image_list)):
        images = []
        for i in range(3):
            row = np.concatenate((image_list[epoch][i * 8:(i + 1) * 8]), axis=1)
            images.append(row)
        img = np.clip(np.concatenate((images[:]), axis=0), 0, 1)
        plt.axis("off")
        show_list.append([plt.imshow(img)])

    ani = animation.ArtistAnimation(fig, show_list, interval=1000, repeat_delay=1000, blit=True)
    ani.save('./dcgan.gif', writer='pillow', fps=1)

showGif(image_list)

# 从文件中获取模型参数并加载到网络中
mindspore.load_checkpoint("./generator.ckpt", generator)

fixed_noise = ops.standard_normal((batch_size, nz, 1, 1))
img64 = generator(fixed_noise).transpose(0, 2, 3, 1).asnumpy()

fig = plt.figure(figsize=(8, 3), dpi=120)
images = []
for i in range(3):
    images.append(np.concatenate((img64[i * 8:(i + 1) * 8]), axis=1))
img = np.clip(np.concatenate((images[:]), axis=0), 0, 1)
plt.axis("off")
plt.imshow(img)
plt.show()