使用自适应判别器增强的数据高效 GAN

作者： András Béres
创建日期 2021/10/28
最后修改日期 2025/01/23
描述： 使用 Caltech Birds 数据集从有限的数据中生成图像。

在 Colab 中查看 • GitHub 源码

简介

GAN

生成对抗网络 (GAN) 是一类流行的生成式深度学习模型，常用于图像生成。它们由一对相互对抗的神经网络组成，称为判别器和生成器。判别器的任务是区分真实图像和生成的（伪造的）图像，而生成器网络则试图通过生成越来越逼真的图像来欺骗判别器。然而，如果生成器太容易或太难被欺骗，它可能无法为生成器提供有用的学习信号，因此训练 GAN 通常被认为是一项困难的任务。

GAN 的数据增强

数据增强是深度学习中常用的一种技术，它是指对输入数据随机应用语义保留变换，以生成多个逼真的版本，从而有效地增加可用的训练数据量。最简单的例子是左右翻转图像，它在保留图像内容的同时生成第二个独特的训练样本。数据增强通常用于监督学习中，以防止过拟合并增强泛化能力。

StyleGAN2-ADA 的作者表明，判别器过拟合可能是 GAN 中的一个问题，尤其是在只有少量训练数据可用时。他们提出了自适应判别器增强来缓解这个问题。

然而，将数据增强应用于 GAN 并非易事。由于生成器是使用判别器的梯度更新的，如果生成的图像被增强，则增强管道必须是可微分的，并且还必须与 GPU 兼容以提高计算效率。幸运的是，Keras 图像增强层 满足这两个要求，因此非常适合此任务。

可逆数据增强

在生成模型中使用数据增强时，一个可能的困难是 “泄露增强”（第 2.2 节） 的问题，即模型生成已经增强的图像。 这意味着它无法将增强与底层数据分布分离开来，这可能是由于使用了不可逆的数据转换造成的。 例如，如果以相等的概率执行 0 度、90 度、180 度或 270 度旋转，则图像的原始方向将无法推断，并且此信息将被破坏。

使数据增强可逆的一个简单技巧是仅以一定的概率应用它们。 这样，图像的原始版本将更常见，并且可以推断出数据分布。 通过正确选择此概率，可以有效地正则化判别器，而不会使增强泄露。

设置

import os

os.environ["KERAS_BACKEND"] = "tensorflow"

import matplotlib.pyplot as plt
import tensorflow as tf
import tensorflow_datasets as tfds

import keras
from keras import ops
from keras import layers

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1738798965.367584   17795 cuda_dnn.cc:8310] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1738798965.374084   17795 cuda_blas.cc:1418] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered

超参数

# data
num_epochs = 10  # train for 400 epochs for good results
image_size = 64
# resolution of Kernel Inception Distance measurement, see related section
kid_image_size = 75
padding = 0.25
dataset_name = "caltech_birds2011"

# adaptive discriminator augmentation
max_translation = 0.125
max_rotation = 0.125
max_zoom = 0.25
target_accuracy = 0.85
integration_steps = 1000

# architecture
noise_size = 64
depth = 4
width = 128
leaky_relu_slope = 0.2
dropout_rate = 0.4

# optimization
batch_size = 128
learning_rate = 2e-4
beta_1 = 0.5  # not using the default value of 0.9 is important
ema = 0.99

数据管道

在此示例中，我们将使用 Caltech Birds (2011) 数据集来生成鸟类图像，这是一个多样化的自然数据集，其中包含少于 6000 张图像用于训练。 当处理如此少量的数据时，必须格外小心以尽可能保持高数据质量。 在此示例中，我们使用提供的鸟类边界框将它们裁剪成正方形，同时在可能的情况下保留其纵横比。

def round_to_int(float_value):
    return ops.cast(ops.round(float_value), "int32")


def preprocess_image(data):
    # unnormalize bounding box coordinates
    height = ops.cast(ops.shape(data["image"])[0], "float32")
    width = ops.cast(ops.shape(data["image"])[1], "float32")
    bounding_box = data["bbox"] * ops.stack([height, width, height, width])

    # calculate center and length of longer side, add padding
    target_center_y = 0.5 * (bounding_box[0] + bounding_box[2])
    target_center_x = 0.5 * (bounding_box[1] + bounding_box[3])
    target_size = ops.maximum(
        (1.0 + padding) * (bounding_box[2] - bounding_box[0]),
        (1.0 + padding) * (bounding_box[3] - bounding_box[1]),
    )

    # modify crop size to fit into image
    target_height = ops.min(
        [target_size, 2.0 * target_center_y, 2.0 * (height - target_center_y)]
    )
    target_width = ops.min(
        [target_size, 2.0 * target_center_x, 2.0 * (width - target_center_x)]
    )

    # crop image, `ops.image.crop_images` only works with non-tensor croppings
    image = ops.slice(
        data["image"],
        start_indices=(
            round_to_int(target_center_y - 0.5 * target_height),
            round_to_int(target_center_x - 0.5 * target_width),
            0,
        ),
        shape=(round_to_int(target_height), round_to_int(target_width), 3),
    )

    # resize and clip
    image = ops.cast(image, "float32")
    image = ops.image.resize(image, [image_size, image_size])

    return ops.clip(image / 255.0, 0.0, 1.0)


def prepare_dataset(split):
    # the validation dataset is shuffled as well, because data order matters
    # for the KID calculation
    return (
        tfds.load(dataset_name, split=split, shuffle_files=True)
        .map(preprocess_image, num_parallel_calls=tf.data.AUTOTUNE)
        .cache()
        .shuffle(10 * batch_size)
        .batch(batch_size, drop_remainder=True)
        .prefetch(buffer_size=tf.data.AUTOTUNE)
    )


train_dataset = prepare_dataset("train")
val_dataset = prepare_dataset("test")

I0000 00:00:1738798971.054632   17795 gpu_device.cc:2022] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 13840 MB memory:  -> device: 0, name: Tesla T4, pci bus id: 0000:00:04.0, compute capability: 7.5

预处理后，训练图像如下所示：

内核初始距离

内核初始距离 (KID) 被提出作为流行的 Frechet 初始距离 (FID) 指标的替代品，用于衡量图像生成质量。 这两个指标都衡量在 InceptionV3 网络（在 ImageNet 上预训练）的表示空间中，生成分布和训练分布之间的差异。

根据论文，提出 KID 是因为 FID 没有无偏估计量，当在较少的图像上测量时，其期望值较高。 KID 更适合小型数据集，因为其期望值不依赖于测量的样本数量。 以我的经验，它在计算上也更轻便，数值上更稳定，并且更易于实现，因为它可以通过批量方式进行估计。

在此示例中，图像以 Inception 网络的最小可能分辨率（75x75 而不是 299x299）进行评估，并且该指标仅在验证集上进行测量，以提高计算效率。

class KID(keras.metrics.Metric):
    def __init__(self, name="kid", **kwargs):
        super().__init__(name=name, **kwargs)

        # KID is estimated per batch and is averaged across batches
        self.kid_tracker = keras.metrics.Mean()

        # a pretrained InceptionV3 is used without its classification layer
        # transform the pixel values to the 0-255 range, then use the same
        # preprocessing as during pretraining
        self.encoder = keras.Sequential(
            [
                layers.InputLayer(input_shape=(image_size, image_size, 3)),
                layers.Rescaling(255.0),
                layers.Resizing(height=kid_image_size, width=kid_image_size),
                layers.Lambda(keras.applications.inception_v3.preprocess_input),
                keras.applications.InceptionV3(
                    include_top=False,
                    input_shape=(kid_image_size, kid_image_size, 3),
                    weights="imagenet",
                ),
                layers.GlobalAveragePooling2D(),
            ],
            name="inception_encoder",
        )

    def polynomial_kernel(self, features_1, features_2):
        feature_dimensions = ops.cast(ops.shape(features_1)[1], "float32")
        return (
            features_1 @ ops.transpose(features_2) / feature_dimensions + 1.0
        ) ** 3.0

    def update_state(self, real_images, generated_images, sample_weight=None):
        real_features = self.encoder(real_images, training=False)
        generated_features = self.encoder(generated_images, training=False)

        # compute polynomial kernels using the two sets of features
        kernel_real = self.polynomial_kernel(real_features, real_features)
        kernel_generated = self.polynomial_kernel(
            generated_features, generated_features
        )
        kernel_cross = self.polynomial_kernel(real_features, generated_features)

        # estimate the squared maximum mean discrepancy using the average kernel values
        batch_size = ops.shape(real_features)[0]
        batch_size_f = ops.cast(batch_size, "float32")
        mean_kernel_real = ops.sum(kernel_real * (1.0 - ops.eye(batch_size))) / (
            batch_size_f * (batch_size_f - 1.0)
        )
        mean_kernel_generated = ops.sum(
            kernel_generated * (1.0 - ops.eye(batch_size))
        ) / (batch_size_f * (batch_size_f - 1.0))
        mean_kernel_cross = ops.mean(kernel_cross)
        kid = mean_kernel_real + mean_kernel_generated - 2.0 * mean_kernel_cross

        # update the average KID estimate
        self.kid_tracker.update_state(kid)

    def result(self):
        return self.kid_tracker.result()

    def reset_state(self):
        self.kid_tracker.reset_state()

自适应判别器增强

StyleGAN2-ADA 的作者建议在训练期间自适应地更改增强概率。 虽然论文中对其解释不同，但他们使用增强概率的 积分控制，以使判别器在真实图像上的准确率保持接近目标值。 请注意，他们控制的变量实际上是判别器 logits 的平均符号（论文中的 r_t），它对应于 2 * 准确率 - 1。

此方法需要两个超参数

1. target_accuracy：判别器在真实图像上的准确率的目标值。 我建议从 80-90% 的范围内选择其值。
2. integration_steps：将 100% 的准确率误差转换为 100% 的增强概率增加所需的更新步数。 为了给出直观的理解，这定义了增强概率变化的速度。 我建议将其设置为相对较高的值（在本例中为 1000），以便仅缓慢调整增强强度。

此过程的主要动机是，目标准确率的最佳值在不同的数据集大小之间是相似的（请参阅论文中的图 4 和图 5），因此无需重新调整，因为该过程会在需要时自动应用更强的数据增强。

# "hard sigmoid", useful for binary accuracy calculation from logits
def step(values):
    # negative values -> 0.0, positive values -> 1.0
    return 0.5 * (1.0 + ops.sign(values))


# augments images with a probability that is dynamically updated during training
class AdaptiveAugmenter(keras.Model):
    def __init__(self):
        super().__init__()

        # stores the current probability of an image being augmented
        self.probability = keras.Variable(0.0)
        self.seed_generator = keras.random.SeedGenerator(42)

        # the corresponding augmentation names from the paper are shown above each layer
        # the authors show (see figure 4), that the blitting and geometric augmentations
        # are the most helpful in the low-data regime
        self.augmenter = keras.Sequential(
            [
                layers.InputLayer(input_shape=(image_size, image_size, 3)),
                # blitting/x-flip:
                layers.RandomFlip("horizontal"),
                # blitting/integer translation:
                layers.RandomTranslation(
                    height_factor=max_translation,
                    width_factor=max_translation,
                    interpolation="nearest",
                ),
                # geometric/rotation:
                layers.RandomRotation(factor=max_rotation),
                # geometric/isotropic and anisotropic scaling:
                layers.RandomZoom(
                    height_factor=(-max_zoom, 0.0), width_factor=(-max_zoom, 0.0)
                ),
            ],
            name="adaptive_augmenter",
        )

    def call(self, images, training):
        if training:
            augmented_images = self.augmenter(images, training=training)

            # during training either the original or the augmented images are selected
            # based on self.probability
            augmentation_values = keras.random.uniform(
                shape=(batch_size, 1, 1, 1), seed=self.seed_generator
            )
            augmentation_bools = ops.less(augmentation_values, self.probability)

            images = ops.where(augmentation_bools, augmented_images, images)
        return images

    def update(self, real_logits):
        current_accuracy = ops.mean(step(real_logits))

        # the augmentation probability is updated based on the discriminator's
        # accuracy on real images
        accuracy_error = current_accuracy - target_accuracy
        self.probability.assign(
            ops.clip(self.probability + accuracy_error / integration_steps, 0.0, 1.0)
        )

网络架构

这里我们指定两个网络的架构

• 生成器：将随机向量映射到图像，该图像应尽可能逼真
• 判别器：将图像映射到标量分数，真实图像的分数应高，生成图像的分数应低

GAN 往往对网络架构很敏感，在此示例中，我实现了 DCGAN 架构，因为它在训练期间相对稳定，同时易于实现。 我们在整个网络中使用恒定数量的过滤器，在生成器的最后一层中使用 sigmoid 而不是 tanh，并使用默认初始化而不是随机正态作为进一步的简化。

作为一种良好的实践，我们禁用了批归一化层中的可学习比例参数，因为一方面，后面的 relu + 卷积层使其变得冗余（如 文档 中所述）。 但也因为它应该根据理论在使用 谱归一化（第 4.1 节） 时禁用，此处未使用谱归一化，但在 GAN 中很常见。 我们还禁用了全连接层和卷积层中的偏差，因为后面的批归一化使其变得冗余。

# DCGAN generator
def get_generator():
    noise_input = keras.Input(shape=(noise_size,))
    x = layers.Dense(4 * 4 * width, use_bias=False)(noise_input)
    x = layers.BatchNormalization(scale=False)(x)
    x = layers.ReLU()(x)
    x = layers.Reshape(target_shape=(4, 4, width))(x)
    for _ in range(depth - 1):
        x = layers.Conv2DTranspose(
            width,
            kernel_size=4,
            strides=2,
            padding="same",
            use_bias=False,
        )(x)
        x = layers.BatchNormalization(scale=False)(x)
        x = layers.ReLU()(x)
    image_output = layers.Conv2DTranspose(
        3,
        kernel_size=4,
        strides=2,
        padding="same",
        activation="sigmoid",
    )(x)

    return keras.Model(noise_input, image_output, name="generator")


# DCGAN discriminator
def get_discriminator():
    image_input = keras.Input(shape=(image_size, image_size, 3))
    x = image_input
    for _ in range(depth):
        x = layers.Conv2D(
            width,
            kernel_size=4,
            strides=2,
            padding="same",
            use_bias=False,
        )(x)
        x = layers.BatchNormalization(scale=False)(x)
        x = layers.LeakyReLU(alpha=leaky_relu_slope)(x)
    x = layers.Flatten()(x)
    x = layers.Dropout(dropout_rate)(x)
    output_score = layers.Dense(1)(x)

    return keras.Model(image_input, output_score, name="discriminator")

GAN 模型

class GAN_ADA(keras.Model):
    def __init__(self):
        super().__init__()

        self.seed_generator = keras.random.SeedGenerator(seed=42)
        self.augmenter = AdaptiveAugmenter()
        self.generator = get_generator()
        self.ema_generator = keras.models.clone_model(self.generator)
        self.discriminator = get_discriminator()

        self.generator.summary()
        self.discriminator.summary()
        # we have created all layers at this point, so we can mark the model
        # as having been built
        self.built = True

    def compile(self, generator_optimizer, discriminator_optimizer, **kwargs):
        super().compile(**kwargs)

        # separate optimizers for the two networks
        self.generator_optimizer = generator_optimizer
        self.discriminator_optimizer = discriminator_optimizer

        self.generator_loss_tracker = keras.metrics.Mean(name="g_loss")
        self.discriminator_loss_tracker = keras.metrics.Mean(name="d_loss")
        self.real_accuracy = keras.metrics.BinaryAccuracy(name="real_acc")
        self.generated_accuracy = keras.metrics.BinaryAccuracy(name="gen_acc")
        self.augmentation_probability_tracker = keras.metrics.Mean(name="aug_p")
        self.kid = KID()

    @property
    def metrics(self):
        return [
            self.generator_loss_tracker,
            self.discriminator_loss_tracker,
            self.real_accuracy,
            self.generated_accuracy,
            self.augmentation_probability_tracker,
            self.kid,
        ]

    def generate(self, batch_size, training):
        latent_samples = keras.random.normal(
            shape=(batch_size, noise_size), seed=self.seed_generator
        )
        # use ema_generator during inference
        if training:
            generated_images = self.generator(latent_samples, training=training)
        else:
            generated_images = self.ema_generator(latent_samples, training=training)
        return generated_images

    def adversarial_loss(self, real_logits, generated_logits):
        # this is usually called the non-saturating GAN loss

        real_labels = ops.ones(shape=(batch_size, 1))
        generated_labels = ops.zeros(shape=(batch_size, 1))

        # the generator tries to produce images that the discriminator considers as real
        generator_loss = keras.losses.binary_crossentropy(
            real_labels, generated_logits, from_logits=True
        )
        # the discriminator tries to determine if images are real or generated
        discriminator_loss = keras.losses.binary_crossentropy(
            ops.concatenate([real_labels, generated_labels], axis=0),
            ops.concatenate([real_logits, generated_logits], axis=0),
            from_logits=True,
        )

        return ops.mean(generator_loss), ops.mean(discriminator_loss)

    def train_step(self, real_images):
        real_images = self.augmenter(real_images, training=True)

        # use persistent gradient tape because gradients will be calculated twice
        with tf.GradientTape(persistent=True) as tape:
            generated_images = self.generate(batch_size, training=True)
            # gradient is calculated through the image augmentation
            generated_images = self.augmenter(generated_images, training=True)

            # separate forward passes for the real and generated images, meaning
            # that batch normalization is applied separately
            real_logits = self.discriminator(real_images, training=True)
            generated_logits = self.discriminator(generated_images, training=True)

            generator_loss, discriminator_loss = self.adversarial_loss(
                real_logits, generated_logits
            )

        # calculate gradients and update weights
        generator_gradients = tape.gradient(
            generator_loss, self.generator.trainable_weights
        )
        discriminator_gradients = tape.gradient(
            discriminator_loss, self.discriminator.trainable_weights
        )
        self.generator_optimizer.apply_gradients(
            zip(generator_gradients, self.generator.trainable_weights)
        )
        self.discriminator_optimizer.apply_gradients(
            zip(discriminator_gradients, self.discriminator.trainable_weights)
        )

        # update the augmentation probability based on the discriminator's performance
        self.augmenter.update(real_logits)

        self.generator_loss_tracker.update_state(generator_loss)
        self.discriminator_loss_tracker.update_state(discriminator_loss)
        self.real_accuracy.update_state(1.0, step(real_logits))
        self.generated_accuracy.update_state(0.0, step(generated_logits))
        self.augmentation_probability_tracker.update_state(self.augmenter.probability)

        # track the exponential moving average of the generator's weights to decrease
        # variance in the generation quality
        for weight, ema_weight in zip(
            self.generator.weights, self.ema_generator.weights
        ):
            ema_weight.assign(ema * ema_weight + (1 - ema) * weight)

        # KID is not measured during the training phase for computational efficiency
        return {m.name: m.result() for m in self.metrics[:-1]}

    def test_step(self, real_images):
        generated_images = self.generate(batch_size, training=False)

        self.kid.update_state(real_images, generated_images)

        # only KID is measured during the evaluation phase for computational efficiency
        return {self.kid.name: self.kid.result()}

    def plot_images(self, epoch=None, logs=None, num_rows=3, num_cols=6, interval=5):
        # plot random generated images for visual evaluation of generation quality
        if epoch is None or (epoch + 1) % interval == 0:
            num_images = num_rows * num_cols
            generated_images = self.generate(num_images, training=False)

            plt.figure(figsize=(num_cols * 2.0, num_rows * 2.0))
            for row in range(num_rows):
                for col in range(num_cols):
                    index = row * num_cols + col
                    plt.subplot(num_rows, num_cols, index + 1)
                    plt.imshow(generated_images[index])
                    plt.axis("off")
            plt.tight_layout()
            plt.show()
            plt.close()

训练

从训练期间的指标可以看出，如果真实准确率（判别器在真实图像上的准确率）低于目标准确率，则增强概率会增加，反之亦然。 以我的经验，在健康的 GAN 训练期间，判别器准确率应保持在 80-95% 的范围内。 低于该范围，判别器太弱，高于该范围，判别器太强。

请注意，我们跟踪生成器权重的指数移动平均值，并将其用于图像生成和 KID 评估。

# create and compile the model
model = GAN_ADA()
model.compile(
    generator_optimizer=keras.optimizers.Adam(learning_rate, beta_1),
    discriminator_optimizer=keras.optimizers.Adam(learning_rate, beta_1),
)

# save the best model based on the validation KID metric
checkpoint_path = "gan_model.weights.h5"
checkpoint_callback = keras.callbacks.ModelCheckpoint(
    filepath=checkpoint_path,
    save_weights_only=True,
    monitor="val_kid",
    mode="min",
    save_best_only=True,
)

# run training and plot generated images periodically
model.fit(
    train_dataset,
    epochs=num_epochs,
    validation_data=val_dataset,
    callbacks=[
        keras.callbacks.LambdaCallback(on_epoch_end=model.plot_images),
        checkpoint_callback,
    ],
)

/usr/local/lib/python3.11/dist-packages/keras/src/layers/core/input_layer.py:27: UserWarning: Argument `input_shape` is deprecated. Use `shape` instead.
  warnings.warn(

/usr/local/lib/python3.11/dist-packages/keras/src/layers/activations/leaky_relu.py:41: UserWarning: Argument `alpha` is deprecated. Use `negative_slope` instead.
  warnings.warn(

Model: "generator"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ input_layer_1 (InputLayer)      │ (None, 64)             │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense (Dense)                   │ (None, 2048)           │       131,072 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization             │ (None, 2048)           │         6,144 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ re_lu (ReLU)                    │ (None, 2048)           │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ reshape (Reshape)               │ (None, 4, 4, 128)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose                │ (None, 8, 8, 128)      │       262,144 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_1           │ (None, 8, 8, 128)      │           384 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ re_lu_1 (ReLU)                  │ (None, 8, 8, 128)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_1              │ (None, 16, 16, 128)    │       262,144 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_2           │ (None, 16, 16, 128)    │           384 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ re_lu_2 (ReLU)                  │ (None, 16, 16, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_2              │ (None, 32, 32, 128)    │       262,144 │
│ (Conv2DTranspose)               │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_3           │ (None, 32, 32, 128)    │           384 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ re_lu_3 (ReLU)                  │ (None, 32, 32, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_transpose_3              │ (None, 64, 64, 3)      │         6,147 │
│ (Conv2DTranspose)               │                        │               │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 930,947 (3.55 MB)

 Trainable params: 926,083 (3.53 MB)

 Non-trainable params: 4,864 (19.00 KB)

Model: "discriminator"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ input_layer_2 (InputLayer)      │ (None, 64, 64, 3)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d (Conv2D)                 │ (None, 32, 32, 128)    │         6,144 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_4           │ (None, 32, 32, 128)    │           384 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu (LeakyReLU)         │ (None, 32, 32, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_1 (Conv2D)               │ (None, 16, 16, 128)    │       262,144 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_5           │ (None, 16, 16, 128)    │           384 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_1 (LeakyReLU)       │ (None, 16, 16, 128)    │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_2 (Conv2D)               │ (None, 8, 8, 128)      │       262,144 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_6           │ (None, 8, 8, 128)      │           384 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_2 (LeakyReLU)       │ (None, 8, 8, 128)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ conv2d_3 (Conv2D)               │ (None, 4, 4, 128)      │       262,144 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ batch_normalization_7           │ (None, 4, 4, 128)      │           384 │
│ (BatchNormalization)            │                        │               │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ leaky_re_lu_3 (LeakyReLU)       │ (None, 4, 4, 128)      │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten (Flatten)               │ (None, 2048)           │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dropout (Dropout)               │ (None, 2048)           │             0 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ (None, 1)              │         2,049 │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 796,161 (3.04 MB)

 Trainable params: 795,137 (3.03 MB)

 Non-trainable params: 1,024 (4.00 KB)

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/inception_v3/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5

    0/87910968 ━━━━━━━━━━━━━━━━━━━━  0s 0s/step



4202496/87910968 [37m━━━━━━━━━━━━━━━━━━━━ 1s 0us/step



12304384/87910968 ━━ [37m━━━━━━━━━━━━━━━━━━ 0s 0us/step



21815296/87910968 ━━━━ [37m━━━━━━━━━━━━━━━━ 0s 0us/step



34373632/87910968 ━━━━━━━ [37m━━━━━━━━━━━━━ 0s 0us/step



47669248/87910968 ━━━━━━━━━━ [37m━━━━━━━━━━ 0s 0us/step



62324736/87910968 ━━━━━━━━━━━━━━ [37m━━━━━━ 0s 0us/step



72032256/87910968 ━━━━━━━━━━━━━━━━ [37m━━━━ 0s 0us/step



82083840/87910968 ━━━━━━━━━━━━━━━━━━ [37m━━ 0s 0us/step



87910968/87910968 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step

Epoch 1/10

E0000 00:00:1738798983.901596   17795 meta_optimizer.cc:966] layout failed: INVALID_ARGUMENT: Size of values 0 does not match size of permutation 4 @ fanin shape inStatefulPartitionedCall/gradient_tape/adaptive_augmenter_3/SelectV2_1-1-TransposeNHWCToNCHW-LayoutOptimizer

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1738798987.822990   17861 cuda_solvers.cc:178] Creating GpuSolver handles for stream 0x9f45670

I0000 00:00:1738798988.976919   17862 cuda_dnn.cc:529] Loaded cuDNN version 90300

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 13:22 18s/step - aug_p: 0.0000e+00 - d_loss: 0.8829 - g_loss: 0.5585 - gen_acc: 0.2812 - real_acc: 0.7031



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 7s 173ms/step - aug_p: 0.0000e+00 - d_loss: 0.7985 - g_loss: 0.8154 - gen_acc: 0.4258 - real_acc: 0.6777



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 8s 201ms/step - aug_p: 0.0000e+00 - d_loss: 0.7488 - g_loss: 0.9481 - gen_acc: 0.5052 - real_acc: 0.6680



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 10s 249ms/step - aug_p: 0.0000e+00 - d_loss: 0.7106 - g_loss: 1.0407 - gen_acc: 0.5586 - real_acc: 0.6680



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 11s 269ms/step - aug_p: 0.0000e+00 - d_loss: 0.6782 - g_loss: 1.1104 - gen_acc: 0.5991 - real_acc: 0.6744



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 10s 273ms/step - aug_p: 0.0000e+00 - d_loss: 0.6506 - g_loss: 1.1692 - gen_acc: 0.6301 - real_acc: 0.6818



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 10s 280ms/step - aug_p: 5.1020e-07 - d_loss: 0.6253 - g_loss: 1.2302 - gen_acc: 0.6558 - real_acc: 0.6902



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 10s 289ms/step - aug_p: 1.4962e-06 - d_loss: 0.6030 - g_loss: 1.2838 - gen_acc: 0.6772 - real_acc: 0.6986



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 10s 297ms/step - aug_p: 3.8570e-06 - d_loss: 0.5829 - g_loss: 1.3313 - gen_acc: 0.6952 - real_acc: 0.7074



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 10s 299ms/step - aug_p: 7.9244e-06 - d_loss: 0.5644 - g_loss: 1.3805 - gen_acc: 0.7109 - real_acc: 0.7162



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 11s 314ms/step - aug_p: 1.3531e-05 - d_loss: 0.5473 - g_loss: 1.4301 - gen_acc: 0.7249 - real_acc: 0.7245



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 11s 333ms/step - aug_p: 2.0444e-05 - d_loss: 0.5318 - g_loss: 1.4743 - gen_acc: 0.7373 - real_acc: 0.7322



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 11s 352ms/step - aug_p: 2.8561e-05 - d_loss: 0.5174 - g_loss: 1.5181 - gen_acc: 0.7485 - real_acc: 0.7394



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 11s 365ms/step - aug_p: 3.7929e-05 - d_loss: 0.5040 - g_loss: 1.5586 - gen_acc: 0.7587 - real_acc: 0.7464



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 11s 379ms/step - aug_p: 4.8560e-05 - d_loss: 0.4914 - g_loss: 1.5977 - gen_acc: 0.7680 - real_acc: 0.7530



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 11s 387ms/step - aug_p: 6.0448e-05 - d_loss: 0.4795 - g_loss: 1.6355 - gen_acc: 0.7766 - real_acc: 0.7593



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 11s 399ms/step - aug_p: 7.3577e-05 - d_loss: 0.4683 - g_loss: 1.6715 - gen_acc: 0.7844 - real_acc: 0.7654



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 11s 414ms/step - aug_p: 8.7805e-05 - d_loss: 0.4578 - g_loss: 1.7063 - gen_acc: 0.7917 - real_acc: 0.7711



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 11s 422ms/step - aug_p: 1.0308e-04 - d_loss: 0.4480 - g_loss: 1.7386 - gen_acc: 0.7984 - real_acc: 0.7765



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 11s 433ms/step - aug_p: 1.1933e-04 - d_loss: 0.4386 - g_loss: 1.7736 - gen_acc: 0.8046 - real_acc: 0.7817



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 11s 444ms/step - aug_p: 1.3652e-04 - d_loss: 0.4297 - g_loss: 1.8065 - gen_acc: 0.8104 - real_acc: 0.7866



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 10s 440ms/step - aug_p: 1.5459e-04 - d_loss: 0.4213 - g_loss: 1.8383 - gen_acc: 0.8158 - real_acc: 0.7913



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 9s 434ms/step - aug_p: 1.7347e-04 - d_loss: 0.4132 - g_loss: 1.8694 - gen_acc: 0.8209 - real_acc: 0.7958



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 9s 429ms/step - aug_p: 1.9312e-04 - d_loss: 0.4056 - g_loss: 1.8988 - gen_acc: 0.8257 - real_acc: 0.8000



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 8s 426ms/step - aug_p: 2.1348e-04 - d_loss: 0.3983 - g_loss: 1.9278 - gen_acc: 0.8302 - real_acc: 0.8041



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 8s 423ms/step - aug_p: 2.3451e-04 - d_loss: 0.3914 - g_loss: 1.9548 - gen_acc: 0.8345 - real_acc: 0.8079



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 8s 427ms/step - aug_p: 2.5614e-04 - d_loss: 0.3848 - g_loss: 1.9828 - gen_acc: 0.8385 - real_acc: 0.8116



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 7s 424ms/step - aug_p: 2.7834e-04 - d_loss: 0.3785 - g_loss: 2.0093 - gen_acc: 0.8423 - real_acc: 0.8151



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 7s 420ms/step - aug_p: 3.0107e-04 - d_loss: 0.3724 - g_loss: 2.0347 - gen_acc: 0.8459 - real_acc: 0.8185



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 6s 416ms/step - aug_p: 3.2432e-04 - d_loss: 0.3666 - g_loss: 2.0599 - gen_acc: 0.8493 - real_acc: 0.8218



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 6s 413ms/step - aug_p: 3.4806e-04 - d_loss: 0.3610 - g_loss: 2.0840 - gen_acc: 0.8526 - real_acc: 0.8249



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 5s 409ms/step - aug_p: 3.7225e-04 - d_loss: 0.3556 - g_loss: 2.1073 - gen_acc: 0.8556 - real_acc: 0.8279



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 5s 406ms/step - aug_p: 3.9686e-04 - d_loss: 0.3505 - g_loss: 2.1300 - gen_acc: 0.8586 - real_acc: 0.8307



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 4s 402ms/step - aug_p: 4.2187e-04 - d_loss: 0.3455 - g_loss: 2.1520 - gen_acc: 0.8614 - real_acc: 0.8335



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 4s 400ms/step - aug_p: 4.4725e-04 - d_loss: 0.3407 - g_loss: 2.1736 - gen_acc: 0.8641 - real_acc: 0.8361



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 3s 397ms/step - aug_p: 4.7297e-04 - d_loss: 0.3361 - g_loss: 2.1947 - gen_acc: 0.8667 - real_acc: 0.8387



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 3s 393ms/step - aug_p: 4.9903e-04 - d_loss: 0.3316 - g_loss: 2.2152 - gen_acc: 0.8691 - real_acc: 0.8411



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 3s 387ms/step - aug_p: 5.2539e-04 - d_loss: 0.3273 - g_loss: 2.2357 - gen_acc: 0.8715 - real_acc: 0.8435



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 2s 382ms/step - aug_p: 5.5206e-04 - d_loss: 0.3231 - g_loss: 2.2554 - gen_acc: 0.8738 - real_acc: 0.8458



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 2s 376ms/step - aug_p: 5.7902e-04 - d_loss: 0.3191 - g_loss: 2.2756 - gen_acc: 0.8759 - real_acc: 0.8480



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1s 371ms/step - aug_p: 6.0626e-04 - d_loss: 0.3151 - g_loss: 2.2954 - gen_acc: 0.8780 - real_acc: 0.8502



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 1s 366ms/step - aug_p: 6.3377e-04 - d_loss: 0.3113 - g_loss: 2.3147 - gen_acc: 0.8800 - real_acc: 0.8522



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 1s 362ms/step - aug_p: 6.6154e-04 - d_loss: 0.3076 - g_loss: 2.3339 - gen_acc: 0.8820 - real_acc: 0.8543



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0s 358ms/step - aug_p: 6.8956e-04 - d_loss: 0.3041 - g_loss: 2.3524 - gen_acc: 0.8839 - real_acc: 0.8562



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0s 354ms/step - aug_p: 7.1780e-04 - d_loss: 0.3006 - g_loss: 2.3703 - gen_acc: 0.8857 - real_acc: 0.8581



46/46 ━━━━━━━━━━━━━━━━━━━━ 0s 350ms/step - aug_p: 7.4625e-04 - d_loss: 0.2973 - g_loss: 2.3871 - gen_acc: 0.8874 - real_acc: 0.8599



46/46 ━━━━━━━━━━━━━━━━━━━━ 61s 958ms/step - aug_p: 7.7349e-04 - d_loss: 0.2942 - g_loss: 2.4032 - gen_acc: 0.8890 - real_acc: 0.8616 - val_kid: 9.1841

Epoch 2/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 22:28 30s/step - aug_p: 0.0051 - d_loss: 0.1030 - g_loss: 8.8928 - gen_acc: 1.0000 - real_acc: 0.9375



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 7s 176ms/step - aug_p: 0.0051 - d_loss: 0.1073 - g_loss: 8.4938 - gen_acc: 0.9980 - real_acc: 0.9297



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7s 176ms/step - aug_p: 0.0051 - d_loss: 0.1074 - g_loss: 7.7540 - gen_acc: 0.9952 - real_acc: 0.9340



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7s 176ms/step - aug_p: 0.0052 - d_loss: 0.1037 - g_loss: 7.2815 - gen_acc: 0.9945 - real_acc: 0.9388



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7s 175ms/step - aug_p: 0.0052 - d_loss: 0.0997 - g_loss: 6.9305 - gen_acc: 0.9943 - real_acc: 0.9432



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7s 175ms/step - aug_p: 0.0052 - d_loss: 0.0960 - g_loss: 6.6418 - gen_acc: 0.9942 - real_acc: 0.9473



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6s 175ms/step - aug_p: 0.0052 - d_loss: 0.0933 - g_loss: 6.4224 - gen_acc: 0.9939 - real_acc: 0.9503



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6s 175ms/step - aug_p: 0.0053 - d_loss: 0.0907 - g_loss: 6.2473 - gen_acc: 0.9937 - real_acc: 0.9530



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6s 175ms/step - aug_p: 0.0053 - d_loss: 0.0885 - g_loss: 6.0970 - gen_acc: 0.9936 - real_acc: 0.9552



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6s 175ms/step - aug_p: 0.0053 - d_loss: 0.0868 - g_loss: 5.9686 - gen_acc: 0.9936 - real_acc: 0.9571



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6s 175ms/step - aug_p: 0.0054 - d_loss: 0.0852 - g_loss: 5.8546 - gen_acc: 0.9936 - real_acc: 0.9588



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 5s 176ms/step - aug_p: 0.0054 - d_loss: 0.0837 - g_loss: 5.7615 - gen_acc: 0.9937 - real_acc: 0.9602



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 5s 175ms/step - aug_p: 0.0054 - d_loss: 0.0825 - g_loss: 5.6750 - gen_acc: 0.9937 - real_acc: 0.9614



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5s 176ms/step - aug_p: 0.0055 - d_loss: 0.0813 - g_loss: 5.5972 - gen_acc: 0.9937 - real_acc: 0.9626



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5s 176ms/step - aug_p: 0.0055 - d_loss: 0.0802 - g_loss: 5.5273 - gen_acc: 0.9938 - real_acc: 0.9636



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5s 176ms/step - aug_p: 0.0055 - d_loss: 0.0792 - g_loss: 5.4619 - gen_acc: 0.9939 - real_acc: 0.9645



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5s 176ms/step - aug_p: 0.0056 - d_loss: 0.0783 - g_loss: 5.4012 - gen_acc: 0.9940 - real_acc: 0.9654



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 4s 176ms/step - aug_p: 0.0056 - d_loss: 0.0775 - g_loss: 5.3477 - gen_acc: 0.9941 - real_acc: 0.9661



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4s 176ms/step - aug_p: 0.0056 - d_loss: 0.0768 - g_loss: 5.2979 - gen_acc: 0.9941 - real_acc: 0.9667



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4s 176ms/step - aug_p: 0.0057 - d_loss: 0.0762 - g_loss: 5.2495 - gen_acc: 0.9941 - real_acc: 0.9673



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4s 176ms/step - aug_p: 0.0057 - d_loss: 0.0758 - g_loss: 5.2113 - gen_acc: 0.9940 - real_acc: 0.9677



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4s 176ms/step - aug_p: 0.0057 - d_loss: 0.0754 - g_loss: 5.1753 - gen_acc: 0.9940 - real_acc: 0.9681



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4s 176ms/step - aug_p: 0.0058 - d_loss: 0.0752 - g_loss: 5.1387 - gen_acc: 0.9940 - real_acc: 0.9684



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3s 176ms/step - aug_p: 0.0058 - d_loss: 0.0749 - g_loss: 5.1112 - gen_acc: 0.9939 - real_acc: 0.9688



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3s 176ms/step - aug_p: 0.0058 - d_loss: 0.0746 - g_loss: 5.0899 - gen_acc: 0.9939 - real_acc: 0.9691



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3s 177ms/step - aug_p: 0.0059 - d_loss: 0.0744 - g_loss: 5.0691 - gen_acc: 0.9939 - real_acc: 0.9693



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3s 177ms/step - aug_p: 0.0059 - d_loss: 0.0743 - g_loss: 5.0465 - gen_acc: 0.9937 - real_acc: 0.9696



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3s 177ms/step - aug_p: 0.0059 - d_loss: 0.0742 - g_loss: 5.0296 - gen_acc: 0.9935 - real_acc: 0.9698



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3s 177ms/step - aug_p: 0.0060 - d_loss: 0.0741 - g_loss: 5.0163 - gen_acc: 0.9934 - real_acc: 0.9701



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2s 177ms/step - aug_p: 0.0060 - d_loss: 0.0740 - g_loss: 5.0018 - gen_acc: 0.9932 - real_acc: 0.9703



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2s 177ms/step - aug_p: 0.0060 - d_loss: 0.0739 - g_loss: 4.9862 - gen_acc: 0.9931 - real_acc: 0.9705



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2s 177ms/step - aug_p: 0.0061 - d_loss: 0.0739 - g_loss: 4.9725 - gen_acc: 0.9929 - real_acc: 0.9707



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2s 177ms/step - aug_p: 0.0061 - d_loss: 0.0739 - g_loss: 4.9583 - gen_acc: 0.9928 - real_acc: 0.9709



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2s 178ms/step - aug_p: 0.0061 - d_loss: 0.0739 - g_loss: 4.9439 - gen_acc: 0.9927 - real_acc: 0.9711



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1s 178ms/step - aug_p: 0.0062 - d_loss: 0.0739 - g_loss: 4.9297 - gen_acc: 0.9926 - real_acc: 0.9712



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1s 178ms/step - aug_p: 0.0062 - d_loss: 0.0740 - g_loss: 4.9151 - gen_acc: 0.9925 - real_acc: 0.9714



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1s 178ms/step - aug_p: 0.0062 - d_loss: 0.0741 - g_loss: 4.9027 - gen_acc: 0.9924 - real_acc: 0.9714



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1s 178ms/step - aug_p: 0.0063 - d_loss: 0.0743 - g_loss: 4.8890 - gen_acc: 0.9921 - real_acc: 0.9715



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1s 178ms/step - aug_p: 0.0063 - d_loss: 0.0748 - g_loss: 4.8802 - gen_acc: 0.9918 - real_acc: 0.9713



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1s 178ms/step - aug_p: 0.0063 - d_loss: 0.0752 - g_loss: 4.8742 - gen_acc: 0.9916 - real_acc: 0.9712



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0s 178ms/step - aug_p: 0.0064 - d_loss: 0.0756 - g_loss: 4.8685 - gen_acc: 0.9914 - real_acc: 0.9710



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0s 178ms/step - aug_p: 0.0064 - d_loss: 0.0759 - g_loss: 4.8620 - gen_acc: 0.9911 - real_acc: 0.9709



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0s 178ms/step - aug_p: 0.0064 - d_loss: 0.0762 - g_loss: 4.8555 - gen_acc: 0.9909 - real_acc: 0.9708



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0s 178ms/step - aug_p: 0.0064 - d_loss: 0.0765 - g_loss: 4.8492 - gen_acc: 0.9907 - real_acc: 0.9707



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0s 178ms/step - aug_p: 0.0065 - d_loss: 0.0768 - g_loss: 4.8424 - gen_acc: 0.9905 - real_acc: 0.9707



46/46 ━━━━━━━━━━━━━━━━━━━━ 0s 178ms/step - aug_p: 0.0065 - d_loss: 0.0771 - g_loss: 4.8357 - gen_acc: 0.9902 - real_acc: 0.9706



46/46 ━━━━━━━━━━━━━━━━━━━━ 43s 280ms/step - aug_p: 0.0065 - d_loss: 0.0774 - g_loss: 4.8293 - gen_acc: 0.9900 - real_acc: 0.9705 - val_kid: 8.8293

Epoch 3/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 5:54 8s/step - aug_p: 0.0105 - d_loss: 0.0941 - g_loss: 3.4148 - gen_acc: 0.9766 - real_acc: 0.9609



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 8s 196ms/step - aug_p: 0.0105 - d_loss: 0.0925 - g_loss: 3.3668 - gen_acc: 0.9805 - real_acc: 0.9668



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 8s 187ms/step - aug_p: 0.0106 - d_loss: 0.0918 - g_loss: 3.3820 - gen_acc: 0.9835 - real_acc: 0.9666



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7s 183ms/step - aug_p: 0.0106 - d_loss: 0.0932 - g_loss: 3.3732 - gen_acc: 0.9847 - real_acc: 0.9661



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7s 182ms/step - aug_p: 0.0106 - d_loss: 0.0941 - g_loss: 3.3531 - gen_acc: 0.9859 - real_acc: 0.9670



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7s 181ms/step - aug_p: 0.0107 - d_loss: 0.0942 - g_loss: 3.3519 - gen_acc: 0.9869 - real_acc: 0.9679



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7s 180ms/step - aug_p: 0.0107 - d_loss: 0.0941 - g_loss: 3.3467 - gen_acc: 0.9877 - real_acc: 0.9690



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6s 179ms/step - aug_p: 0.0107 - d_loss: 0.0944 - g_loss: 3.3438 - gen_acc: 0.9882 - real_acc: 0.9693



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6s 179ms/step - aug_p: 0.0107 - d_loss: 0.0947 - g_loss: 3.3384 - gen_acc: 0.9886 - real_acc: 0.9696



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6s 180ms/step - aug_p: 0.0108 - d_loss: 0.0948 - g_loss: 3.3468 - gen_acc: 0.9889 - real_acc: 0.9694



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6s 180ms/step - aug_p: 0.0108 - d_loss: 0.0952 - g_loss: 3.3443 - gen_acc: 0.9888 - real_acc: 0.9695



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6s 180ms/step - aug_p: 0.0108 - d_loss: 0.0955 - g_loss: 3.3676 - gen_acc: 0.9887 - real_acc: 0.9693



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 5s 180ms/step - aug_p: 0.0109 - d_loss: 0.0954 - g_loss: 3.3959 - gen_acc: 0.9888 - real_acc: 0.9693



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5s 180ms/step - aug_p: 0.0109 - d_loss: 0.0957 - g_loss: 3.4125 - gen_acc: 0.9883 - real_acc: 0.9694



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5s 180ms/step - aug_p: 0.0109 - d_loss: 0.0963 - g_loss: 3.4419 - gen_acc: 0.9880 - real_acc: 0.9688



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5s 180ms/step - aug_p: 0.0110 - d_loss: 0.0969 - g_loss: 3.4641 - gen_acc: 0.9876 - real_acc: 0.9684



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5s 180ms/step - aug_p: 0.0110 - d_loss: 0.0972 - g_loss: 3.4867 - gen_acc: 0.9873 - real_acc: 0.9681



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5s 180ms/step - aug_p: 0.0110 - d_loss: 0.0975 - g_loss: 3.5046 - gen_acc: 0.9869 - real_acc: 0.9679



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4s 180ms/step - aug_p: 0.0110 - d_loss: 0.0977 - g_loss: 3.5235 - gen_acc: 0.9866 - real_acc: 0.9678



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4s 180ms/step - aug_p: 0.0111 - d_loss: 0.0978 - g_loss: 3.5387 - gen_acc: 0.9863 - real_acc: 0.9677



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4s 180ms/step - aug_p: 0.0111 - d_loss: 0.0980 - g_loss: 3.5544 - gen_acc: 0.9861 - real_acc: 0.9676



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4s 179ms/step - aug_p: 0.0111 - d_loss: 0.0983 - g_loss: 3.5646 - gen_acc: 0.9857 - real_acc: 0.9675



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4s 179ms/step - aug_p: 0.0112 - d_loss: 0.0990 - g_loss: 3.5834 - gen_acc: 0.9853 - real_acc: 0.9670



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3s 179ms/step - aug_p: 0.0112 - d_loss: 0.0995 - g_loss: 3.6027 - gen_acc: 0.9850 - real_acc: 0.9665



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3s 179ms/step - aug_p: 0.0112 - d_loss: 0.1001 - g_loss: 3.6171 - gen_acc: 0.9845 - real_acc: 0.9662



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3s 179ms/step - aug_p: 0.0112 - d_loss: 0.1006 - g_loss: 3.6374 - gen_acc: 0.9840 - real_acc: 0.9659



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3s 179ms/step - aug_p: 0.0113 - d_loss: 0.1009 - g_loss: 3.6630 - gen_acc: 0.9836 - real_acc: 0.9656



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3s 179ms/step - aug_p: 0.0113 - d_loss: 0.1012 - g_loss: 3.6907 - gen_acc: 0.9833 - real_acc: 0.9654



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3s 179ms/step - aug_p: 0.0113 - d_loss: 0.1014 - g_loss: 3.7165 - gen_acc: 0.9830 - real_acc: 0.9652



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2s 179ms/step - aug_p: 0.0114 - d_loss: 0.1016 - g_loss: 3.7387 - gen_acc: 0.9827 - real_acc: 0.9651



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2s 179ms/step - aug_p: 0.0114 - d_loss: 0.1016 - g_loss: 3.7601 - gen_acc: 0.9824 - real_acc: 0.9650



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2s 179ms/step - aug_p: 0.0114 - d_loss: 0.1017 - g_loss: 3.7799 - gen_acc: 0.9822 - real_acc: 0.9649



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2s 179ms/step - aug_p: 0.0114 - d_loss: 0.1017 - g_loss: 3.7963 - gen_acc: 0.9820 - real_acc: 0.9649



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2s 179ms/step - aug_p: 0.0115 - d_loss: 0.1019 - g_loss: 3.8154 - gen_acc: 0.9818 - real_acc: 0.9647



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1s 179ms/step - aug_p: 0.0115 - d_loss: 0.1020 - g_loss: 3.8348 - gen_acc: 0.9816 - real_acc: 0.9645



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1s 179ms/step - aug_p: 0.0115 - d_loss: 0.1022 - g_loss: 3.8515 - gen_acc: 0.9813 - real_acc: 0.9644



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1s 179ms/step - aug_p: 0.0115 - d_loss: 0.1025 - g_loss: 3.8702 - gen_acc: 0.9810 - real_acc: 0.9642



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1s 179ms/step - aug_p: 0.0116 - d_loss: 0.1027 - g_loss: 3.8891 - gen_acc: 0.9807 - real_acc: 0.9640



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1s 179ms/step - aug_p: 0.0116 - d_loss: 0.1032 - g_loss: 3.9048 - gen_acc: 0.9803 - real_acc: 0.9638



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1s 179ms/step - aug_p: 0.0116 - d_loss: 0.1036 - g_loss: 3.9227 - gen_acc: 0.9799 - real_acc: 0.9636



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0s 179ms/step - aug_p: 0.0117 - d_loss: 0.1040 - g_loss: 3.9415 - gen_acc: 0.9796 - real_acc: 0.9633



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0s 179ms/step - aug_p: 0.0117 - d_loss: 0.1044 - g_loss: 3.9588 - gen_acc: 0.9792 - real_acc: 0.9631



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0s 179ms/step - aug_p: 0.0117 - d_loss: 0.1048 - g_loss: 3.9748 - gen_acc: 0.9789 - real_acc: 0.9629



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0s 179ms/step - aug_p: 0.0117 - d_loss: 0.1052 - g_loss: 3.9895 - gen_acc: 0.9785 - real_acc: 0.9627



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0s 179ms/step - aug_p: 0.0118 - d_loss: 0.1055 - g_loss: 4.0041 - gen_acc: 0.9782 - real_acc: 0.9626



46/46 ━━━━━━━━━━━━━━━━━━━━ 0s 179ms/step - aug_p: 0.0118 - d_loss: 0.1058 - g_loss: 4.0177 - gen_acc: 0.9779 - real_acc: 0.9624



46/46 ━━━━━━━━━━━━━━━━━━━━ 22s 315ms/step - aug_p: 0.0118 - d_loss: 0.1061 - g_loss: 4.0306 - gen_acc: 0.9776 - real_acc: 0.9623 - val_kid: 8.4585

Epoch 4/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 11s 263ms/step - aug_p: 0.0154 - d_loss: 0.1223 - g_loss: 2.5203 - gen_acc: 0.9688 - real_acc: 1.0000



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 7s 180ms/step - aug_p: 0.0154 - d_loss: 0.1018 - g_loss: 3.6445 - gen_acc: 0.9766 - real_acc: 0.9980



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0155 - d_loss: 0.0925 - g_loss: 4.2071 - gen_acc: 0.9809 - real_acc: 0.9926



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0155 - d_loss: 0.0875 - g_loss: 4.3535 - gen_acc: 0.9827 - real_acc: 0.9910



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0155 - d_loss: 0.0850 - g_loss: 4.3580 - gen_acc: 0.9843 - real_acc: 0.9900



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0156 - d_loss: 0.0830 - g_loss: 4.3789 - gen_acc: 0.9856 - real_acc: 0.9889



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0156 - d_loss: 0.0821 - g_loss: 4.3592 - gen_acc: 0.9864 - real_acc: 0.9879



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0156 - d_loss: 0.0814 - g_loss: 4.3377 - gen_acc: 0.9871 - real_acc: 0.9870



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0156 - d_loss: 0.0811 - g_loss: 4.3049 - gen_acc: 0.9876 - real_acc: 0.9864



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0157 - d_loss: 0.0807 - g_loss: 4.2813 - gen_acc: 0.9881 - real_acc: 0.9859



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0157 - d_loss: 0.0804 - g_loss: 4.2560 - gen_acc: 0.9884 - real_acc: 0.9855



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0157 - d_loss: 0.0804 - g_loss: 4.2284 - gen_acc: 0.9886 - real_acc: 0.9851



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0158 - d_loss: 0.0806 - g_loss: 4.2117 - gen_acc: 0.9888 - real_acc: 0.9847



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0158 - d_loss: 0.0807 - g_loss: 4.1925 - gen_acc: 0.9887 - real_acc: 0.9844



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0158 - d_loss: 0.0807 - g_loss: 4.1790 - gen_acc: 0.9888 - real_acc: 0.9841



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0159 - d_loss: 0.0807 - g_loss: 4.1631 - gen_acc: 0.9887 - real_acc: 0.9840



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0159 - d_loss: 0.0807 - g_loss: 4.1518 - gen_acc: 0.9887 - real_acc: 0.9838



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0159 - d_loss: 0.0807 - g_loss: 4.1398 - gen_acc: 0.9887 - real_acc: 0.9837



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0160 - d_loss: 0.0809 - g_loss: 4.1255 - gen_acc: 0.9887 - real_acc: 0.9837



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0160 - d_loss: 0.0810 - g_loss: 4.1171 - gen_acc: 0.9887 - real_acc: 0.9835



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0160 - d_loss: 0.0810 - g_loss: 4.1116 - gen_acc: 0.9886 - real_acc: 0.9835



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0161 - d_loss: 0.0811 - g_loss: 4.1037 - gen_acc: 0.9885 - real_acc: 0.9834



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0161 - d_loss: 0.0812 - g_loss: 4.1013 - gen_acc: 0.9885 - real_acc: 0.9833



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0161 - d_loss: 0.0813 - g_loss: 4.1000 - gen_acc: 0.9884 - real_acc: 0.9832



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0162 - d_loss: 0.0814 - g_loss: 4.0967 - gen_acc: 0.9883 - real_acc: 0.9832



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0162 - d_loss: 0.0815 - g_loss: 4.0951 - gen_acc: 0.9882 - real_acc: 0.9831



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0162 - d_loss: 0.0815 - g_loss: 4.0930 - gen_acc: 0.9882 - real_acc: 0.9830



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0163 - d_loss: 0.0817 - g_loss: 4.0887 - gen_acc: 0.9880 - real_acc: 0.9830



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0163 - d_loss: 0.0818 - g_loss: 4.0890 - gen_acc: 0.9879 - real_acc: 0.9829



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0163 - d_loss: 0.0819 - g_loss: 4.0918 - gen_acc: 0.9878 - real_acc: 0.9828



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0164 - d_loss: 0.0821 - g_loss: 4.0923 - gen_acc: 0.9877 - real_acc: 0.9826



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0164 - d_loss: 0.0823 - g_loss: 4.0957 - gen_acc: 0.9876 - real_acc: 0.9826



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0164 - d_loss: 0.0824 - g_loss: 4.1014 - gen_acc: 0.9874 - real_acc: 0.9825



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0165 - d_loss: 0.0824 - g_loss: 4.1072 - gen_acc: 0.9873 - real_acc: 0.9824



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 180毫秒/步 - aug_p: 0.0165 - d_loss: 0.0825 - g_loss: 4.1116 - gen_acc: 0.9872 - real_acc: 0.9823



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 180毫秒/步 - aug_p: 0.0165 - d_loss: 0.0826 - g_loss: 4.1168 - gen_acc: 0.9871 - real_acc: 0.9822



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 181毫秒/步 - aug_p: 0.0166 - d_loss: 0.0827 - g_loss: 4.1217 - gen_acc: 0.9870 - real_acc: 0.9822



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0166 - d_loss: 0.0827 - g_loss: 4.1262 - gen_acc: 0.9869 - real_acc: 0.9821



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0166 - d_loss: 0.0828 - g_loss: 4.1298 - gen_acc: 0.9868 - real_acc: 0.9820



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1秒 180毫秒/步 - aug_p: 0.0167 - d_loss: 0.0828 - g_loss: 4.1333 - gen_acc: 0.9868 - real_acc: 0.9820



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0秒 181毫秒/步 - aug_p: 0.0167 - d_loss: 0.0828 - g_loss: 4.1361 - gen_acc: 0.9867 - real_acc: 0.9819



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 181毫秒/步 - aug_p: 0.0167 - d_loss: 0.0828 - g_loss: 4.1389 - gen_acc: 0.9866 - real_acc: 0.9819



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 181毫秒/步 - aug_p: 0.0168 - d_loss: 0.0828 - g_loss: 4.1408 - gen_acc: 0.9866 - real_acc: 0.9819



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 181毫秒/步 - aug_p: 0.0168 - d_loss: 0.0828 - g_loss: 4.1438 - gen_acc: 0.9865 - real_acc: 0.9818



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 181毫秒/步 - aug_p: 0.0168 - d_loss: 0.0828 - g_loss: 4.1466 - gen_acc: 0.9865 - real_acc: 0.9818



46/46 ━━━━━━━━━━━━━━━━━━━━ 0秒 181毫秒/步 - aug_p: 0.0168 - d_loss: 0.0829 - g_loss: 4.1480 - gen_acc: 0.9864 - real_acc: 0.9818



46/46 ━━━━━━━━━━━━━━━━━━━━ 14秒 316毫秒/步 - aug_p: 0.0169 - d_loss: 0.0829 - g_loss: 4.1493 - gen_acc: 0.9863 - real_acc: 0.9817 - val_kid: 6.6764

Epoch 5/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 10秒 237毫秒/步 - aug_p: 0.0212 - d_loss: 0.3046 - g_loss: 11.2403 - gen_acc: 1.0000 - real_acc: 0.7734



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 8秒 197毫秒/步 - aug_p: 0.0212 - d_loss: 0.2549 - g_loss: 10.8464 - gen_acc: 1.0000 - real_acc: 0.8086



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 8秒 190毫秒/步 - aug_p: 0.0212 - d_loss: 0.2217 - g_loss: 10.0394 - gen_acc: 0.9983 - real_acc: 0.8359



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 187毫秒/步 - aug_p: 0.0212 - d_loss: 0.2183 - g_loss: 9.2019 - gen_acc: 0.9753 - real_acc: 0.8560



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 185毫秒/步 - aug_p: 0.0212 - d_loss: 0.2125 - g_loss: 8.8056 - gen_acc: 0.9652 - real_acc: 0.8676



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 185毫秒/步 - aug_p: 0.0213 - d_loss: 0.2060 - g_loss: 8.5755 - gen_acc: 0.9606 - real_acc: 0.8755



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7秒 184毫秒/步 - aug_p: 0.0213 - d_loss: 0.1995 - g_loss: 8.3695 - gen_acc: 0.9579 - real_acc: 0.8823



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7秒 184毫秒/步 - aug_p: 0.0213 - d_loss: 0.1938 - g_loss: 8.1574 - gen_acc: 0.9561 - real_acc: 0.8884



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 184毫秒/步 - aug_p: 0.0214 - d_loss: 0.1881 - g_loss: 7.9590 - gen_acc: 0.9552 - real_acc: 0.8939



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 184毫秒/步 - aug_p: 0.0214 - d_loss: 0.1827 - g_loss: 7.7719 - gen_acc: 0.9550 - real_acc: 0.8989



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 183毫秒/步 - aug_p: 0.0214 - d_loss: 0.1785 - g_loss: 7.5867 - gen_acc: 0.9546 - real_acc: 0.9034



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 183毫秒/步 - aug_p: 0.0214 - d_loss: 0.1748 - g_loss: 7.4330 - gen_acc: 0.9546 - real_acc: 0.9072



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 183毫秒/步 - aug_p: 0.0215 - d_loss: 0.1717 - g_loss: 7.2895 - gen_acc: 0.9548 - real_acc: 0.9103



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 183毫秒/步 - aug_p: 0.0215 - d_loss: 0.1693 - g_loss: 7.1489 - gen_acc: 0.9544 - real_acc: 0.9132



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 182毫秒/步 - aug_p: 0.0215 - d_loss: 0.1674 - g_loss: 7.0344 - gen_acc: 0.9543 - real_acc: 0.9153



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 182毫秒/步 - aug_p: 0.0215 - d_loss: 0.1654 - g_loss: 6.9321 - gen_acc: 0.9544 - real_acc: 0.9173



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 182毫秒/步 - aug_p: 0.0216 - d_loss: 0.1637 - g_loss: 6.8304 - gen_acc: 0.9541 - real_acc: 0.9191



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 183毫秒/步 - aug_p: 0.0216 - d_loss: 0.1620 - g_loss: 6.7449 - gen_acc: 0.9540 - real_acc: 0.9209



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 183毫秒/步 - aug_p: 0.0216 - d_loss: 0.1603 - g_loss: 6.6702 - gen_acc: 0.9540 - real_acc: 0.9225



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0217 - d_loss: 0.1587 - g_loss: 6.5977 - gen_acc: 0.9541 - real_acc: 0.9240



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0217 - d_loss: 0.1572 - g_loss: 6.5271 - gen_acc: 0.9542 - real_acc: 0.9255



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0217 - d_loss: 0.1556 - g_loss: 6.4626 - gen_acc: 0.9544 - real_acc: 0.9269



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0217 - d_loss: 0.1540 - g_loss: 6.4028 - gen_acc: 0.9546 - real_acc: 0.9282



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0218 - d_loss: 0.1525 - g_loss: 6.3440 - gen_acc: 0.9548 - real_acc: 0.9295



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 183毫秒/步 - aug_p: 0.0218 - d_loss: 0.1510 - g_loss: 6.2898 - gen_acc: 0.9551 - real_acc: 0.9307



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 183毫秒/步 - aug_p: 0.0218 - d_loss: 0.1495 - g_loss: 6.2380 - gen_acc: 0.9554 - real_acc: 0.9318



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 183毫秒/步 - aug_p: 0.0219 - d_loss: 0.1481 - g_loss: 6.1880 - gen_acc: 0.9558 - real_acc: 0.9330



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 182毫秒/步 - aug_p: 0.0219 - d_loss: 0.1468 - g_loss: 6.1413 - gen_acc: 0.9561 - real_acc: 0.9340



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 183毫秒/步 - aug_p: 0.0219 - d_loss: 0.1454 - g_loss: 6.0966 - gen_acc: 0.9565 - real_acc: 0.9350



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 182毫秒/步 - aug_p: 0.0220 - d_loss: 0.1441 - g_loss: 6.0534 - gen_acc: 0.9569 - real_acc: 0.9360



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 183毫秒/步 - aug_p: 0.0220 - d_loss: 0.1428 - g_loss: 6.0122 - gen_acc: 0.9573 - real_acc: 0.9370



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 182毫秒/步 - aug_p: 0.0220 - d_loss: 0.1415 - g_loss: 5.9738 - gen_acc: 0.9577 - real_acc: 0.9379



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 182毫秒/步 - aug_p: 0.0220 - d_loss: 0.1403 - g_loss: 5.9369 - gen_acc: 0.9581 - real_acc: 0.9388



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 182毫秒/步 - aug_p: 0.0221 - d_loss: 0.1390 - g_loss: 5.9020 - gen_acc: 0.9585 - real_acc: 0.9396



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 2秒 182毫秒/步 - aug_p: 0.0221 - d_loss: 0.1378 - g_loss: 5.8680 - gen_acc: 0.9589 - real_acc: 0.9404



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 182毫秒/步 - aug_p: 0.0221 - d_loss: 0.1366 - g_loss: 5.8355 - gen_acc: 0.9592 - real_acc: 0.9412



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 182毫秒/步 - aug_p: 0.0222 - d_loss: 0.1355 - g_loss: 5.8042 - gen_acc: 0.9596 - real_acc: 0.9420



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 182毫秒/步 - aug_p: 0.0222 - d_loss: 0.1344 - g_loss: 5.7737 - gen_acc: 0.9600 - real_acc: 0.9427



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 182毫秒/步 - aug_p: 0.0222 - d_loss: 0.1333 - g_loss: 5.7447 - gen_acc: 0.9604 - real_acc: 0.9434



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1秒 182毫秒/步 - aug_p: 0.0223 - d_loss: 0.1323 - g_loss: 5.7161 - gen_acc: 0.9608 - real_acc: 0.9441



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0秒 182毫秒/步 - aug_p: 0.0223 - d_loss: 0.1313 - g_loss: 5.6892 - gen_acc: 0.9611 - real_acc: 0.9447



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 182毫秒/步 - aug_p: 0.0223 - d_loss: 0.1304 - g_loss: 5.6621 - gen_acc: 0.9615 - real_acc: 0.9453



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 182毫秒/步 - aug_p: 0.0223 - d_loss: 0.1296 - g_loss: 5.6390 - gen_acc: 0.9618 - real_acc: 0.9458



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 182毫秒/步 - aug_p: 0.0224 - d_loss: 0.1288 - g_loss: 5.6185 - gen_acc: 0.9621 - real_acc: 0.9463



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 182毫秒/步 - aug_p: 0.0224 - d_loss: 0.1280 - g_loss: 5.5982 - gen_acc: 0.9623 - real_acc: 0.9468



46/46 ━━━━━━━━━━━━━━━━━━━━ 0秒 182毫秒/步 - aug_p: 0.0224 - d_loss: 0.1273 - g_loss: 5.5795 - gen_acc: 0.9626 - real_acc: 0.9473



46/46 ━━━━━━━━━━━━━━━━━━━━ 14秒 317毫秒/步 - aug_p: 0.0225 - d_loss: 0.1265 - g_loss: 5.5616 - gen_acc: 0.9629 - real_acc: 0.9478 - val_kid: 4.7496

Epoch 6/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 10秒 236毫秒/步 - aug_p: 0.0268 - d_loss: 0.0745 - g_loss: 5.1780 - gen_acc: 0.9922 - real_acc: 0.9688



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 8秒 184毫秒/步 - aug_p: 0.0269 - d_loss: 0.0774 - g_loss: 4.5412 - gen_acc: 0.9883 - real_acc: 0.9766



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 183毫秒/步 - aug_p: 0.0269 - d_loss: 0.0743 - g_loss: 4.5406 - gen_acc: 0.9887 - real_acc: 0.9783



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 184毫秒/步 - aug_p: 0.0269 - d_loss: 0.0724 - g_loss: 4.5764 - gen_acc: 0.9896 - real_acc: 0.9779



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 183毫秒/步 - aug_p: 0.0270 - d_loss: 0.0732 - g_loss: 4.5209 - gen_acc: 0.9882 - real_acc: 0.9785



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 182毫秒/步 - aug_p: 0.0270 - d_loss: 0.0738 - g_loss: 4.5449 - gen_acc: 0.9878 - real_acc: 0.9782



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7秒 182毫秒/步 - aug_p: 0.0270 - d_loss: 0.0747 - g_loss: 4.5880 - gen_acc: 0.9878 - real_acc: 0.9769



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0271 - d_loss: 0.0766 - g_loss: 4.5791 - gen_acc: 0.9857 - real_acc: 0.9763



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 183毫秒/步 - aug_p: 0.0271 - d_loss: 0.0777 - g_loss: 4.6269 - gen_acc: 0.9844 - real_acc: 0.9757



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0271 - d_loss: 0.0786 - g_loss: 4.7075 - gen_acc: 0.9836 - real_acc: 0.9749



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0271 - d_loss: 0.0792 - g_loss: 4.7786 - gen_acc: 0.9826 - real_acc: 0.9745



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0272 - d_loss: 0.0793 - g_loss: 4.8440 - gen_acc: 0.9820 - real_acc: 0.9744



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 183毫秒/步 - aug_p: 0.0272 - d_loss: 0.0792 - g_loss: 4.9001 - gen_acc: 0.9816 - real_acc: 0.9744



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 182毫秒/步 - aug_p: 0.0272 - d_loss: 0.0789 - g_loss: 4.9354 - gen_acc: 0.9814 - real_acc: 0.9745



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 182毫秒/步 - aug_p: 0.0273 - d_loss: 0.0785 - g_loss: 4.9643 - gen_acc: 0.9813 - real_acc: 0.9747



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 182毫秒/步 - aug_p: 0.0273 - d_loss: 0.0781 - g_loss: 4.9864 - gen_acc: 0.9814 - real_acc: 0.9749



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 182毫秒/步 - aug_p: 0.0273 - d_loss: 0.0778 - g_loss: 4.9973 - gen_acc: 0.9814 - real_acc: 0.9751



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 182毫秒/步 - aug_p: 0.0274 - d_loss: 0.0774 - g_loss: 5.0125 - gen_acc: 0.9815 - real_acc: 0.9753



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0274 - d_loss: 0.0770 - g_loss: 5.0280 - gen_acc: 0.9816 - real_acc: 0.9755



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0274 - d_loss: 0.0765 - g_loss: 5.0398 - gen_acc: 0.9818 - real_acc: 0.9757



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0275 - d_loss: 0.0760 - g_loss: 5.0455 - gen_acc: 0.9819 - real_acc: 0.9759



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 182毫秒/步 - aug_p: 0.0275 - d_loss: 0.0756 - g_loss: 5.0535 - gen_acc: 0.9820 - real_acc: 0.9760



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4秒 181毫秒/步 - aug_p: 0.0275 - d_loss: 0.0752 - g_loss: 5.0590 - gen_acc: 0.9822 - real_acc: 0.9762



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0276 - d_loss: 0.0749 - g_loss: 5.0595 - gen_acc: 0.9823 - real_acc: 0.9763



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0276 - d_loss: 0.0746 - g_loss: 5.0650 - gen_acc: 0.9825 - real_acc: 0.9764



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0276 - d_loss: 0.0743 - g_loss: 5.0742 - gen_acc: 0.9826 - real_acc: 0.9765



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0277 - d_loss: 0.0740 - g_loss: 5.0823 - gen_acc: 0.9828 - real_acc: 0.9766



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0277 - d_loss: 0.0737 - g_loss: 5.0871 - gen_acc: 0.9829 - real_acc: 0.9767



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0277 - d_loss: 0.0734 - g_loss: 5.0913 - gen_acc: 0.9831 - real_acc: 0.9768



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0278 - d_loss: 0.0731 - g_loss: 5.0957 - gen_acc: 0.9832 - real_acc: 0.9769



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0278 - d_loss: 0.0727 - g_loss: 5.0986 - gen_acc: 0.9834 - real_acc: 0.9770



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0278 - d_loss: 0.0725 - g_loss: 5.0992 - gen_acc: 0.9835 - real_acc: 0.9771



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0278 - d_loss: 0.0722 - g_loss: 5.1012 - gen_acc: 0.9836 - real_acc: 0.9772



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0279 - d_loss: 0.0719 - g_loss: 5.1022 - gen_acc: 0.9838 - real_acc: 0.9773



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 181毫秒/步 - aug_p: 0.0279 - d_loss: 0.0718 - g_loss: 5.1007 - gen_acc: 0.9838 - real_acc: 0.9773



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 181毫秒/步 - aug_p: 0.0279 - d_loss: 0.0717 - g_loss: 5.1026 - gen_acc: 0.9839 - real_acc: 0.9773



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 181毫秒/步 - aug_p: 0.0280 - d_loss: 0.0716 - g_loss: 5.1070 - gen_acc: 0.9840 - real_acc: 0.9772



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 181毫秒/步 - aug_p: 0.0280 - d_loss: 0.0715 - g_loss: 5.1124 - gen_acc: 0.9840 - real_acc: 0.9772



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 181毫秒/步 - aug_p: 0.0280 - d_loss: 0.0714 - g_loss: 5.1178 - gen_acc: 0.9841 - real_acc: 0.9773



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1秒 181毫秒/步 - aug_p: 0.0281 - d_loss: 0.0712 - g_loss: 5.1221 - gen_acc: 0.9842 - real_acc: 0.9773



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0秒 181毫秒/步 - aug_p: 0.0281 - d_loss: 0.0710 - g_loss: 5.1258 - gen_acc: 0.9843 - real_acc: 0.9773



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 181毫秒/步 - aug_p: 0.0281 - d_loss: 0.0708 - g_loss: 5.1290 - gen_acc: 0.9843 - real_acc: 0.9773



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 181毫秒/步 - aug_p: 0.0282 - d_loss: 0.0707 - g_loss: 5.1315 - gen_acc: 0.9844 - real_acc: 0.9774



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 181毫秒/步 - aug_p: 0.0282 - d_loss: 0.0705 - g_loss: 5.1332 - gen_acc: 0.9845 - real_acc: 0.9774



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 181毫秒/步 - aug_p: 0.0282 - d_loss: 0.0703 - g_loss: 5.1347 - gen_acc: 0.9845 - real_acc: 0.9775



46/46 ━━━━━━━━━━━━━━━━━━━━ 0秒 181毫秒/步 - aug_p: 0.0283 - d_loss: 0.0701 - g_loss: 5.1357 - gen_acc: 0.9846 - real_acc: 0.9775



46/46 ━━━━━━━━━━━━━━━━━━━━ 12秒 267毫秒/步 - aug_p: 0.0283 - d_loss: 0.0699 - g_loss: 5.1367 - gen_acc: 0.9846 - real_acc: 0.9776 - val_kid: 6.2893

Epoch 7/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 7秒 174毫秒/步 - aug_p: 0.0328 - d_loss: 0.0456 - g_loss: 3.5202 - gen_acc: 1.0000 - real_acc: 1.0000



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 7秒 179毫秒/步 - aug_p: 0.0329 - d_loss: 0.0466 - g_loss: 3.7961 - gen_acc: 0.9980 - real_acc: 0.9980



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 179毫秒/步 - aug_p: 0.0329 - d_loss: 0.0471 - g_loss: 3.9462 - gen_acc: 0.9970 - real_acc: 0.9961



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 179毫秒/步 - aug_p: 0.0329 - d_loss: 0.0469 - g_loss: 4.0184 - gen_acc: 0.9967 - real_acc: 0.9946



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0330 - d_loss: 0.0463 - g_loss: 4.0670 - gen_acc: 0.9968 - real_acc: 0.9941



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0330 - d_loss: 0.0458 - g_loss: 4.1012 - gen_acc: 0.9969 - real_acc: 0.9938



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0330 - d_loss: 0.0451 - g_loss: 4.1240 - gen_acc: 0.9970 - real_acc: 0.9937



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0331 - d_loss: 0.0444 - g_loss: 4.1347 - gen_acc: 0.9971 - real_acc: 0.9938



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0331 - d_loss: 0.0438 - g_loss: 4.1433 - gen_acc: 0.9971 - real_acc: 0.9937



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0331 - d_loss: 0.0434 - g_loss: 4.1560 - gen_acc: 0.9970 - real_acc: 0.9936



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0332 - d_loss: 0.0431 - g_loss: 4.1654 - gen_acc: 0.9969 - real_acc: 0.9936



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0332 - d_loss: 0.0429 - g_loss: 4.1695 - gen_acc: 0.9969 - real_acc: 0.9935



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0333 - d_loss: 0.0428 - g_loss: 4.1758 - gen_acc: 0.9969 - real_acc: 0.9934



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0333 - d_loss: 0.0427 - g_loss: 4.1789 - gen_acc: 0.9969 - real_acc: 0.9932



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0333 - d_loss: 0.0426 - g_loss: 4.1799 - gen_acc: 0.9970 - real_acc: 0.9929



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0334 - d_loss: 0.0425 - g_loss: 4.1823 - gen_acc: 0.9970 - real_acc: 0.9927



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0334 - d_loss: 0.0425 - g_loss: 4.1836 - gen_acc: 0.9970 - real_acc: 0.9926



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0334 - d_loss: 0.0425 - g_loss: 4.1854 - gen_acc: 0.9971 - real_acc: 0.9923



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0335 - d_loss: 0.0426 - g_loss: 4.1843 - gen_acc: 0.9971 - real_acc: 0.9921



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0335 - d_loss: 0.0427 - g_loss: 4.1873 - gen_acc: 0.9971 - real_acc: 0.9920



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0335 - d_loss: 0.0427 - g_loss: 4.1927 - gen_acc: 0.9972 - real_acc: 0.9918



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0336 - d_loss: 0.0428 - g_loss: 4.1952 - gen_acc: 0.9972 - real_acc: 0.9916



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0336 - d_loss: 0.0428 - g_loss: 4.2017 - gen_acc: 0.9972 - real_acc: 0.9915



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0336 - d_loss: 0.0428 - g_loss: 4.2106 - gen_acc: 0.9972 - real_acc: 0.9914



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0337 - d_loss: 0.0428 - g_loss: 4.2181 - gen_acc: 0.9972 - real_acc: 0.9913



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0337 - d_loss: 0.0428 - g_loss: 4.2229 - gen_acc: 0.9972 - real_acc: 0.9912



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0337 - d_loss: 0.0429 - g_loss: 4.2318 - gen_acc: 0.9972 - real_acc: 0.9911



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0338 - d_loss: 0.0429 - g_loss: 4.2416 - gen_acc: 0.9972 - real_acc: 0.9910



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0338 - d_loss: 0.0430 - g_loss: 4.2491 - gen_acc: 0.9971 - real_acc: 0.9909



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0338 - d_loss: 0.0430 - g_loss: 4.2604 - gen_acc: 0.9971 - real_acc: 0.9908



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0339 - d_loss: 0.0431 - g_loss: 4.2736 - gen_acc: 0.9971 - real_acc: 0.9907



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0339 - d_loss: 0.0432 - g_loss: 4.2834 - gen_acc: 0.9970 - real_acc: 0.9906



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0339 - d_loss: 0.0439 - g_loss: 4.3010 - gen_acc: 0.9968 - real_acc: 0.9901



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0340 - d_loss: 0.0444 - g_loss: 4.3187 - gen_acc: 0.9967 - real_acc: 0.9897



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 180毫秒/步 - aug_p: 0.0340 - d_loss: 0.0455 - g_loss: 4.3319 - gen_acc: 0.9961 - real_acc: 0.9892



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 180毫秒/步 - aug_p: 0.0340 - d_loss: 0.0464 - g_loss: 4.3508 - gen_acc: 0.9956 - real_acc: 0.9889



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0341 - d_loss: 0.0474 - g_loss: 4.3765 - gen_acc: 0.9951 - real_acc: 0.9884



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0341 - d_loss: 0.0483 - g_loss: 4.4070 - gen_acc: 0.9947 - real_acc: 0.9880



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0341 - d_loss: 0.0492 - g_loss: 4.4400 - gen_acc: 0.9943 - real_acc: 0.9875



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1秒 180毫秒/步 - aug_p: 0.0342 - d_loss: 0.0499 - g_loss: 4.4739 - gen_acc: 0.9939 - real_acc: 0.9872



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0秒 180毫秒/步 - aug_p: 0.0342 - d_loss: 0.0506 - g_loss: 4.5070 - gen_acc: 0.9935 - real_acc: 0.9868



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 180毫秒/步 - aug_p: 0.0342 - d_loss: 0.0513 - g_loss: 4.5375 - gen_acc: 0.9932 - real_acc: 0.9865



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 180毫秒/步 - aug_p: 0.0343 - d_loss: 0.0519 - g_loss: 4.5646 - gen_acc: 0.9929 - real_acc: 0.9862



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 180毫秒/步 - aug_p: 0.0343 - d_loss: 0.0525 - g_loss: 4.5904 - gen_acc: 0.9925 - real_acc: 0.9859



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 180毫秒/步 - aug_p: 0.0343 - d_loss: 0.0530 - g_loss: 4.6149 - gen_acc: 0.9923 - real_acc: 0.9857



46/46 ━━━━━━━━━━━━━━━━━━━━ 0秒 180毫秒/步 - aug_p: 0.0344 - d_loss: 0.0536 - g_loss: 4.6368 - gen_acc: 0.9920 - real_acc: 0.9854



46/46 ━━━━━━━━━━━━━━━━━━━━ 13秒 294毫秒/步 - aug_p: 0.0344 - d_loss: 0.0542 - g_loss: 4.6579 - gen_acc: 0.9917 - real_acc: 0.9852 - val_kid: 6.7378

Epoch 8/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 7秒 167毫秒/步 - aug_p: 0.0384 - d_loss: 0.1191 - g_loss: 4.3279 - gen_acc: 1.0000 - real_acc: 0.9219



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 7秒 179毫秒/步 - aug_p: 0.0384 - d_loss: 0.1470 - g_loss: 3.7525 - gen_acc: 0.9590 - real_acc: 0.9219



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0384 - d_loss: 0.1768 - g_loss: 4.0819 - gen_acc: 0.9544 - real_acc: 0.8950



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0384 - d_loss: 0.1801 - g_loss: 4.1693 - gen_acc: 0.9551 - real_acc: 0.8910



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0384 - d_loss: 0.1829 - g_loss: 4.1280 - gen_acc: 0.9491 - real_acc: 0.8934



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0384 - d_loss: 0.1828 - g_loss: 4.2346 - gen_acc: 0.9471 - real_acc: 0.8949



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7秒 180毫秒/步 - aug_p: 0.0385 - d_loss: 0.1806 - g_loss: 4.3823 - gen_acc: 0.9470 - real_acc: 0.8968



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0385 - d_loss: 0.1765 - g_loss: 4.5079 - gen_acc: 0.9478 - real_acc: 0.8997



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0385 - d_loss: 0.1723 - g_loss: 4.5814 - gen_acc: 0.9486 - real_acc: 0.9028



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0385 - d_loss: 0.1679 - g_loss: 4.6213 - gen_acc: 0.9496 - real_acc: 0.9061



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0385 - d_loss: 0.1637 - g_loss: 4.6466 - gen_acc: 0.9507 - real_acc: 0.9092



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 180毫秒/步 - aug_p: 0.0386 - d_loss: 0.1595 - g_loss: 4.6599 - gen_acc: 0.9520 - real_acc: 0.9122



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0386 - d_loss: 0.1561 - g_loss: 4.6625 - gen_acc: 0.9531 - real_acc: 0.9148



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0386 - d_loss: 0.1535 - g_loss: 4.6513 - gen_acc: 0.9537 - real_acc: 0.9172



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0386 - d_loss: 0.1530 - g_loss: 4.6600 - gen_acc: 0.9544 - real_acc: 0.9175



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0387 - d_loss: 0.1612 - g_loss: 4.6490 - gen_acc: 0.9512 - real_acc: 0.9180



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0387 - d_loss: 0.1727 - g_loss: 4.6715 - gen_acc: 0.9488 - real_acc: 0.9157



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0387 - d_loss: 0.1825 - g_loss: 4.7072 - gen_acc: 0.9469 - real_acc: 0.9135



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0387 - d_loss: 0.1904 - g_loss: 4.7428 - gen_acc: 0.9454 - real_acc: 0.9118



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 179毫秒/步 - aug_p: 0.0387 - d_loss: 0.1970 - g_loss: 4.7693 - gen_acc: 0.9440 - real_acc: 0.9106



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 179毫秒/步 - aug_p: 0.0387 - d_loss: 0.2029 - g_loss: 4.7854 - gen_acc: 0.9424 - real_acc: 0.9098



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0387 - d_loss: 0.2079 - g_loss: 4.7960 - gen_acc: 0.9409 - real_acc: 0.9092



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0388 - d_loss: 0.2119 - g_loss: 4.8033 - gen_acc: 0.9397 - real_acc: 0.9090



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 179毫秒/步 - aug_p: 0.0388 - d_loss: 0.2153 - g_loss: 4.8076 - gen_acc: 0.9387 - real_acc: 0.9088



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 179毫秒/步 - aug_p: 0.0388 - d_loss: 0.2182 - g_loss: 4.8077 - gen_acc: 0.9378 - real_acc: 0.9087



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 179毫秒/步 - aug_p: 0.0388 - d_loss: 0.2207 - g_loss: 4.8051 - gen_acc: 0.9371 - real_acc: 0.9087



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 179毫秒/步 - aug_p: 0.0388 - d_loss: 0.2229 - g_loss: 4.8007 - gen_acc: 0.9365 - real_acc: 0.9086



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0388 - d_loss: 0.2249 - g_loss: 4.7934 - gen_acc: 0.9360 - real_acc: 0.9086



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0388 - d_loss: 0.2265 - g_loss: 4.7860 - gen_acc: 0.9355 - real_acc: 0.9086



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0389 - d_loss: 0.2278 - g_loss: 4.7775 - gen_acc: 0.9352 - real_acc: 0.9087



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0389 - d_loss: 0.2290 - g_loss: 4.7677 - gen_acc: 0.9349 - real_acc: 0.9087



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0389 - d_loss: 0.2299 - g_loss: 4.7575 - gen_acc: 0.9347 - real_acc: 0.9089



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0389 - d_loss: 0.2305 - g_loss: 4.7470 - gen_acc: 0.9346 - real_acc: 0.9091



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0389 - d_loss: 0.2310 - g_loss: 4.7363 - gen_acc: 0.9345 - real_acc: 0.9093



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 180毫秒/步 - aug_p: 0.0389 - d_loss: 0.2314 - g_loss: 4.7249 - gen_acc: 0.9344 - real_acc: 0.9095



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 180毫秒/步 - aug_p: 0.0389 - d_loss: 0.2317 - g_loss: 4.7149 - gen_acc: 0.9344 - real_acc: 0.9098



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0390 - d_loss: 0.2319 - g_loss: 4.7045 - gen_acc: 0.9345 - real_acc: 0.9101



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0390 - d_loss: 0.2319 - g_loss: 4.6937 - gen_acc: 0.9345 - real_acc: 0.9104



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 179毫秒/步 - aug_p: 0.0390 - d_loss: 0.2319 - g_loss: 4.6838 - gen_acc: 0.9346 - real_acc: 0.9107



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1秒 179毫秒/步 - aug_p: 0.0390 - d_loss: 0.2318 - g_loss: 4.6734 - gen_acc: 0.9347 - real_acc: 0.9110



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0秒 179毫秒/步 - aug_p: 0.0390 - d_loss: 0.2316 - g_loss: 4.6636 - gen_acc: 0.9349 - real_acc: 0.9114



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 179毫秒/步 - aug_p: 0.0390 - d_loss: 0.2313 - g_loss: 4.6532 - gen_acc: 0.9350 - real_acc: 0.9117



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 180毫秒/步 - aug_p: 0.0391 - d_loss: 0.2310 - g_loss: 4.6442 - gen_acc: 0.9352 - real_acc: 0.9120



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 180毫秒/步 - aug_p: 0.0391 - d_loss: 0.2306 - g_loss: 4.6361 - gen_acc: 0.9354 - real_acc: 0.9124



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 180毫秒/步 - aug_p: 0.0391 - d_loss: 0.2302 - g_loss: 4.6279 - gen_acc: 0.9356 - real_acc: 0.9127



46/46 ━━━━━━━━━━━━━━━━━━━━ 0秒 180毫秒/步 - aug_p: 0.0391 - d_loss: 0.2297 - g_loss: 4.6201 - gen_acc: 0.9358 - real_acc: 0.9131



46/46 ━━━━━━━━━━━━━━━━━━━━ 12秒 266毫秒/步 - aug_p: 0.0391 - d_loss: 0.2292 - g_loss: 4.6126 - gen_acc: 0.9361 - real_acc: 0.9134 - val_kid: 5.7109

Epoch 9/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 8秒 180毫秒/步 - aug_p: 0.0422 - d_loss: 0.0668 - g_loss: 3.8939 - gen_acc: 0.9922 - real_acc: 0.9922



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 8秒 187毫秒/步 - aug_p: 0.0422 - d_loss: 0.0676 - g_loss: 3.8295 - gen_acc: 0.9863 - real_acc: 0.9941



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 185毫秒/步 - aug_p: 0.0422 - d_loss: 0.0659 - g_loss: 3.8676 - gen_acc: 0.9865 - real_acc: 0.9944



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 183毫秒/步 - aug_p: 0.0423 - d_loss: 0.0703 - g_loss: 3.8084 - gen_acc: 0.9831 - real_acc: 0.9928



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 183毫秒/步 - aug_p: 0.0423 - d_loss: 0.0755 - g_loss: 3.9384 - gen_acc: 0.9821 - real_acc: 0.9880



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 183毫秒/步 - aug_p: 0.0423 - d_loss: 0.0781 - g_loss: 4.0291 - gen_acc: 0.9818 - real_acc: 0.9846



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7秒 182毫秒/步 - aug_p: 0.0424 - d_loss: 0.0831 - g_loss: 4.0366 - gen_acc: 0.9779 - real_acc: 0.9828



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0424 - d_loss: 0.0888 - g_loss: 4.1585 - gen_acc: 0.9756 - real_acc: 0.9782



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0424 - d_loss: 0.0921 - g_loss: 4.3258 - gen_acc: 0.9744 - real_acc: 0.9750



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0424 - d_loss: 0.0937 - g_loss: 4.4967 - gen_acc: 0.9737 - real_acc: 0.9729



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0425 - d_loss: 0.0944 - g_loss: 4.6444 - gen_acc: 0.9732 - real_acc: 0.9715



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 181毫秒/步 - aug_p: 0.0425 - d_loss: 0.0945 - g_loss: 4.7625 - gen_acc: 0.9730 - real_acc: 0.9706



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0425 - d_loss: 0.0943 - g_loss: 4.8487 - gen_acc: 0.9728 - real_acc: 0.9701



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0425 - d_loss: 0.0940 - g_loss: 4.9110 - gen_acc: 0.9726 - real_acc: 0.9698



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0426 - d_loss: 0.0935 - g_loss: 4.9645 - gen_acc: 0.9725 - real_acc: 0.9696



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0426 - d_loss: 0.0931 - g_loss: 5.0047 - gen_acc: 0.9726 - real_acc: 0.9694



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0426 - d_loss: 0.0930 - g_loss: 5.0287 - gen_acc: 0.9723 - real_acc: 0.9693



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0426 - d_loss: 0.0941 - g_loss: 5.0578 - gen_acc: 0.9721 - real_acc: 0.9679



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 181毫秒/步 - aug_p: 0.0427 - d_loss: 0.0976 - g_loss: 5.0709 - gen_acc: 0.9702 - real_acc: 0.9669



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 181毫秒/步 - aug_p: 0.0427 - d_loss: 0.1023 - g_loss: 5.0961 - gen_acc: 0.9687 - real_acc: 0.9645



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 181毫秒/步 - aug_p: 0.0427 - d_loss: 0.1064 - g_loss: 5.1232 - gen_acc: 0.9674 - real_acc: 0.9623



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 181毫秒/步 - aug_p: 0.0427 - d_loss: 0.1101 - g_loss: 5.1442 - gen_acc: 0.9662 - real_acc: 0.9603



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4秒 181毫秒/步 - aug_p: 0.0428 - d_loss: 0.1136 - g_loss: 5.1570 - gen_acc: 0.9649 - real_acc: 0.9587



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0428 - d_loss: 0.1166 - g_loss: 5.1674 - gen_acc: 0.9638 - real_acc: 0.9573



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0428 - d_loss: 0.1192 - g_loss: 5.1751 - gen_acc: 0.9628 - real_acc: 0.9561



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0428 - d_loss: 0.1216 - g_loss: 5.1786 - gen_acc: 0.9620 - real_acc: 0.9550



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0428 - d_loss: 0.1238 - g_loss: 5.1785 - gen_acc: 0.9612 - real_acc: 0.9539



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0429 - d_loss: 0.1258 - g_loss: 5.1765 - gen_acc: 0.9605 - real_acc: 0.9530



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0429 - d_loss: 0.1276 - g_loss: 5.1726 - gen_acc: 0.9599 - real_acc: 0.9521



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0429 - d_loss: 0.1294 - g_loss: 5.1667 - gen_acc: 0.9595 - real_acc: 0.9513



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0429 - d_loss: 0.1309 - g_loss: 5.1594 - gen_acc: 0.9590 - real_acc: 0.9506



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0429 - d_loss: 0.1323 - g_loss: 5.1512 - gen_acc: 0.9587 - real_acc: 0.9500



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0429 - d_loss: 0.1335 - g_loss: 5.1414 - gen_acc: 0.9584 - real_acc: 0.9494



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 180毫秒/步 - aug_p: 0.0430 - d_loss: 0.1346 - g_loss: 5.1320 - gen_acc: 0.9582 - real_acc: 0.9489



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 180毫秒/步 - aug_p: 0.0430 - d_loss: 0.1356 - g_loss: 5.1216 - gen_acc: 0.9580 - real_acc: 0.9484



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 180毫秒/步 - aug_p: 0.0430 - d_loss: 0.1365 - g_loss: 5.1109 - gen_acc: 0.9579 - real_acc: 0.9479



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0430 - d_loss: 0.1373 - g_loss: 5.0996 - gen_acc: 0.9578 - real_acc: 0.9475



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0430 - d_loss: 0.1379 - g_loss: 5.0882 - gen_acc: 0.9577 - real_acc: 0.9472



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 180毫秒/步 - aug_p: 0.0431 - d_loss: 0.1385 - g_loss: 5.0769 - gen_acc: 0.9577 - real_acc: 0.9468



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1秒 180毫秒/步 - aug_p: 0.0431 - d_loss: 0.1391 - g_loss: 5.0648 - gen_acc: 0.9577 - real_acc: 0.9466



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0秒 180毫秒/步 - aug_p: 0.0431 - d_loss: 0.1395 - g_loss: 5.0535 - gen_acc: 0.9577 - real_acc: 0.9463



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 180毫秒/步 - aug_p: 0.0431 - d_loss: 0.1400 - g_loss: 5.0419 - gen_acc: 0.9576 - real_acc: 0.9461



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 180毫秒/步 - aug_p: 0.0431 - d_loss: 0.1403 - g_loss: 5.0307 - gen_acc: 0.9577 - real_acc: 0.9459



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 180毫秒/步 - aug_p: 0.0431 - d_loss: 0.1406 - g_loss: 5.0198 - gen_acc: 0.9577 - real_acc: 0.9458



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 180毫秒/步 - aug_p: 0.0432 - d_loss: 0.1408 - g_loss: 5.0087 - gen_acc: 0.9577 - real_acc: 0.9456



46/46 ━━━━━━━━━━━━━━━━━━━━ 0秒 180毫秒/步 - aug_p: 0.0432 - d_loss: 0.1410 - g_loss: 4.9981 - gen_acc: 0.9578 - real_acc: 0.9455



46/46 ━━━━━━━━━━━━━━━━━━━━ 14秒 300毫秒/步 - aug_p: 0.0432 - d_loss: 0.1411 - g_loss: 4.9879 - gen_acc: 0.9579 - real_acc: 0.9455 - val_kid: 3.6018

Epoch 10/10

1/46 [37m━━━━━━━━━━━━━━━━━━━━ 5分15秒 7秒/步 - aug_p: 0.0464 - d_loss: 0.0324 - g_loss: 4.1750 - gen_acc: 1.0000 - real_acc: 0.9922



2/46 [37m━━━━━━━━━━━━━━━━━━━━ 8秒 195毫秒/步 - aug_p: 0.0464 - d_loss: 0.0337 - g_loss: 4.0349 - gen_acc: 0.9980 - real_acc: 0.9941



3/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 8秒 186毫秒/步 - aug_p: 0.0464 - d_loss: 0.0367 - g_loss: 4.0199 - gen_acc: 0.9978 - real_acc: 0.9918



4/46 ━ [37m━━━━━━━━━━━━━━━━━━━ 7秒 184毫秒/步 - aug_p: 0.0465 - d_loss: 0.0374 - g_loss: 4.0297 - gen_acc: 0.9979 - real_acc: 0.9909



5/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 183毫秒/步 - aug_p: 0.0465 - d_loss: 0.0380 - g_loss: 4.0271 - gen_acc: 0.9980 - real_acc: 0.9902



6/46 ━━ [37m━━━━━━━━━━━━━━━━━━ 7秒 183毫秒/步 - aug_p: 0.0465 - d_loss: 0.0383 - g_loss: 4.0130 - gen_acc: 0.9981 - real_acc: 0.9901



7/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 7秒 183毫秒/步 - aug_p: 0.0466 - d_loss: 0.0385 - g_loss: 4.0148 - gen_acc: 0.9982 - real_acc: 0.9901



8/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0466 - d_loss: 0.0389 - g_loss: 4.0141 - gen_acc: 0.9983 - real_acc: 0.9902



9/46 ━━━ [37m━━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0467 - d_loss: 0.0393 - g_loss: 4.0076 - gen_acc: 0.9984 - real_acc: 0.9903



10/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0467 - d_loss: 0.0397 - g_loss: 4.0031 - gen_acc: 0.9985 - real_acc: 0.9903



11/46 ━━━━ [37m━━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0467 - d_loss: 0.0402 - g_loss: 3.9981 - gen_acc: 0.9985 - real_acc: 0.9902



12/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 6秒 182毫秒/步 - aug_p: 0.0468 - d_loss: 0.0406 - g_loss: 3.9968 - gen_acc: 0.9985 - real_acc: 0.9902



13/46 ━━━━━ [37m━━━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0468 - d_loss: 0.0411 - g_loss: 3.9967 - gen_acc: 0.9985 - real_acc: 0.9899



14/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0468 - d_loss: 0.0418 - g_loss: 3.9930 - gen_acc: 0.9984 - real_acc: 0.9897



15/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0469 - d_loss: 0.0428 - g_loss: 3.9956 - gen_acc: 0.9982 - real_acc: 0.9893



16/46 ━━━━━━ [37m━━━━━━━━━━━━━━ 5秒 181毫秒/步 - aug_p: 0.0469 - d_loss: 0.0436 - g_loss: 3.9957 - gen_acc: 0.9980 - real_acc: 0.9890



17/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0469 - d_loss: 0.0443 - g_loss: 3.9983 - gen_acc: 0.9978 - real_acc: 0.9887



18/46 ━━━━━━━ [37m━━━━━━━━━━━━━ 5秒 180毫秒/步 - aug_p: 0.0470 - d_loss: 0.0450 - g_loss: 3.9978 - gen_acc: 0.9977 - real_acc: 0.9885



19/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0470 - d_loss: 0.0457 - g_loss: 3.9987 - gen_acc: 0.9976 - real_acc: 0.9883



20/46 ━━━━━━━━ [37m━━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0470 - d_loss: 0.0464 - g_loss: 3.9966 - gen_acc: 0.9974 - real_acc: 0.9880



21/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0471 - d_loss: 0.0472 - g_loss: 3.9956 - gen_acc: 0.9973 - real_acc: 0.9877



22/46 ━━━━━━━━━ [37m━━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0471 - d_loss: 0.0482 - g_loss: 3.9910 - gen_acc: 0.9969 - real_acc: 0.9874



23/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 4秒 180毫秒/步 - aug_p: 0.0471 - d_loss: 0.0501 - g_loss: 3.9936 - gen_acc: 0.9965 - real_acc: 0.9862



24/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0472 - d_loss: 0.0532 - g_loss: 3.9900 - gen_acc: 0.9949 - real_acc: 0.9853



25/46 ━━━━━━━━━━ [37m━━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0472 - d_loss: 0.0576 - g_loss: 3.9964 - gen_acc: 0.9935 - real_acc: 0.9832



26/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 180毫秒/步 - aug_p: 0.0472 - d_loss: 0.0624 - g_loss: 3.9986 - gen_acc: 0.9917 - real_acc: 0.9813



27/46 ━━━━━━━━━━━ [37m━━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0472 - d_loss: 0.0667 - g_loss: 4.0030 - gen_acc: 0.9901 - real_acc: 0.9795



28/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0473 - d_loss: 0.0707 - g_loss: 4.0083 - gen_acc: 0.9887 - real_acc: 0.9778



29/46 ━━━━━━━━━━━━ [37m━━━━━━━━ 3秒 181毫秒/步 - aug_p: 0.0473 - d_loss: 0.0744 - g_loss: 4.0128 - gen_acc: 0.9873 - real_acc: 0.9762



30/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0473 - d_loss: 0.0776 - g_loss: 4.0161 - gen_acc: 0.9862 - real_acc: 0.9748



31/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0473 - d_loss: 0.0806 - g_loss: 4.0186 - gen_acc: 0.9851 - real_acc: 0.9735



32/46 ━━━━━━━━━━━━━ [37m━━━━━━━ 2秒 181毫秒/步 - aug_p: 0.0474 - d_loss: 0.0832 - g_loss: 4.0199 - gen_acc: 0.9841 - real_acc: 0.9724



33/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 182毫秒/步 - aug_p: 0.0474 - d_loss: 0.0856 - g_loss: 4.0204 - gen_acc: 0.9832 - real_acc: 0.9714



34/46 ━━━━━━━━━━━━━━ [37m━━━━━━ 2秒 182毫秒/步 - aug_p: 0.0474 - d_loss: 0.0878 - g_loss: 4.0206 - gen_acc: 0.9825 - real_acc: 0.9705



35/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 182毫秒/步 - aug_p: 0.0474 - d_loss: 0.0898 - g_loss: 4.0206 - gen_acc: 0.9818 - real_acc: 0.9697



36/46 ━━━━━━━━━━━━━━━ [37m━━━━━ 1秒 182毫秒/步 - aug_p: 0.0475 - d_loss: 0.0916 - g_loss: 4.0200 - gen_acc: 0.9811 - real_acc: 0.9690



37/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 182毫秒/步 - aug_p: 0.0475 - d_loss: 0.0933 - g_loss: 4.0193 - gen_acc: 0.9805 - real_acc: 0.9683



38/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 182毫秒/步 - aug_p: 0.0475 - d_loss: 0.0948 - g_loss: 4.0185 - gen_acc: 0.9800 - real_acc: 0.9677



39/46 ━━━━━━━━━━━━━━━━ [37m━━━━ 1秒 182毫秒/步 - aug_p: 0.0475 - d_loss: 0.0961 - g_loss: 4.0171 - gen_acc: 0.9796 - real_acc: 0.9672



40/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 1秒 182毫秒/步 - aug_p: 0.0475 - d_loss: 0.0974 - g_loss: 4.0158 - gen_acc: 0.9791 - real_acc: 0.9667



41/46 ━━━━━━━━━━━━━━━━━ [37m━━━ 0秒 182毫秒/步 - aug_p: 0.0476 - d_loss: 0.0985 - g_loss: 4.0146 - gen_acc: 0.9787 - real_acc: 0.9662



42/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 182毫秒/步 - aug_p: 0.0476 - d_loss: 0.0995 - g_loss: 4.0133 - gen_acc: 0.9784 - real_acc: 0.9658



43/46 ━━━━━━━━━━━━━━━━━━ [37m━━ 0秒 182毫秒/步 - aug_p: 0.0476 - d_loss: 0.1005 - g_loss: 4.0119 - gen_acc: 0.9781 - real_acc: 0.9655



44/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 182毫秒/步 - aug_p: 0.0476 - d_loss: 0.1013 - g_loss: 4.0102 - gen_acc: 0.9778 - real_acc: 0.9652



45/46 ━━━━━━━━━━━━━━━━━━━ [37m━ 0秒 182毫秒/步 - aug_p: 0.0476 - d_loss: 0.1021 - g_loss: 4.0083 - gen_acc: 0.9775 - real_acc: 0.9649



46/46 ━━━━━━━━━━━━━━━━━━━━ 0秒 182毫秒/步 - aug_p: 0.0477 - d_loss: 0.1028 - g_loss: 4.0070 - gen_acc: 0.9773 - real_acc: 0.9647



46/46 ━━━━━━━━━━━━━━━━━━━━ 21秒 304毫秒/步 - aug_p: 0.0477 - d_loss: 0.1035 - g_loss: 4.0058 - gen_acc: 0.9771 - real_acc: 0.9644 - val_kid: 3.0212

<keras.src.callbacks.history.History at 0x794f705d3390>

推理

# load the best model and generate images
model.load_weights(checkpoint_path)
model.plot_images()

结果

通过运行 400 个 epoch 的训练（在 Colab 笔记本中需要 2-3 小时），可以使用此代码示例获得高质量的图像生成。

在 400 个 epoch 训练中随机一批图像的演变过程（ema=0.999 用于动画平滑）：

选定图像批次之间的潜在空间插值：

我还建议尝试在其他数据集上进行训练，例如 CelebA。根据我的经验，无需更改任何超参数即可获得良好的结果（尽管判别器增强可能不是必要的）。

GAN 技巧和诀窍

我希望通过此示例找到 GAN 的易于实现性和生成质量之间的良好平衡。在准备过程中，我使用此仓库进行了多次消融实验。

在本节中，我将按主观重要性顺序列出经验教训和我的建议。

我建议查看 DCGAN 论文、此 NeurIPS 演讲 和此 大规模 GAN 研究，以了解其他人对此主题的看法。

架构技巧

• 分辨率：在较高分辨率下训练 GAN 往往更困难，我建议最初在 32x32 或 64x64 分辨率下进行实验。
• 初始化：如果在训练初期看到强烈的彩色图案，则可能是初始化问题。将层的 kernel_initializer 参数设置为 随机正态分布，并减小标准差（推荐值：0.02，遵循 DCGAN），直到问题消失。
• 上采样：生成器中有两种主要的上采样方法。转置卷积速度更快，但可能导致棋盘格伪影，可以通过使用可被步幅整除的内核大小来减少伪影（对于步幅为 2，推荐内核大小为 4）。上采样 + 标准卷积的质量可能略低，但棋盘格伪影不是问题。我建议对其使用最近邻插值而不是双线性插值。
• 判别器中的批归一化：有时影响很大，我建议两种方式都尝试一下。
• 谱归一化：一种流行的 GAN 训练技术，可以帮助提高稳定性。我建议禁用批归一化的可学习尺度参数以及谱归一化。
• 残差连接：虽然残差判别器的行为类似，但根据我的经验，残差生成器更难训练。然而，它们对于训练大型和深层架构是必要的。我建议从非残差架构开始。
• dropout：根据我的经验，在判别器的最后一层之前使用 dropout 可以提高生成质量。推荐的 dropout 率低于 0.5。
• leaky ReLU：在判别器中使用 leaky ReLU 激活函数，以使其梯度不那么稀疏。推荐的斜率/alpha 值为 0.2，遵循 DCGAN。

算法技巧

• 损失函数：多年来，人们提出了许多用于训练 GAN 的损失函数，承诺提高性能和稳定性。我在此仓库中实现了其中的 5 个，我的经验与这项 GAN 研究一致：似乎没有哪种损失函数能始终优于默认的非饱和 GAN 损失。我建议将此作为默认值。
• Adam 的 beta_1 参数：Adam 中的 beta_1 参数可以解释为平均梯度估计的动量。DCGAN 中提出使用 0.5 甚至 0.0 而不是默认值 0.9，这很重要。此示例在使用其默认值的情况下将无法工作。
• 为生成图像和真实图像分别进行批归一化：判别器的前向传播对于生成图像和真实图像应该是分开的。否则可能会导致伪影（在我的情况下为 45 度条纹）和性能下降。
• 生成器权重的指数移动平均：这有助于减少 KID 测量的方差，并有助于平均训练期间的快速调色板变化。
• 生成器和判别器使用不同的学习率：如果资源充足，可以帮助分别调整两个网络的学习率。一个类似的想法是，对于另一个网络的每次更新，多次更新其中一个网络（通常是判别器）的权重。我建议对于两个网络都使用相同的 2e-4 学习率 (Adam)，遵循 DCGAN，并且仅默认更新它们一次。
• 标签噪声：单边标签平滑（对真实标签使用小于 1.0 的值）或向标签添加噪声可以正则化判别器，使其不会过度自信，但在我的情况下，它们并没有提高性能。
• 自适应数据增强：由于它为训练过程添加了另一个动态组件，因此默认情况下禁用它，并且仅在其他组件已经运行良好时才启用它。

相关工作

其他 GAN 相关的 Keras 代码示例

• DCGAN + CelebA
• WGAN + FashionMNIST
• WGAN + Molecules
• ConditionalGAN + MNIST
• CycleGAN + Horse2Zebra
• StyleGAN

现代 GAN 架构系列

• SAGAN, BigGAN
• ProgressiveGAN, StyleGAN, StyleGAN2, StyleGAN2-ADA, AliasFreeGAN

关于判别器数据增强的同期论文： 1, 2, 3

最近关于 GAN 的文献综述： 演讲