{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import tensorflow as tf\nfrom tensorflow import keras\nfrom tensorflow.keras import layers\nimport tensorflow_addons as tfa\nfrom kaggle_datasets import KaggleDatasets\nimport matplotlib.pyplot as plt\nimport numpy as np\nimport re\nimport os\nimport math\nimport random\nimport cv2\n\ntry:\n    tpu = tf.distribute.cluster_resolver.TPUClusterResolver()\n    print('Device:', tpu.master())\n    tf.config.experimental_connect_to_cluster(tpu)\n    tf.tpu.experimental.initialize_tpu_system(tpu)\n    strategy = tf.distribute.experimental.TPUStrategy(tpu)\nexcept:\n    strategy = tf.distribute.get_strategy()","metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# 读取数据\n\n读取jpg文件","metadata":{}},{"cell_type":"code","source":"GCS_PATH = KaggleDatasets().get_gcs_path('gan-getting-started')\nfn_monet = tf.io.gfile.glob(str(GCS_PATH + '/monet_jpg/*.jpg'))\nfn_photo = tf.io.gfile.glob(str(GCS_PATH + '/photo_jpg/*.jpg'))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"BATCH_SIZE =  4\n\n\ndef parse_function(filename):\n    image_string = tf.io.read_file(filename)\n    image = tf.image.decode_jpeg(image_string, channels=3)\n    image = (tf.cast(image,tf.float32)/ 127.5) - 1\n    image = tf.reshape(image, [256, 256,3])\n    return image\n\ndef data_augment(image):\n    p_rotate = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_spatial = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    p_crop = tf.random.uniform([], 0, 1.0, dtype=tf.float32)\n    if p_crop > .5:\n        image = tf.image.resize(image, [286, 286])\n        image = tf.image.random_crop(image, size=[256, 256, 3])\n        if p_crop > .9:\n            image = tf.image.resize(image, [300, 300])\n            image = tf.image.random_crop(image, size=[256, 256, 3])\n    \n    if p_rotate > .9:\n        image = tf.image.rot90(image, k=3)\n    elif p_rotate > .7:\n        image = tf.image.rot90(image, k=2)\n    elif p_rotate > .5:\n        image = tf.image.rot90(image, k=1)\n        \n    if p_spatial > .6:\n        image = tf.image.random_flip_left_right(image)\n        image = tf.image.random_flip_up_down(image)\n        if p_spatial > .9:\n            image = tf.image.transpose(image)\n    \n    return image\n\nnum_parallel_calls=tf.data.experimental.AUTOTUNE\ndef getSet(filenames):\n    dataset = tf.data.Dataset.from_tensor_slices(filenames)\n    dataset = dataset.shuffle(len(filenames))\n    dataset = dataset.map(parse_function, num_parallel_calls)\n    dataset = dataset.map(data_augment, num_parallel_calls)\n    dataset = dataset.repeat()\n    dataset = dataset.batch(BATCH_SIZE,drop_remainder=True)\n    dataset = dataset.cache()\n    dataset = dataset.prefetch(num_parallel_calls)\n    return dataset\n\nmonet_ds=getSet(fn_monet)\nphoto_ds=getSet(fn_photo)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# 生成器的实现\n\n生成器使用U-Net来实现。U-Net主要被用于图像分割。由于是个图像到图像的网络，因此在CycleGAN中可以被用于生成器\n\nU-Net首先对图像进行下采样，方法是使用步长为2的$4\\times 4$**卷积**，使用padding=\"same\"，将使得卷积后图片的尺寸变为“原尺寸/步长“的向上取整（如果padding=\"valid\"，则图片的尺寸会变为“(原尺寸-卷积核尺寸+1)/步长“的向上取整）。第一次卷积，使用64个卷积核将3通道变成64通道，之后每一次使用的卷积核数目是上一次的两倍，从而将通道数增大一倍，最大增大到512通道。**保存下采样的每一步的结果**，在之后会用到。\n\n等到把图片大小卷积到$1\\times 1$后，转为对图片进行升采样，方法是使用步长为2的$4\\times 4$**逆卷积**，使用padding=\"same\"，将使得卷积后图片的尺寸变为“原尺寸\\*步长“（如果padding=\"valid\"，则图片的尺寸会变为“原尺寸\\*步长-卷积核尺寸+1“）最后一次卷积，使用3个卷积核将64通道变成3通道，之前每一次使用的卷积核数目是前一次的1/2倍，从而将通道数减少一半，从512通道开始减少。上采样过程中，每次会沿通道维度拼接下采样过程中对应步的结果。","metadata":{}},{"cell_type":"code","source":"OUTPUT_CHANNELS = 3\n\ndef downsample(filters, size, apply_instancenorm=True):\n    initializer = tf.random_normal_initializer(0., 0.02)\n    gamma_init = keras.initializers.RandomNormal(mean=0.0, stddev=0.02)\n\n    result = keras.Sequential()\n    result.add(layers.Conv2D(filters, size, strides=2, padding='same',kernel_initializer=initializer, use_bias=False))\n\n    if apply_instancenorm:\n        result.add(tfa.layers.InstanceNormalization(gamma_initializer=gamma_init))\n\n    result.add(layers.LeakyReLU())\n\n    return result","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def upsample(filters, size, apply_dropout=False):\n    initializer = tf.random_normal_initializer(0., 0.02)\n    gamma_init = keras.initializers.RandomNormal(mean=0.0, stddev=0.02)\n\n    result = keras.Sequential()\n    result.add(layers.Conv2DTranspose(filters, size, strides=2,padding='same',kernel_initializer=initializer,use_bias=False))\n\n    result.add(tfa.layers.InstanceNormalization(gamma_initializer=gamma_init))\n\n    if apply_dropout:\n        result.add(layers.Dropout(0.5))\n\n    result.add(layers.ReLU())\n\n    return result","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def Generator():\n    inputs = layers.Input(shape=[256,256,3])\n\n    # bs = batch size\n    #下采样层列表\n    down_stack = [\n        downsample(64, 4, apply_instancenorm=False), # (bs, 128, 128, 64) 256/2向上取整（因为padding是same，否则(256-4+1)/2向上取整）\n        downsample(128, 4), # (bs, 64, 64, 128)\n        downsample(256, 4), # (bs, 32, 32, 256)\n        downsample(512, 4), # (bs, 16, 16, 512)\n        downsample(512, 4), # (bs, 8, 8, 512)\n        downsample(512, 4), # (bs, 4, 4, 512)\n        downsample(512, 4), # (bs, 2, 2, 512)\n        downsample(512, 4), # (bs, 1, 1, 512)\n    ]\n    #上采样层列表\n    up_stack = [\n        upsample(512, 4, apply_dropout=True), # (bs, 2, 2, 1024) 1*2（因为padding是same，否则(1*2-4+1)），此外还拼接了一个张量，因此通道数增加\n        upsample(512, 4, apply_dropout=True), # (bs, 4, 4, 1024)\n        upsample(512, 4, apply_dropout=True), # (bs, 8, 8, 1024)\n        upsample(512, 4), # (bs, 16, 16, 1024)\n        upsample(256, 4), # (bs, 32, 32, 512)\n        upsample(128, 4), # (bs, 64, 64, 256)\n        upsample(64, 4), # (bs, 128, 128, 128)\n    ]\n\n    initializer = tf.random_normal_initializer(0., 0.02)\n    last = layers.Conv2DTranspose(OUTPUT_CHANNELS, 4,strides=2,padding='same',kernel_initializer=initializer,activation='tanh') # (bs, 256, 256, 3)，最后一层，重新变成原来大小\n\n    x = inputs\n\n    #进行下采样\n    skips = []\n    for down in down_stack:\n        x = down(x)\n        skips.append(x)\n\n    skips = reversed(skips[:-1])#最下层的在最前\n\n    #进行上采样，并且进行跨接\n    for up, skip in zip(up_stack, skips):\n        x = up(x)\n        x = layers.Concatenate()([x, skip])\n\n    x = last(x)\n\n    return keras.Model(inputs=inputs, outputs=x)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"tf.keras.utils.plot_model(Generator(), show_shapes=True, dpi=64)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# 鉴别器的实现\n\n鉴别器使用的是普通的下采样卷积神经网络，下采样3次后，输出一个$30\\times30$的矩阵作为结果，具有指示真实分类的较高像素值和指示假分类的较低值","metadata":{}},{"cell_type":"code","source":"def Discriminator():\n    initializer = tf.random_normal_initializer(0., 0.02)\n    gamma_init = keras.initializers.RandomNormal(mean=0.0, stddev=0.02)\n\n    inp = layers.Input(shape=[256, 256, 3], name='input_image')\n\n    x = inp\n\n    down1 = downsample(64, 4, False)(x) # (bs, 128, 128, 64)\n    down2 = downsample(128, 4)(down1) # (bs, 64, 64, 128)\n    down3 = downsample(256, 4)(down2) # (bs, 32, 32, 256)\n\n    zero_pad1 = layers.ZeroPadding2D()(down3) # (bs, 34, 34, 256)\n    conv = layers.Conv2D(512, 4, strides=1,kernel_initializer=initializer,use_bias=False)(zero_pad1) # (bs, 31, 31, 512)\n\n    norm1 = tfa.layers.InstanceNormalization(gamma_initializer=gamma_init)(conv)\n\n    leaky_relu = layers.LeakyReLU()(norm1)\n\n    zero_pad2 = layers.ZeroPadding2D()(leaky_relu) # (bs, 33, 33, 512)\n\n    last = layers.Conv2D(1, 4, strides=1,\n                         kernel_initializer=initializer)(zero_pad2) # (bs, 30, 30, 1)该鉴别器不输出单个节点，而是输出较小的2D图像，其具有指示真实分类的较高像素值和指示假分类的较低值\n\n    return tf.keras.Model(inputs=inp, outputs=last)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"tf.keras.utils.plot_model(Discriminator(), show_shapes=True, dpi=64)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# CycleGAN\n\nCycleGAN包含两个生成器$G$和$F$，对应两个鉴别器$D_X$和$D_Y$\n\n* 生成器：$G(x)$，$x$为分布$p_x$中取样的样本，目标是其生成的样本和分布$p_y$中取样的样本$y$尽可能接近\n* 鉴别器：$D_Y(y)$，目标是尽可能区分$G(x)$和$y$，输出属于$G(p_x)$而不属于$p_y$的概率\n\n**对抗性损失**1：$L_{GAN}(G,D_Y)=E_{y\\sim p_y}(logD_Y(y))+E_{x\\sim p_x}(logD_Y(G(x)))$\n\n* 生成器：$F(y)$，$y$为分布$p_y$中取样的样本，目标是其生成的样本和分布$p_x$中取样的样本$x$尽可能接近\n* 鉴别器：$D_X(x)$，目标是尽可能区分$ F(y)$和$x$，输出属于$F(p_y)$而不属于$p_x$的概率\n\n**对抗性损失**2：$L_{GAN}(F,D_X)=E_{x\\sim p_x}(logD_X(x))+E_{y\\sim p_y}(logD_X(F(y)))$\n\n尽管上述的对抗性损失能够让生成器$G$和生成器$F$学习到$p_x$和$p_y$，但是却没有保证从$x$得到$G(x)$时图像的内容不变，因为$G(x)$只需要符合$p_y$即可，并没有对其施加约束，所以$x$到$G(x)$包含很多种可能的映射\n\n为此，使用**循环一致性损失**来作为约束，使得$G$生成的$G(x)$在内容上仍然能和$x$保持一致\n\n循环一致性损失：$L_{cyc}(G,F)=E_{x\\sim p_x}[||F(G(x))-x||_1]+E_{y\\sim p_y}[||G(F(y))-y||_1]$\n\n总体损失：$L(G,F,D_X,D_Y)=L_{GAN}(G,D_Y)+L_{GAN}(F,D_X)+\\lambda L_{cyc}(G,F)$\n\n以上是CycleGAN论文中的内容，实现的时候我们实际上是这么做的：\n\n* 对鉴别器$Y $而言，要最小化的损失函数为：$D_Y(y)$接近0的程度+$D_Y(G(x))$接近1的程度\n* 对鉴别器$X $而言，要最小化的损失函数为：$D_X(x)$接近0的程度+$D_X(F(y))$接近1的程度\n* 循环一致性损失为：$F(G(x))$与$x$的距离（像素平均）+$G(F(y))$与$y$的距离（像素平均）\n* 对生成器$G$而言，要最小化的损失函数为：$D_Y(G(x))$接近0的程度+$\\lambda$*循环一致性损失\n* 对生成器$F$而言，要最小化的损失函数为：$D_X(F(y))$接近0的程度+$\\lambda$*循环一致性损失\n\n我们让鉴别器输出一个$30\\times30$的矩阵作为结果，并**与全0矩阵或全1矩阵计算交叉熵**，以衡量输出结果接近0或接近1的程度\n\n本代码还包括一个自我一致性损失，这是论文中没有的部分，计算的是$G(y)$与$y$的距离（像素平均）+$F(x)$与$x$的距离（像素平均），同样乘以$\\lambda$后加到了生成器的损失函数上","metadata":{}},{"cell_type":"code","source":"class CycleGan(keras.Model):\n    def __init__(self,monet_generator,photo_generator,monet_discriminator,photo_discriminator,lambda_cycle=15):\n        super(CycleGan, self).__init__()\n        self.m_gen = monet_generator\n        self.p_gen = photo_generator\n        self.m_disc = monet_discriminator\n        self.p_disc = photo_discriminator\n        self.lambda_cycle = lambda_cycle\n        \n    def compile(self,m_gen_optimizer,p_gen_optimizer,m_disc_optimizer,p_disc_optimizer,gen_loss_fn,disc_loss_fn,cycle_loss_fn,identity_loss_fn):\n        super(CycleGan, self).compile()\n        self.m_gen_optimizer = m_gen_optimizer\n        self.p_gen_optimizer = p_gen_optimizer\n        self.m_disc_optimizer = m_disc_optimizer\n        self.p_disc_optimizer = p_disc_optimizer\n        \n        self.gen_loss_fn = gen_loss_fn\n        self.disc_loss_fn = disc_loss_fn\n        self.cycle_loss_fn = cycle_loss_fn\n        self.identity_loss_fn = identity_loss_fn\n        \n    def train_step(self, batch_data):\n        real_monet, real_photo = batch_data\n        #real_monet为y，real_photo为x，m_gen为G，p_gen为F，p_disc为DX，m_disc为DY\n        \n        with tf.GradientTape(persistent=True) as tape:\n            fake_monet = self.m_gen(real_photo, training=True)#G(x)\n            cycled_photo = self.p_gen(fake_monet, training=True)#F(G(x))\n\n            fake_photo = self.p_gen(real_monet, training=True)#F(y)\n            cycled_monet = self.m_gen(fake_photo, training=True)#G(F(y))\n\n            same_monet = self.m_gen(real_monet, training=True)#G(y)\n            same_photo = self.p_gen(real_photo, training=True)#F(x)\n\n            disc_real_monet = self.m_disc(real_monet, training=True)#DY(y)\n            disc_real_photo = self.p_disc(real_photo, training=True)#DX(x)\n\n            disc_fake_monet = self.m_disc(fake_monet, training=True)#DY(G(x))\n            disc_fake_photo = self.p_disc(fake_photo, training=True)#DX(F(y))\n\n            monet_gen_loss = self.gen_loss_fn(disc_fake_monet)#用于训练生成器的损失函数（基本部分）\n            photo_gen_loss = self.gen_loss_fn(disc_fake_photo)#用于训练生成器的损失函数（基本部分）\n\n            total_cycle_loss = self.cycle_loss_fn(real_monet, cycled_monet, self.lambda_cycle) + self.cycle_loss_fn(real_photo, cycled_photo, self.lambda_cycle)#循环一致性损失，乘以lambda\n\n            # evaluates total generator loss\n            total_monet_gen_loss = monet_gen_loss + total_cycle_loss + self.identity_loss_fn(real_monet, same_monet, self.lambda_cycle)#用于训练生成器的总损失函数\n            total_photo_gen_loss = photo_gen_loss + total_cycle_loss + self.identity_loss_fn(real_photo, same_photo, self.lambda_cycle)#用于训练生成器的总损失函数\n\n            # evaluates discriminator loss\n            monet_disc_loss = self.disc_loss_fn(disc_real_monet, disc_fake_monet)#用于训练鉴别器的损失函数\n            photo_disc_loss = self.disc_loss_fn(disc_real_photo, disc_fake_photo)#用于训练鉴别器的损失函数\n\n        # 损失函数对网络当中的参数求梯度（之前由网络输出结果时，设置了training为True，因此可以对网络中的参数求梯度）\n        monet_generator_gradients = tape.gradient(total_monet_gen_loss,self.m_gen.trainable_variables)#生成器的总损失函数对生成器的所有参数求梯度\n        photo_generator_gradients = tape.gradient(total_photo_gen_loss,self.p_gen.trainable_variables)#生成器的总损失函数对生成器的所有参数求梯度\n        monet_discriminator_gradients = tape.gradient(monet_disc_loss,self.m_disc.trainable_variables)#鉴别器的损失函数对鉴别器的所有参数求梯度\n        photo_discriminator_gradients = tape.gradient(photo_disc_loss,self.p_disc.trainable_variables)#鉴别器的损失函数对鉴别器的所有参数求梯度\n\n        self.m_gen_optimizer.apply_gradients(zip(monet_generator_gradients,self.m_gen.trainable_variables))#优化器向前迭代一步\n        self.p_gen_optimizer.apply_gradients(zip(photo_generator_gradients,self.p_gen.trainable_variables))#优化器向前迭代一步\n        self.m_disc_optimizer.apply_gradients(zip(monet_discriminator_gradients,self.m_disc.trainable_variables))#优化器向前迭代一步\n        self.p_disc_optimizer.apply_gradients(zip(photo_discriminator_gradients,self.p_disc.trainable_variables))#优化器向前迭代一步\n        \n        return {\"monet_gen_loss\": total_monet_gen_loss,\"photo_gen_loss\": total_photo_gen_loss,\"monet_disc_loss\": monet_disc_loss,\"photo_disc_loss\": photo_disc_loss}","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# 损失函数\n\n用于训练鉴别器的损失函数","metadata":{}},{"cell_type":"code","source":"with strategy.scope():\n    def discriminator_loss(real, generated):\n        real_loss = tf.keras.losses.BinaryCrossentropy(from_logits=True, reduction=tf.keras.losses.Reduction.NONE)(tf.ones_like(real), real)#对于真实样本，通过鉴别器后与全1矩阵计算交叉熵，使得D对真实数据输出尽可能接近1\n        generated_loss = tf.keras.losses.BinaryCrossentropy(from_logits=True, reduction=tf.keras.losses.Reduction.NONE)(tf.zeros_like(generated), generated)#对于生成样本，通过鉴别器后与全0矩阵计算交叉熵，使得D对生成数据输出尽可能接近0\n        total_disc_loss = real_loss + generated_loss\n        return total_disc_loss * 0.5","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"用于训练生成器的损失函数（基本部分）","metadata":{}},{"cell_type":"code","source":"with strategy.scope():\n    def generator_loss(generated):\n        return tf.keras.losses.BinaryCrossentropy(from_logits=True, reduction=tf.keras.losses.Reduction.NONE)(tf.ones_like(generated), generated)#对于生成样本，通过鉴别器后与全1矩阵计算交叉熵，使得D对生成数据输出尽可能接近1","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"循环一致性损失","metadata":{}},{"cell_type":"code","source":"with strategy.scope():\n    def calc_cycle_loss(real_image, cycled_image, LAMBDA):\n        loss1 = tf.reduce_mean(tf.abs(real_image - cycled_image))#循环一致性损失\n        return LAMBDA * loss1","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"自我一致损失（论文中没有的部分）","metadata":{}},{"cell_type":"code","source":"with strategy.scope():\n    def identity_loss(real_image, same_image, LAMBDA):\n        loss = tf.reduce_mean(tf.abs(real_image - same_image))\n        return LAMBDA * 0.5 * loss","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# 训练","metadata":{}},{"cell_type":"code","source":"with strategy.scope():\n    monet_generator_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)\n    photo_generator_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)\n    monet_discriminator_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)\n    photo_discriminator_optimizer = tf.keras.optimizers.Adam(2e-4, beta_1=0.5)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"with strategy.scope():\n    monet_generator=Generator()\n    photo_generator=Generator()\n    monet_discriminator=Discriminator()\n    photo_discriminator=Discriminator()\nwith strategy.scope():\n    cycle_gan_model = CycleGan(monet_generator,photo_generator,monet_discriminator,photo_discriminator)\n    cycle_gan_model.compile(\n            m_gen_optimizer = monet_generator_optimizer,\n            p_gen_optimizer = photo_generator_optimizer,\n            m_disc_optimizer = monet_discriminator_optimizer,\n            p_disc_optimizer = photo_discriminator_optimizer,\n            gen_loss_fn = generator_loss,\n            disc_loss_fn = discriminator_loss,\n            cycle_loss_fn = calc_cycle_loss,\n            identity_loss_fn = identity_loss\n        )","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"steps_per_epoch=(max(len(fn_monet), len(fn_photo)))//BATCH_SIZE\ncycle_gan_model.fit(tf.data.Dataset.zip((monet_ds, photo_ds)),epochs=25,steps_per_epoch=steps_per_epoch)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# 可视化结果","metadata":{}},{"cell_type":"code","source":"testset = tf.data.Dataset.from_tensor_slices(fn_photo)\ntestset = testset.shuffle(len(fn_photo))\ntestset = testset.map(parse_function, num_parallel_calls)\ntestset=testset.batch(1)\ntestset = testset.prefetch(num_parallel_calls)\n\n\n_, ax = plt.subplots(5, 2, figsize=(20, 20))\nfor i, img in enumerate(testset.take(5)):\n    prediction = monet_generator(img, training=False)[0].numpy()\n    prediction = (prediction * 127.5 + 127.5).astype(np.uint8)\n    img = (img[0] * 127.5 + 127.5).numpy().astype(np.uint8)\n\n    ax[i, 0].imshow(img)\n    ax[i, 1].imshow(prediction)\n    ax[i, 0].set_title(\"Input Photo\")\n    ax[i, 1].set_title(\"Monet-esque\")\n    ax[i, 0].axis(\"off\")\n    ax[i, 1].axis(\"off\")\nplt.show()","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# 提交","metadata":{}},{"cell_type":"code","source":"import PIL\n\ndef predict_and_save(input_ds, generator_model, output_path):\n    i = 1\n    for img in input_ds:\n        prediction = generator_model(img, training=False)[0].numpy() # make predition\n        prediction = (prediction * 127.5 + 127.5).astype(np.uint8)   # re-scale\n        im = PIL.Image.fromarray(prediction)\n        im.save(output_path+\"/\"+str(i)+'.jpg')\n        i += 1","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import os\nos.makedirs('../images/') # Create folder to save generated images\n\npredict_and_save(testset, monet_generator, '../images/')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import shutil\nshutil.make_archive('/kaggle/working/images/', 'zip', '../images')\n\nprint(f\"Generated samples: {len([name for name in os.listdir('../images/') if os.path.isfile(os.path.join('../images/', name))])}\")","metadata":{"trusted":true},"execution_count":null,"outputs":[]}]}