{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.7.12","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":13836,"databundleVersionId":1718836,"sourceType":"competition"}],"dockerImageVersionId":30299,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import tensorflow as tf\nimport tensorflow_addons as tfa\nimport math\n\nimport pandas as pd\nimport cv2\nfrom PIL import Image\nimport numpy as np\nimport os\nimport matplotlib.pyplot as plt\nfrom kaggle_datasets import KaggleDatasets","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"try:\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()\nprint('Number of replicas:', strategy.num_replicas_in_sync)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"IMAGE_SIZE = [512, 512]\nINPUT_SIZE = [256, 256] #just for fast training\nAUTO = tf.data.experimental.AUTOTUNE\n\ndef data_pipeline(filenames, label_num, batch_size, is_train=True):\n    label_num = tf.constant(label_num, tf.int32)\n    dataset = tf.data.TFRecordDataset(filenames, num_parallel_reads=AUTO)\n    dataset = dataset.map(read_tfrecord, num_parallel_calls=AUTO)\n    dataset = dataset.filter(lambda x, y: y == label_num)# only seletcted label number can go through\n    dataset = dataset.map(lambda x, y: x, num_parallel_calls=AUTO)# do not use label\n    if is_train:\n        dataset = dataset.map(data_augmentation, num_parallel_calls=AUTO)\n        dataset = dataset.shuffle(64)\n        dataset = dataset.repeat()\n    dataset = dataset.map(resize_and_normalize, num_parallel_calls=AUTO)   \n    dataset = dataset.batch(batch_size)\n    dataset = dataset.prefetch(AUTO)\n    return dataset\n\ndef read_tfrecord(example):\n    tfrecord_format = {\n        \"image\": tf.io.FixedLenFeature([], tf.string),\n        \"target\": tf.io.FixedLenFeature([], tf.int64)\n        }\n    example = tf.io.parse_single_example(example, tfrecord_format)\n    img = decode_img(example['image'])\n    label = tf.cast(example['target'], tf.int32)\n    return img, label\n\ndef decode_img(img):\n    img = tf.image.decode_jpeg(img, channels=3)\n    img = tf.cast(img, tf.float32)\n    img = tf.reshape(img, [*IMAGE_SIZE, 3])\n    return img\n\ndef data_augmentation(img):\n    # horizontal flip\n    if tf.random.uniform(()) > 0.5:\n        img = img[:, ::-1, :]\n    \n    # vertical flip\n    if tf.random.uniform(()) > 0.5:\n        img = img[::-1, :, :]\n        \n    # random crop\n    if tf.random.uniform(()) > 0.3:\n        height = tf.random.uniform(shape=(), minval=450, maxval=500)\n        width = tf.random.uniform(shape=(), minval=450, maxval=500)\n        img = tf.image.random_crop(img, [height,width,3])\n    \n    # change brightness and contrast\n    contrast_rate = tf.random.uniform(shape=(), minval=0.8, maxval=1.2)\n    img = tf.image.adjust_contrast(img, contrast_rate)\n\n    brightness_shift = tf.random.uniform(shape=(), minval=-20, maxval=20)\n    img = tf.image.adjust_brightness(img, brightness_shift)\n    img = tf.clip_by_value(img, 0, 255)\n\n    # change hue\n    hue_rate = tf.random.uniform(shape=(), minval=-0.05, maxval=0.05)\n    img = tf.image.adjust_hue(img, hue_rate)\n\n    return img\n      \ndef resize_and_normalize(img):\n    img = img[tf.newaxis, ...]\n    img = tf.image.resize(img, tuple(INPUT_SIZE))[0,:,:,:]\n    img = (img/127.5) - 1\n    return img","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def build_gan_models():\n    generator_x2y = build_generator()\n    generator_y2x = build_generator()\n    discriminator_x = build_discriminator()\n    discriminator_y = build_discriminator()\n    return generator_x2y, generator_y2x, discriminator_x, discriminator_y\n\n\ndef downsample(filters, size, apply_instancenorm=True):\n    initializer = tf.random_normal_initializer(0., 0.02)\n    gamma_init = tf.keras.initializers.RandomNormal(mean=0.0, stddev=0.02)\n\n    result = tf.keras.Sequential()\n    result.add(tf.keras.layers.Conv2D(filters, size, strides=2, padding='same',\n                             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(tf.keras.layers.LeakyReLU())\n\n    return result\n\ndef upsample(filters, size, apply_dropout=False):\n    initializer = tf.random_normal_initializer(0., 0.02)\n    gamma_init = tf.keras.initializers.RandomNormal(mean=0.0, stddev=0.02)\n\n    result = tf.keras.Sequential()\n    result.add(tf.keras.layers.Conv2DTranspose(filters, size, strides=2,\n                                      padding='same',\n                                      kernel_initializer=initializer,\n                                      use_bias=False))\n\n    result.add(tfa.layers.InstanceNormalization(gamma_initializer=gamma_init))\n\n    if apply_dropout:\n        result.add(tf.keras.layers.Dropout(0.5))\n\n    result.add(tf.keras.layers.ReLU())\n\n    return result\n\ndef build_generator():\n    inputs = tf.keras.layers.Input(shape=[256,256,3])\n\n    down_stack = [\n        downsample(64, 4, apply_instancenorm=False),\n        downsample(128, 4),\n        downsample(256, 4),\n        downsample(512, 4),\n        downsample(512, 4),\n        downsample(512, 4),\n        downsample(512, 4),\n        downsample(512, 4),\n    ]\n\n    up_stack = [\n        upsample(512, 4, apply_dropout=True),\n        upsample(512, 4, apply_dropout=True),\n        upsample(512, 4, apply_dropout=True),\n        upsample(512, 4),\n        upsample(256, 4),\n        upsample(128, 4),\n        upsample(64, 4),\n    ]\n\n    initializer = tf.random_normal_initializer(0., 0.02)\n    last = tf.keras.layers.Conv2DTranspose(3, 4,\n                                  strides=2,\n                                  padding='same',\n                                  kernel_initializer=initializer,\n                                  activation='tanh')\n\n    x = inputs\n\n    # Downsampling through the model\n    skips = []\n    for down in down_stack:\n        x = down(x)\n        skips.append(x)\n\n    skips = reversed(skips[:-1])\n\n    # Upsampling and establishing the skip connections\n    for up, skip in zip(up_stack, skips):\n        x = up(x)\n        x = tf.keras.layers.Concatenate()([x, skip])\n\n    x = last(x)\n\n    return tf.keras.Model(inputs=inputs, outputs=x)\n\ndef build_discriminator():\n    initializer = tf.random_normal_initializer(0., 0.02)\n    gamma_init = tf.keras.initializers.RandomNormal(mean=0.0, stddev=0.02)\n\n    inputs = tf.keras.layers.Input(shape=[256, 256, 3], name='input_image')\n    \n    # I use pretrained model of VGG instead of the model given in the original notebook,\n    # because this classification task is much more difficult than the original task(Monet or photo). \n    pretrained_model = tf.keras.applications.VGG16(weights=\"imagenet\", include_top=False ,input_shape=[256,256, 3])\n    x = pretrained_model(inputs)\n    \"\"\"\n    \n    down1 = downsample(64, 4, False)(inputs)\n    down2 = downsample(128, 4)(down1)\n    down3 = downsample(256, 4)(down2)\n\n    zero_pad1 = tf.keras.layers.ZeroPadding2D()(down3)\n    conv = tf.keras.layers.Conv2D(512, 4, strides=1,\n                         kernel_initializer=initializer,\n                         use_bias=False)(zero_pad1)\n\n    norm1 = tfa.layers.InstanceNormalization(gamma_initializer=gamma_init)(conv)\n\n    leaky_relu = tf.keras.layers.LeakyReLU()(norm1)\n\n    zero_pad2 = tf.keras.layers.ZeroPadding2D()(leaky_relu)\n\n    last = tf.keras.layers.Conv2D(1, 4, strides=1,\n                         kernel_initializer=initializer)(zero_pad2)\n    \"\"\"\n    last =  tf.keras.layers.Conv2D(1, 4, strides=1, kernel_initializer=initializer)(x)\n    \n    return tf.keras.Model(inputs=inputs, outputs=last)\n\n\ndef discriminator_loss(real, generated):\n    bc_loss = tf.keras.losses.BinaryCrossentropy(from_logits=True, reduction=tf.keras.losses.Reduction.NONE)\n    real_loss = bc_loss(tf.ones_like(real), real)# real as real\n    generated_loss = bc_loss(tf.zeros_like(generated), generated)# fake as fake\n    #loss = tf.reduce_mean(tf.abs(real_image - same_image))\n    return (real_loss + generated_loss) * 0.5\n\ndef generator_loss(generated):\n    bc_loss = tf.keras.losses.BinaryCrossentropy(from_logits=True, reduction=tf.keras.losses.Reduction.NONE)# fake as real\n    return bc_loss(tf.ones_like(generated), generated)\n\ndef calc_cycle_loss(real_image, cycled_image):\n    loss = tf.reduce_mean(tf.abs(real_image - cycled_image))\n    return 10 * loss\n\ndef identity_loss(real_image, same_image):\n    loss = tf.reduce_mean(tf.abs(real_image - same_image))\n    return 5 * loss\n\n\nclass CycleGan(tf.keras.Model):\n    def __init__(\n        self,\n        generator_x2y,\n        generator_y2x,\n        discriminator_x,\n        discriminator_y,\n    ):\n        super(CycleGan, self).__init__()\n        self.generator_x2y = generator_x2y\n        self.generator_y2x = generator_y2x\n        self.discriminator_x = discriminator_x\n        self.discriminator_y = discriminator_y\n        \n    def compile(\n        self,\n        generator_x2y_optimizer,\n        generator_y2x_optimizer,\n        discriminator_x_optimizer,\n        discriminator_y_optimizer,\n        gen_loss_fn,\n        disc_loss_fn,\n        cycle_loss_fn,\n        identity_loss_fn\n    ):\n        super(CycleGan, self).compile()\n        self.generator_x2y_optimizer = generator_x2y_optimizer\n        self.generator_y2x_optimizer = generator_y2x_optimizer\n        self.discriminator_x_optimizer = discriminator_x_optimizer\n        self.discriminator_y_optimizer = discriminator_y_optimizer\n        self.generator_loss = gen_loss_fn\n        self.discriminator_loss = disc_loss_fn\n        self.cycle_loss = cycle_loss_fn\n        self.identity_loss = identity_loss_fn\n        \n    def train_step(self, batch_data):\n        real_x, real_y = batch_data\n        with tf.GradientTape(persistent=True) as tape:\n            # Generator G translates X -> Y\n            # Generator F translates Y -> X.\n\n            fake_y = self.generator_x2y(real_x, training=True)\n            cycled_x = self.generator_y2x(fake_y, training=True)\n\n            fake_x = self.generator_y2x(real_y, training=True)\n            cycled_y = self.generator_x2y(fake_x, training=True)\n\n            # same_x and same_y are used for identity loss. in half cycle, do not change original image\n            same_x = self.generator_y2x(real_x, training=True)\n            same_y = self.generator_x2y(real_y, training=True)\n\n            # discriminate 4 images in total\n            disc_real_x = self.discriminator_x(real_x, training=True)\n            disc_real_y = self.discriminator_y(real_y, training=True)\n\n            disc_fake_x = self.discriminator_x(fake_x, training=True)\n            disc_fake_y = self.discriminator_y(fake_y, training=True)\n\n            # calculate the loss\n            gen_x2y_loss = self.generator_loss(disc_fake_y)\n            gen_y2x_loss = self.generator_loss(disc_fake_x)\n\n            total_cycle_loss = self.cycle_loss(real_x, cycled_x) + self.cycle_loss(real_y, cycled_y)\n\n            # Total generator loss = adversarial loss + cycle loss\n            total_gen_x2y_loss = gen_x2y_loss + total_cycle_loss + self.identity_loss(real_y, same_y)\n            total_gen_y2x_loss = gen_y2x_loss + total_cycle_loss + self.identity_loss(real_x, same_x)\n\n            disc_x_loss = self.discriminator_loss(disc_real_x, disc_fake_x)\n            disc_y_loss = self.discriminator_loss(disc_real_y, disc_fake_y)\n      \n        # Calculate the gradients for generator and discriminator\n        generator_x2y_gradients = tape.gradient(total_gen_x2y_loss, \n                                              self.generator_x2y.trainable_variables)\n        generator_y2x_gradients = tape.gradient(total_gen_y2x_loss, \n                                              self.generator_y2x.trainable_variables)\n        \n        discriminator_x_gradients = tape.gradient(disc_x_loss, \n                                                  self.discriminator_x.trainable_variables)\n        discriminator_y_gradients = tape.gradient(disc_y_loss, \n                                                  self.discriminator_y.trainable_variables)\n      \n        # Apply the gradients to the optimizer\n        generator_x2y_optimizer.apply_gradients(zip(generator_x2y_gradients, \n                                                  self.generator_x2y.trainable_variables))\n    \n        generator_y2x_optimizer.apply_gradients(zip(generator_y2x_gradients, \n                                                  self.generator_y2x.trainable_variables))\n      \n        discriminator_x_optimizer.apply_gradients(zip(discriminator_x_gradients,\n                                                      self.discriminator_x.trainable_variables))\n        \n        discriminator_y_optimizer.apply_gradients(zip(discriminator_y_gradients,\n                                                      discriminator_y.trainable_variables))\n    \n    \n        return {\n                \"gen_x2y_loss\": total_gen_x2y_loss,\n                \"gen_y2x_loss\": total_gen_y2x_loss,\n                \"disc_x_loss\": disc_x_loss,\n                \"disc_y_loss\": disc_y_loss\n                }","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"batch_size = 1\nGCS_PATH = KaggleDatasets().get_gcs_path()\nfilenames = tf.io.gfile.glob(GCS_PATH + '/train_tfrecords/ld_train*.tfrec')\nCBSD_data = data_pipeline(filenames, label_num=1, batch_size=batch_size, is_train=True)\nHealthy_data = data_pipeline(filenames, label_num=4, batch_size=batch_size, is_train=True)\n\nnum_samples = 4\nfig, axes = plt.subplots(2,num_samples, figsize=(12,6))\nfor i, data in enumerate(CBSD_data.take(num_samples)):\n    axes[0, i].imshow(data[0].numpy() * 0.5 + 0.5)\n    axes[0, i].title.set_text(\"CBSD\")\n    \nfor i, data in enumerate(Healthy_data.take(num_samples)):\n    axes[1, i].imshow(data[0].numpy() * 0.5 + 0.5)\n    axes[1, i].title.set_text(\"Healthy\")    \n\nfig.tight_layout()\nplt.show()\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data_num = 2000\nlearning_rate = 1e-4\nbatch_size = 1\n\nwith strategy.scope(): \n    generator_x2y, generator_y2x, discriminator_x, discriminator_y = build_gan_models()\n\n\n    # loss and learning rate\n    generator_x2y_optimizer = tf.keras.optimizers.Adam(learning_rate, beta_1=0.5)\n    generator_y2x_optimizer = tf.keras.optimizers.Adam(learning_rate, beta_1=0.5)\n\n    discriminator_x_optimizer = tf.keras.optimizers.Adam(learning_rate/10, beta_1=0.5)\n    discriminator_y_optimizer = tf.keras.optimizers.Adam(learning_rate/10, beta_1=0.5)\n\n    cycle_gan_model = CycleGan(generator_x2y, generator_y2x, discriminator_x, discriminator_y)\n\n\n    def compile_and_fit(epochs):\n        cycle_gan_model.compile(generator_x2y_optimizer, generator_y2x_optimizer,\n                                discriminator_x, discriminator_y,\n                                generator_loss,\n                                discriminator_loss,\n                                calc_cycle_loss,\n                                identity_loss\n                                )\n        cycle_gan_model.fit(tf.data.Dataset.zip((CBSD_data, Healthy_data)),\n                            batch_size = batch_size,\n                            steps_per_epoch = data_num//batch_size, \n                            epochs=epochs\n                           )    \n\n\n    # I don't want to break the pretrained weight of disciminator at initial epochs.\n    discriminator_x.trainable = False\n    discriminator_y.trainable = False\n    compile_and_fit(epochs=3)\n\n    discriminator_x.trainable = True\n    discriminator_y.trainable = True\n    compile_and_fit(epochs=10)\n\n    cycle_gan_model.save_weights(\"gan_10ep_CBSD_healthy.h5\")","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def generate_images(model, test_input):\n    prediction = model(test_input)\n    plt.figure(figsize=(10, 10))\n    display_list = [test_input[0], prediction[0]]\n    title = ['Original Image', 'Generated Image']\n    for i in range(2):\n        plt.subplot(1, 2, i+1)\n        plt.title(title[i])\n        plt.imshow(display_list[i] * 0.5 + 0.5)\n        plt.axis('off')\n    plt.show()\n\nfor inputs in CBSD_data.take(12):\n    generate_images(generator_x2y, inputs)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for inputs in Healthy_data.take(12):\n    generate_images(generator_y2x, inputs) ","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Colors and patterns of leaves look slightly changing. They are somewhat unhealthy.","metadata":{}},{"cell_type":"code","source":"CBB_data = data_pipeline(filenames, label_num=0, batch_size=batch_size, is_train=True)\nwith strategy.scope(): \n    generator_x2y, generator_y2x, discriminator_x, discriminator_y = build_gan_models()\n\n    generator_x2y_optimizer = tf.keras.optimizers.Adam(learning_rate, beta_1=0.5)\n    generator_y2x_optimizer = tf.keras.optimizers.Adam(learning_rate, beta_1=0.5)\n    discriminator_x_optimizer = tf.keras.optimizers.Adam(learning_rate/10, beta_1=0.5)\n    discriminator_y_optimizer = tf.keras.optimizers.Adam(learning_rate/10, beta_1=0.5)\n    \n    cycle_gan_model = CycleGan(generator_x2y, generator_y2x, discriminator_x, discriminator_y)\n\n    def compile_and_fit(epochs):\n        cycle_gan_model.compile(generator_x2y_optimizer, generator_y2x_optimizer,\n                                discriminator_x, discriminator_y,\n                                generator_loss,\n                                discriminator_loss,\n                                calc_cycle_loss,\n                                identity_loss\n                                )\n        cycle_gan_model.fit(tf.data.Dataset.zip((CBB_data, Healthy_data)),\n                            batch_size = batch_size,\n                            steps_per_epoch = data_num//batch_size, \n                            epochs=epochs\n                           )    \n\n\n    # I don't want to break the pretrained weight of disciminator at initial epochs.\n    discriminator_x.trainable = False\n    discriminator_y.trainable = False\n    compile_and_fit(epochs=3)\n\n    discriminator_x.trainable = True\n    discriminator_y.trainable = True\n    compile_and_fit(epochs=10)\n\n    cycle_gan_model.save_weights(\"gan_10ep_CBB_healthy.h5\")\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for inputs in CBB_data.take(12):\n    generate_images(generator_x2y, inputs)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for inputs in Healthy_data.take(12):\n    generate_images(generator_y2x, inputs)","metadata":{"trusted":true},"execution_count":null,"outputs":[]}]}