{"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":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nimport tensorflow as tf\nfrom tensorflow import keras\nfrom tensorflow.keras import layers\nimport matplotlib.pyplot as plt\nimport glob\nimport time\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\n# for dirname, _, filenames in os.walk('/kaggle/input'):\n#     for filename in filenames:\n#         print(os.path.join(dirname, filename))\n\n# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2022-10-07T04:56:58.212613Z","iopub.execute_input":"2022-10-07T04:56:58.213053Z","iopub.status.idle":"2022-10-07T04:57:03.528717Z","shell.execute_reply.started":"2022-10-07T04:56:58.212959Z","shell.execute_reply":"2022-10-07T04:57:03.527766Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# From official tutorial of Monet-GAN competition","metadata":{}},{"cell_type":"code","source":"image_feature_description={\n    'image_name': tf.io.FixedLenFeature([], tf.string),\n    'image': tf.io.FixedLenFeature([], tf.string),\n    'target' : tf.io.FixedLenFeature([], tf.string)\n}","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:12.072387Z","iopub.execute_input":"2022-10-07T04:57:12.073758Z","iopub.status.idle":"2022-10-07T04:57:12.079159Z","shell.execute_reply.started":"2022-10-07T04:57:12.073706Z","shell.execute_reply":"2022-10-07T04:57:12.078103Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"IMAGE_SIZE = [256, 256]\ndef decode_image(image):\n    image = tf.image.decode_jpeg(image, channels=3)\n    image = (tf.cast(image, tf.float32) / 127.5) - 1\n    image = tf.reshape(image, [*IMAGE_SIZE,3])\n    return image","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:13.896751Z","iopub.execute_input":"2022-10-07T04:57:13.897115Z","iopub.status.idle":"2022-10-07T04:57:13.903557Z","shell.execute_reply.started":"2022-10-07T04:57:13.897083Z","shell.execute_reply":"2022-10-07T04:57:13.902486Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def parse_image_function(example_proto):\n    example= tf.io.parse_single_example(example_proto, image_feature_description)\n    image = decode_image(example['image'])\n    return image","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:14.536455Z","iopub.execute_input":"2022-10-07T04:57:14.537134Z","iopub.status.idle":"2022-10-07T04:57:14.542799Z","shell.execute_reply.started":"2022-10-07T04:57:14.537093Z","shell.execute_reply":"2022-10-07T04:57:14.541682Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"AUTOTUNE = tf.data.AUTOTUNE\ndef load_data(filename):\n    raw_ds = tf.data.TFRecordDataset(filename)\n    raw_ds = raw_ds.map(parse_image_function, num_parallel_calls=AUTOTUNE)\n    return raw_ds","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:15.151379Z","iopub.execute_input":"2022-10-07T04:57:15.151994Z","iopub.status.idle":"2022-10-07T04:57:15.157162Z","shell.execute_reply.started":"2022-10-07T04:57:15.151963Z","shell.execute_reply":"2022-10-07T04:57:15.155980Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"MONET_FILENAMES = os.listdir(\"../input/gan-getting-started/monet_tfrec\")\nprint(len(MONET_FILENAMES))","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:17.021926Z","iopub.execute_input":"2022-10-07T04:57:17.022763Z","iopub.status.idle":"2022-10-07T04:57:17.033425Z","shell.execute_reply.started":"2022-10-07T04:57:17.022727Z","shell.execute_reply":"2022-10-07T04:57:17.032418Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"monet_tfrec = \"../input/gan-getting-started/monet_tfrec\"\nMONET_PATHS = [os.path.join(monet_tfrec,filename) for filename in MONET_FILENAMES]","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:17.416196Z","iopub.execute_input":"2022-10-07T04:57:17.417780Z","iopub.status.idle":"2022-10-07T04:57:17.422533Z","shell.execute_reply.started":"2022-10-07T04:57:17.417740Z","shell.execute_reply":"2022-10-07T04:57:17.421341Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"monet_ds = load_data(MONET_PATHS).batch(1)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:18.016228Z","iopub.execute_input":"2022-10-07T04:57:18.016921Z","iopub.status.idle":"2022-10-07T04:57:20.901989Z","shell.execute_reply.started":"2022-10-07T04:57:18.016880Z","shell.execute_reply":"2022-10-07T04:57:20.900987Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"sample = next(iter(monet_ds))\nplt.imshow(sample[0]*0.5 + 0.5)\nplt.axis('off')","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:27.617180Z","iopub.execute_input":"2022-10-07T04:57:27.617915Z","iopub.status.idle":"2022-10-07T04:57:28.054096Z","shell.execute_reply.started":"2022-10-07T04:57:27.617878Z","shell.execute_reply":"2022-10-07T04:57:28.053076Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(sample.shape)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:30.461525Z","iopub.execute_input":"2022-10-07T04:57:30.461899Z","iopub.status.idle":"2022-10-07T04:57:30.467338Z","shell.execute_reply.started":"2022-10-07T04:57:30.461868Z","shell.execute_reply":"2022-10-07T04:57:30.466055Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# (28,28) dataset for tutorial DCGAN checking","metadata":{}},{"cell_type":"code","source":"IMG_SIZE=28\nresize = keras.layers.Resizing(IMG_SIZE,IMG_SIZE)\ntutorial_ds = monet_ds.map(lambda x : resize(x))","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:32.736989Z","iopub.execute_input":"2022-10-07T04:57:32.737793Z","iopub.status.idle":"2022-10-07T04:57:32.774332Z","shell.execute_reply.started":"2022-10-07T04:57:32.737749Z","shell.execute_reply":"2022-10-07T04:57:32.773437Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"sample = next(iter(tutorial_ds))","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:57:34.106781Z","iopub.execute_input":"2022-10-07T04:57:34.107129Z","iopub.status.idle":"2022-10-07T04:57:34.140401Z","shell.execute_reply.started":"2022-10-07T04:57:34.107097Z","shell.execute_reply":"2022-10-07T04:57:34.139509Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.imshow(sample[0,:,:,:]*0.5 + 0.5)\nplt.axis('off')","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:58:13.021929Z","iopub.execute_input":"2022-10-07T04:58:13.022311Z","iopub.status.idle":"2022-10-07T04:58:13.112472Z","shell.execute_reply.started":"2022-10-07T04:58:13.022260Z","shell.execute_reply":"2022-10-07T04:58:13.111190Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Implementing DCGAN models\n## Generator model\n### Input : random noise ([1,100]) -> output (256,256,3) image\nUpsampling using combinations of Conv2DTranspose, BatchNorm, LeakyReLU and an input Dense layer","metadata":{}},{"cell_type":"code","source":"noise_shape = 100\n\ndef generator_model():\n\n    model = keras.Sequential()\n    model.add(layers.Dense(8*8*512, use_bias= False, input_shape=(noise_shape,)))\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    model.add(layers.Reshape((8,8,512)))\n\n    assert model.output_shape==(None, 8,8,512)\n\n    model.add(layers.Conv2DTranspose(filters= 256, kernel_size=(5,5), strides=(2,2), padding='same', use_bias=False))\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    assert model.output_shape == (None, 16,16,256)\n\n    model.add(layers.Conv2DTranspose(filters= 128, kernel_size=(5,5), strides=(2,2), padding='same', use_bias=False))\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    assert model.output_shape == (None, 32,32,128)\n\n    model.add(layers.Conv2DTranspose(filters= 64, kernel_size=(5,5), strides=(2,2), padding='same', use_bias=False, activation='tanh'))\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    assert model.output_shape == (None, 64,64,64)\n\n    model.add(layers.Conv2DTranspose(filters= 32, kernel_size=(5,5), strides=(2,2), padding='same', use_bias=False, activation='tanh'))\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    assert model.output_shape == (None, 128,128,32)\n\n    model.add(layers.Conv2DTranspose(filters= 3, kernel_size=(5,5), strides=(2,2), padding='same', use_bias=False, activation='tanh'))\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n    assert model.output_shape == (None, 256,256,3)\n    \n    return model","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:58:26.512825Z","iopub.execute_input":"2022-10-07T04:58:26.513182Z","iopub.status.idle":"2022-10-07T04:58:26.526211Z","shell.execute_reply.started":"2022-10-07T04:58:26.513152Z","shell.execute_reply":"2022-10-07T04:58:26.525238Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def tutorial_generator_model():\n    model = tf.keras.Sequential()\n    model.add(layers.Dense(7*7*256, use_bias=False, input_shape=(100,)))\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    model.add(layers.Reshape((7, 7, 256)))\n    assert model.output_shape == (None, 7, 7, 256)  # Note: None is the batch size\n\n    model.add(layers.Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False))\n    assert model.output_shape == (None, 7, 7, 128)\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    model.add(layers.Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False))\n    assert model.output_shape == (None, 14, 14, 64)\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    model.add(layers.Conv2DTranspose(3, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh'))\n    assert model.output_shape == (None, 28, 28, 3)\n    \n    return model","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:58:29.501958Z","iopub.execute_input":"2022-10-07T04:58:29.502334Z","iopub.status.idle":"2022-10-07T04:58:29.512412Z","shell.execute_reply.started":"2022-10-07T04:58:29.502296Z","shell.execute_reply":"2022-10-07T04:58:29.511246Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generator = generator_model()\nnoise = tf.random.normal([1,100])\ngenerated_image= generator(noise, training=False)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T06:06:00.866000Z","iopub.execute_input":"2022-10-07T06:06:00.866393Z","iopub.status.idle":"2022-10-07T06:06:01.015293Z","shell.execute_reply.started":"2022-10-07T06:06:00.866359Z","shell.execute_reply":"2022-10-07T06:06:01.014397Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(generated_image.shape)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:58:42.860332Z","iopub.execute_input":"2022-10-07T04:58:42.860749Z","iopub.status.idle":"2022-10-07T04:58:42.866966Z","shell.execute_reply.started":"2022-10-07T04:58:42.860711Z","shell.execute_reply":"2022-10-07T04:58:42.865932Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generated_image[0,:,:,0].shape","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:58:42.868469Z","iopub.execute_input":"2022-10-07T04:58:42.869121Z","iopub.status.idle":"2022-10-07T04:58:42.877831Z","shell.execute_reply.started":"2022-10-07T04:58:42.869074Z","shell.execute_reply":"2022-10-07T04:58:42.876609Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def generate_images(r=1,c=1, train_dim=100, is_norm=False):\n    plt.figure(figsize=(4*r,5*c))    \n    for i in range(1,1+r*c):        \n        noise = tf.random.normal([1,train_dim])\n        plt.subplot(r,c,i)\n        if(is_norm):\n            plt.imshow(generator(noise)[0,:,:,:]*255)\n        else:            \n            plt.imshow(generator(noise)[0,:,:,:]*0.5 + 0.5)\n        plt.axis('off')\ngenerate_images()","metadata":{"execution":{"iopub.status.busy":"2022-10-07T06:06:10.230805Z","iopub.execute_input":"2022-10-07T06:06:10.231670Z","iopub.status.idle":"2022-10-07T06:06:10.353723Z","shell.execute_reply.started":"2022-10-07T06:06:10.231635Z","shell.execute_reply":"2022-10-07T06:06:10.352573Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generator = tutorial_generator_model()\nnoise = tf.random.normal([1,100])\ngenerated_image= generator(noise, training=False)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T06:06:27.915904Z","iopub.execute_input":"2022-10-07T06:06:27.916450Z","iopub.status.idle":"2022-10-07T06:06:28.015737Z","shell.execute_reply.started":"2022-10-07T06:06:27.916408Z","shell.execute_reply":"2022-10-07T06:06:28.014820Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(generated_image.shape)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:59:08.076806Z","iopub.execute_input":"2022-10-07T04:59:08.077171Z","iopub.status.idle":"2022-10-07T04:59:08.086554Z","shell.execute_reply.started":"2022-10-07T04:59:08.077137Z","shell.execute_reply":"2022-10-07T04:59:08.085431Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images()","metadata":{"execution":{"iopub.status.busy":"2022-10-07T06:06:33.145504Z","iopub.execute_input":"2022-10-07T06:06:33.145867Z","iopub.status.idle":"2022-10-07T06:06:33.238372Z","shell.execute_reply.started":"2022-10-07T06:06:33.145835Z","shell.execute_reply":"2022-10-07T06:06:33.237148Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Discriminator\n## Takes (256,256,3) image and gives bool (0/1) output, predicting whether input_image is fake or real","metadata":{}},{"cell_type":"code","source":"def discriminator_model():\n    model = keras.Sequential()\n    model.add(layers.Conv2D(filters=64, kernel_size=(5,5), strides=(2,2), padding='same', input_shape=[256,256,3]))\n    \n    model.add(layers.LeakyReLU())\n    model.add(layers.Dropout(0.3))\n    \n    model.add(layers.Conv2D(filters=128, kernel_size=(5,5), strides=(2,2), padding='same'))\n    model.add(layers.LeakyReLU())\n    model.add(layers.Dropout(0.3))\n    \n    model.add(layers.Flatten())\n    model.add(layers.Dense(1))\n    \n    return model","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:59:21.617208Z","iopub.execute_input":"2022-10-07T04:59:21.617900Z","iopub.status.idle":"2022-10-07T04:59:21.625215Z","shell.execute_reply.started":"2022-10-07T04:59:21.617864Z","shell.execute_reply":"2022-10-07T04:59:21.624260Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def tutorial_discriminator_model():\n    model = tf.keras.Sequential()\n    model.add(layers.Conv2D(64, (5, 5), strides=(2, 2), padding='same',\n                                     input_shape=[28, 28, 3]))\n    model.add(layers.LeakyReLU())\n    model.add(layers.Dropout(0.3))\n\n    model.add(layers.Conv2D(128, (5, 5), strides=(2, 2), padding='same'))\n    model.add(layers.LeakyReLU())\n    model.add(layers.Dropout(0.3))\n\n    model.add(layers.Flatten())\n    model.add(layers.Dense(1))\n\n    return model","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:59:24.136634Z","iopub.execute_input":"2022-10-07T04:59:24.137591Z","iopub.status.idle":"2022-10-07T04:59:24.144900Z","shell.execute_reply.started":"2022-10-07T04:59:24.137535Z","shell.execute_reply":"2022-10-07T04:59:24.143920Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"discriminator = tutorial_discriminator_model()\ngen_pred = discriminator(generated_image, training=False)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:59:29.116819Z","iopub.execute_input":"2022-10-07T04:59:29.117183Z","iopub.status.idle":"2022-10-07T04:59:29.189790Z","shell.execute_reply.started":"2022-10-07T04:59:29.117150Z","shell.execute_reply":"2022-10-07T04:59:29.188891Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(gen_pred)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:59:31.082374Z","iopub.execute_input":"2022-10-07T04:59:31.082735Z","iopub.status.idle":"2022-10-07T04:59:31.089744Z","shell.execute_reply.started":"2022-10-07T04:59:31.082704Z","shell.execute_reply":"2022-10-07T04:59:31.088478Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Defining loss and optimizations","metadata":{}},{"cell_type":"markdown","source":"## Loss","metadata":{}},{"cell_type":"code","source":"cross_entropy_loss = keras.losses.BinaryCrossentropy(from_logits=True)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:59:34.286687Z","iopub.execute_input":"2022-10-07T04:59:34.287042Z","iopub.status.idle":"2022-10-07T04:59:34.291794Z","shell.execute_reply.started":"2022-10-07T04:59:34.287010Z","shell.execute_reply":"2022-10-07T04:59:34.290743Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Optimizers","metadata":{}},{"cell_type":"code","source":"generator_optimizer = keras.optimizers.Adam(1e-4)\ndiscriminator_optimizer = keras.optimizers.Adam(1e-4)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:59:37.761769Z","iopub.execute_input":"2022-10-07T04:59:37.762130Z","iopub.status.idle":"2022-10-07T04:59:37.767175Z","shell.execute_reply.started":"2022-10-07T04:59:37.762096Z","shell.execute_reply":"2022-10-07T04:59:37.766083Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Training step and Training loop","metadata":{}},{"cell_type":"code","source":"@tf.function\ndef train_step(image_batch, train_dim):\n    noise = tf.random.normal([1,train_dim])\n    \n    with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:            \n        generated_image = generator(noise, training=True)\n\n        discriminator_pred_gen = discriminator(generated_image, training=True)\n\n        discriminator_pred_ori= discriminator(image_batch, training=True)\n\n\n        discriminator_loss = cross_entropy_loss(tf.ones_like(discriminator_pred_ori),discriminator_pred_ori) + cross_entropy_loss(tf.zeros_like(discriminator_pred_gen), discriminator_pred_gen)\n        generator_loss = cross_entropy_loss(tf.ones_like(discriminator_pred_gen), discriminator_pred_gen)\n        \n    generator_grad = gen_tape.gradient(generator_loss, generator.trainable_variables)\n    discriminator_grad = disc_tape.gradient(discriminator_loss, discriminator.trainable_variables)\n\n    generator_optimizer.apply_gradients(zip(generator_grad,generator.trainable_variables))\n    discriminator_optimizer.apply_gradients(zip(discriminator_grad,discriminator.trainable_variables))    ","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:31:34.227658Z","iopub.execute_input":"2022-10-07T05:31:34.228028Z","iopub.status.idle":"2022-10-07T05:31:34.238809Z","shell.execute_reply.started":"2022-10-07T05:31:34.227996Z","shell.execute_reply":"2022-10-07T05:31:34.237949Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Training loop","metadata":{}},{"cell_type":"code","source":"def train(dataset, epochs, train_dim=100):\n    for epoch in range(epochs):\n        start = time.time()\n        \n        for image_batch in dataset:\n            train_step(image_batch, train_dim)\n        \n        print(f\"Time for epoch {epoch+ 1} is : {time.time()-start}\")","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:31:41.212204Z","iopub.execute_input":"2022-10-07T05:31:41.212907Z","iopub.status.idle":"2022-10-07T05:31:41.218630Z","shell.execute_reply.started":"2022-10-07T05:31:41.212871Z","shell.execute_reply":"2022-10-07T05:31:41.217481Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"normalization_layer = keras.layers.Rescaling(1./255)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T04:59:50.261848Z","iopub.execute_input":"2022-10-07T04:59:50.262199Z","iopub.status.idle":"2022-10-07T04:59:50.267961Z","shell.execute_reply.started":"2022-10-07T04:59:50.262169Z","shell.execute_reply":"2022-10-07T04:59:50.266987Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"EPOCHS= 50\ngenerator = generator_model()\ndiscriminator = discriminator_model()\ntrain(monet_ds,EPOCHS)","metadata":{"scrolled":true,"execution":{"iopub.status.busy":"2022-10-07T04:59:52.776663Z","iopub.execute_input":"2022-10-07T04:59:52.777022Z","iopub.status.idle":"2022-10-07T05:02:37.677024Z","shell.execute_reply.started":"2022-10-07T04:59:52.776991Z","shell.execute_reply":"2022-10-07T05:02:37.675975Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images()","metadata":{"scrolled":true,"_kg_hide-output":false,"execution":{"iopub.status.busy":"2022-10-07T05:02:45.591817Z","iopub.execute_input":"2022-10-07T05:02:45.592183Z","iopub.status.idle":"2022-10-07T05:02:45.762723Z","shell.execute_reply.started":"2022-10-07T05:02:45.592151Z","shell.execute_reply":"2022-10-07T05:02:45.761545Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(5,5)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:03:30.137484Z","iopub.execute_input":"2022-10-07T05:03:30.138223Z","iopub.status.idle":"2022-10-07T05:03:32.492576Z","shell.execute_reply.started":"2022-10-07T05:03:30.138184Z","shell.execute_reply":"2022-10-07T05:03:32.487366Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"monet_ds_norm = monet_ds.map(lambda x : normalization_layer(x))","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:03:49.346791Z","iopub.execute_input":"2022-10-07T05:03:49.347146Z","iopub.status.idle":"2022-10-07T05:03:49.371896Z","shell.execute_reply.started":"2022-10-07T05:03:49.347115Z","shell.execute_reply":"2022-10-07T05:03:49.371012Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"EPOCHS= 50\ngenerator = generator_model()\ndiscriminator = discriminator_model()\ntrain(monet_ds_norm,EPOCHS)","metadata":{"scrolled":true,"execution":{"iopub.status.busy":"2022-10-07T05:03:52.617367Z","iopub.execute_input":"2022-10-07T05:03:52.617795Z","iopub.status.idle":"2022-10-07T05:06:35.933564Z","shell.execute_reply.started":"2022-10-07T05:03:52.617758Z","shell.execute_reply":"2022-10-07T05:06:35.932558Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(1,1,is_norm=True)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:06:39.736737Z","iopub.execute_input":"2022-10-07T05:06:39.737106Z","iopub.status.idle":"2022-10-07T05:06:39.886925Z","shell.execute_reply.started":"2022-10-07T05:06:39.737067Z","shell.execute_reply":"2022-10-07T05:06:39.885408Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"tutorial_ds_norm = tutorial_ds.map(lambda x : normalization_layer(x))","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:06:44.011693Z","iopub.execute_input":"2022-10-07T05:06:44.012051Z","iopub.status.idle":"2022-10-07T05:06:44.025252Z","shell.execute_reply.started":"2022-10-07T05:06:44.012020Z","shell.execute_reply":"2022-10-07T05:06:44.024327Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# I have to re-execute definition of train_step() function to retrain another model of different dimensions  without making `tf.config.run_functions_eagerly(True)` and losing performance\n\n[Relevant github issue](https://github.com/tensorflow/tensorflow/issues/27120)\n\n[Relevant thread explanation which worked](https://github.com/tensorflow/tensorflow/issues/27120#issuecomment-500971183)","metadata":{}},{"cell_type":"code","source":"# tf.config.run_functions_eagerly(True)\n\n@tf.function\ndef train_step(image_batch, train_dim):\n    noise = tf.random.normal([1,train_dim])\n    \n    with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:            \n        generated_image = generator(noise, training=True)\n\n        discriminator_pred_gen = discriminator(generated_image, training=True)\n\n        discriminator_pred_ori= discriminator(image_batch, training=True)\n\n\n        discriminator_loss = cross_entropy_loss(tf.ones_like(discriminator_pred_ori),discriminator_pred_ori) + cross_entropy_loss(tf.zeros_like(discriminator_pred_gen), discriminator_pred_gen)\n        generator_loss = cross_entropy_loss(tf.ones_like(discriminator_pred_gen), discriminator_pred_gen)\n        \n    generator_grad = gen_tape.gradient(generator_loss, generator.trainable_variables)\n    discriminator_grad = disc_tape.gradient(discriminator_loss, discriminator.trainable_variables)\n\n    generator_optimizer.apply_gradients(zip(generator_grad,generator.trainable_variables))\n    discriminator_optimizer.apply_gradients(zip(discriminator_grad,discriminator.trainable_variables)) \n    \ndef train(dataset, epochs, train_dim=100):\n    for epoch in range(epochs):\n        start = time.time()\n        \n        for image_batch in dataset:\n            train_step(image_batch, train_dim)\n        \n        print(f\"Time for epoch {epoch+ 1} is : {time.time()-start}\")","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:30:24.252062Z","iopub.execute_input":"2022-10-07T05:30:24.252435Z","iopub.status.idle":"2022-10-07T05:30:24.257115Z","shell.execute_reply.started":"2022-10-07T05:30:24.252402Z","shell.execute_reply":"2022-10-07T05:30:24.255873Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"EPOCHS= 50\ngenerator= tutorial_generator_model()\ndiscriminator = tutorial_discriminator_model()\n\ntrain(tutorial_ds,EPOCHS)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:13:21.636593Z","iopub.execute_input":"2022-10-07T05:13:21.636951Z","iopub.status.idle":"2022-10-07T05:26:21.044519Z","shell.execute_reply.started":"2022-10-07T05:13:21.636919Z","shell.execute_reply":"2022-10-07T05:26:21.042552Z"},"scrolled":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(1,1)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:30:34.472230Z","iopub.execute_input":"2022-10-07T05:30:34.472841Z","iopub.status.idle":"2022-10-07T05:30:34.563644Z","shell.execute_reply.started":"2022-10-07T05:30:34.472806Z","shell.execute_reply":"2022-10-07T05:30:34.561973Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(3,3)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:30:53.912295Z","iopub.execute_input":"2022-10-07T05:30:53.913229Z","iopub.status.idle":"2022-10-07T05:30:54.387246Z","shell.execute_reply.started":"2022-10-07T05:30:53.913179Z","shell.execute_reply":"2022-10-07T05:30:54.386209Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"","metadata":{}},{"cell_type":"code","source":"EPOCHS= 50\ngenerator= tutorial_generator_model()\ndiscriminator = tutorial_discriminator_model()\ntrain(tutorial_ds_norm,EPOCHS)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:31:54.547534Z","iopub.execute_input":"2022-10-07T05:31:54.547906Z","iopub.status.idle":"2022-10-07T05:33:09.933567Z","shell.execute_reply.started":"2022-10-07T05:31:54.547875Z","shell.execute_reply":"2022-10-07T05:33:09.932574Z"},"scrolled":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(1,1,is_norm=True)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:35:04.212418Z","iopub.execute_input":"2022-10-07T05:35:04.212802Z","iopub.status.idle":"2022-10-07T05:35:04.305268Z","shell.execute_reply.started":"2022-10-07T05:35:04.212771Z","shell.execute_reply":"2022-10-07T05:35:04.304084Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(3,3)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:35:19.621784Z","iopub.execute_input":"2022-10-07T05:35:19.622155Z","iopub.status.idle":"2022-10-07T05:35:20.345195Z","shell.execute_reply.started":"2022-10-07T05:35:19.622122Z","shell.execute_reply":"2022-10-07T05:35:20.344253Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def updated_generator_model():\n    model = tf.keras.Sequential()\n    model.add(layers.Dense(4*4*128, use_bias=False, input_shape=(128,)))\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU(alpha=0.2))\n\n    model.add(layers.Reshape((4,4,128)))\n\n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2DTranspose(128,4, strides=(2,2), padding= 'same', kernel_initializer= initializer, use_bias= False))\n#     print(model.output_shape)\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2DTranspose(128,4, strides=(2,2), padding= 'same', kernel_initializer= initializer, use_bias= False))\n#     print(model.output_shape)\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2DTranspose(128//2,4, strides=(2,2), padding= 'same', kernel_initializer= initializer, use_bias= False))\n#     print(model.output_shape)\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2DTranspose(128//4, 4, strides=(2,2), padding= 'same', kernel_initializer= initializer, use_bias= False))\n#     print(model.output_shape)\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2DTranspose(128//8, 4, strides=(2,2), padding= 'same', kernel_initializer= initializer, use_bias= False))\n#     print(model.output_shape)\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n\n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2DTranspose(3,4, strides=(2,2), padding= 'same', kernel_initializer= initializer, use_bias= False, activation='tanh'))\n#     print(model.output_shape)\n    model.add(layers.BatchNormalization())\n    model.add(layers.LeakyReLU())\n    \n    return model","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:35:28.492398Z","iopub.execute_input":"2022-10-07T05:35:28.493098Z","iopub.status.idle":"2022-10-07T05:35:28.506884Z","shell.execute_reply.started":"2022-10-07T05:35:28.493061Z","shell.execute_reply":"2022-10-07T05:35:28.505717Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"noise = tf.random.normal([1,128])\ngenerator = updated_generator_model()\ngenerated_image = generator(noise, training=False)\nprint(generated_image.shape)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:35:34.812627Z","iopub.execute_input":"2022-10-07T05:35:34.812988Z","iopub.status.idle":"2022-10-07T05:35:35.006735Z","shell.execute_reply.started":"2022-10-07T05:35:34.812957Z","shell.execute_reply":"2022-10-07T05:35:35.005728Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(1,1)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:35:38.516721Z","iopub.execute_input":"2022-10-07T05:35:38.517090Z","iopub.status.idle":"2022-10-07T05:35:38.739307Z","shell.execute_reply.started":"2022-10-07T05:35:38.517057Z","shell.execute_reply":"2022-10-07T05:35:38.738231Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def updated_discriminator_model():\n    model = tf.keras.Sequential()\n    \n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2D(64, 4, strides=(2, 2), padding='same', kernel_initializer=initializer,\n                                     input_shape=[256, 256, 3], use_bias=False))\n    \n    model.add(layers.LeakyReLU())\n    model.add(layers.Dropout(0.3))\n    \n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2D(128, 4, strides=(2, 2), padding='same', kernel_initializer=initializer, use_bias=False))\n    \n    model.add(layers.LeakyReLU())\n    model.add(layers.Dropout(0.3))\n    \n    initializer = tf.random_normal_initializer(0., 0.02)\n    model.add(layers.Conv2D(256, 4, strides=(2, 2), padding='same', kernel_initializer=initializer, use_bias=False))\n    \n    model.add(layers.LeakyReLU())\n    model.add(layers.Dropout(0.3))\n    \n    model.add(layers.ZeroPadding2D())\n    initializer = tf.random_normal_initializer(0., 0.02)\n    \n    model.add(layers.Conv2D(512, 4, strides=1, kernel_initializer=initializer, use_bias=False))\n    model.add(layers.LeakyReLU())\n    model.add(layers.ZeroPadding2D())\n    \n    model.add(layers.Conv2D(1, 4, strides=1, kernel_initializer=initializer, use_bias=False))\n    model.add(layers.LeakyReLU(alpha=0.2))\n    \n    model.add(layers.Flatten())\n    model.add(layers.Dense(1, activation='sigmoid'))\n    \n    return model\n    ","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:35:43.246627Z","iopub.execute_input":"2022-10-07T05:35:43.247752Z","iopub.status.idle":"2022-10-07T05:35:43.259172Z","shell.execute_reply.started":"2022-10-07T05:35:43.247705Z","shell.execute_reply":"2022-10-07T05:35:43.257991Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"discriminator = updated_discriminator_model()\npreds = discriminator(generated_image, training=False)\nprint(preds)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:35:47.901744Z","iopub.execute_input":"2022-10-07T05:35:47.902116Z","iopub.status.idle":"2022-10-07T05:35:48.023784Z","shell.execute_reply.started":"2022-10-07T05:35:47.902082Z","shell.execute_reply":"2022-10-07T05:35:48.022763Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generator = updated_generator_model()\ndiscriminator = updated_discriminator_model()","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:35:59.062728Z","iopub.execute_input":"2022-10-07T05:35:59.063157Z","iopub.status.idle":"2022-10-07T05:35:59.301223Z","shell.execute_reply.started":"2022-10-07T05:35:59.063119Z","shell.execute_reply":"2022-10-07T05:35:59.300062Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Re-executing same train_step functions for issue already mentioned above.","metadata":{}},{"cell_type":"code","source":"@tf.function\ndef train_step(image_batch, train_dim):\n    noise = tf.random.normal([1,train_dim])\n    \n    with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:            \n        generated_image = generator(noise, training=True)\n\n        discriminator_pred_gen = discriminator(generated_image, training=True)\n\n        discriminator_pred_ori= discriminator(image_batch, training=True)\n\n\n        discriminator_loss = cross_entropy_loss(tf.ones_like(discriminator_pred_ori),discriminator_pred_ori) + cross_entropy_loss(tf.zeros_like(discriminator_pred_gen), discriminator_pred_gen)\n        generator_loss = cross_entropy_loss(tf.ones_like(discriminator_pred_gen), discriminator_pred_gen)\n        \n    generator_grad = gen_tape.gradient(generator_loss, generator.trainable_variables)\n    discriminator_grad = disc_tape.gradient(discriminator_loss, discriminator.trainable_variables)\n\n    generator_optimizer.apply_gradients(zip(generator_grad,generator.trainable_variables))\n    discriminator_optimizer.apply_gradients(zip(discriminator_grad,discriminator.trainable_variables))    \n    \ndef train(dataset, epochs, train_dim=100):\n    for epoch in range(epochs):\n        start = time.time()\n        \n        for image_batch in dataset:\n            train_step(image_batch, train_dim)\n        \n        print(f\"Time for epoch {epoch+ 1} is : {time.time()-start}\")","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:38:29.437706Z","iopub.execute_input":"2022-10-07T05:38:29.438083Z","iopub.status.idle":"2022-10-07T05:38:29.473962Z","shell.execute_reply.started":"2022-10-07T05:38:29.438051Z","shell.execute_reply":"2022-10-07T05:38:29.473049Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"EPOCHS= 50\nTRAIN_DIM= 128\ncross_entropy_loss = keras.losses.BinaryCrossentropy()\ntrain(monet_ds,EPOCHS, TRAIN_DIM)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:38:40.067207Z","iopub.execute_input":"2022-10-07T05:38:40.067902Z","iopub.status.idle":"2022-10-07T05:42:51.222099Z","shell.execute_reply.started":"2022-10-07T05:38:40.067865Z","shell.execute_reply":"2022-10-07T05:42:51.220966Z"},"scrolled":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(1,1, TRAIN_DIM)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:43:13.236763Z","iopub.execute_input":"2022-10-07T05:43:13.237148Z","iopub.status.idle":"2022-10-07T05:43:13.383139Z","shell.execute_reply.started":"2022-10-07T05:43:13.237113Z","shell.execute_reply":"2022-10-07T05:43:13.382234Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(5,5, TRAIN_DIM)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:43:18.966710Z","iopub.execute_input":"2022-10-07T05:43:18.967085Z","iopub.status.idle":"2022-10-07T05:43:21.270863Z","shell.execute_reply.started":"2022-10-07T05:43:18.967050Z","shell.execute_reply":"2022-10-07T05:43:21.269935Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"@tf.function\ndef train_step(image_batch, train_dim):\n    noise = tf.random.normal([1,train_dim])\n    \n    with tf.GradientTape() as gen_tape, tf.GradientTape() as disc_tape:            \n        generated_image = generator(noise, training=True)\n\n        discriminator_pred_gen = discriminator(generated_image, training=True)\n\n        discriminator_pred_ori= discriminator(image_batch, training=True)\n\n\n        discriminator_loss = cross_entropy_loss(tf.ones_like(discriminator_pred_ori),discriminator_pred_ori) + cross_entropy_loss(tf.zeros_like(discriminator_pred_gen), discriminator_pred_gen)\n        generator_loss = cross_entropy_loss(tf.ones_like(discriminator_pred_gen), discriminator_pred_gen)\n        \n    generator_grad = gen_tape.gradient(generator_loss, generator.trainable_variables)\n    discriminator_grad = disc_tape.gradient(discriminator_loss, discriminator.trainable_variables)\n\n    generator_optimizer.apply_gradients(zip(generator_grad,generator.trainable_variables))\n    discriminator_optimizer.apply_gradients(zip(discriminator_grad,discriminator.trainable_variables))    \n    \ndef train(dataset, epochs, train_dim=100):\n    for epoch in range(epochs):\n        start = time.time()\n        \n        for image_batch in dataset:\n            train_step(image_batch, train_dim)\n        \n        print(f\"Time for epoch {epoch+ 1} is : {time.time()-start}\")","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:44:08.282800Z","iopub.execute_input":"2022-10-07T05:44:08.283199Z","iopub.status.idle":"2022-10-07T05:44:08.294219Z","shell.execute_reply.started":"2022-10-07T05:44:08.283164Z","shell.execute_reply":"2022-10-07T05:44:08.293123Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"EPOCHS= 50\nTRAIN_DIM= 128\n\ngenerator = updated_generator_model()\ndiscriminator = updated_discriminator_model()\n\ncross_entropy_loss = keras.losses.BinaryCrossentropy()\n\ngenerator = updated_generator_model()\ndiscriminator = updated_discriminator_model()\n\ntrain(monet_ds_norm,EPOCHS, TRAIN_DIM)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:44:12.251749Z","iopub.execute_input":"2022-10-07T05:44:12.252134Z","iopub.status.idle":"2022-10-07T05:48:23.744062Z","shell.execute_reply.started":"2022-10-07T05:44:12.252100Z","shell.execute_reply":"2022-10-07T05:48:23.743071Z"},"scrolled":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"generate_images(1,1, TRAIN_DIM,is_norm=True)","metadata":{"execution":{"iopub.status.busy":"2022-10-07T05:48:27.921899Z","iopub.execute_input":"2022-10-07T05:48:27.922360Z","iopub.status.idle":"2022-10-07T05:48:28.116360Z","shell.execute_reply.started":"2022-10-07T05:48:27.922317Z","shell.execute_reply":"2022-10-07T05:48:28.115209Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Not Working ??","metadata":{}},{"cell_type":"markdown","source":"# References\n## 1. [Official Competition Starter tutorial](https://www.kaggle.com/code/amyjang/monet-cyclegan-tutorial/notebook)\n## 2. [Tensorflow DCGAN tutorial](https://www.tensorflow.org/tutorials/generative/dcgan)","metadata":{}}]}