Deep Autoencoders in Keras - conv-tf2-v3-bc.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# https://blog.keras.io/building-autoencoders-in-keras.html\n",
    "import tensorflow as tf\n",
    "Model           = tf.keras.models.Model\n",
    "Input           = tf.keras.layers.Input\n",
    "#Dense           = tf.keras.layers.Dense\n",
    "Conv2D          = tf.keras.layers.Conv2D\n",
    "MaxPooling2D    = tf.keras.layers.MaxPooling2D\n",
    "UpSampling2D    = tf.keras.layers.UpSampling2D\n",
    "mnist           = tf.keras.datasets.mnist\n",
    "save_model      = tf.keras.models.save_model\n",
    "load_model      = tf.keras.models.load_model\n",
    "TensorBoard     = tf.keras.callbacks.TensorBoard\n",
    "ModelCheckpoint = tf.keras.callbacks.ModelCheckpoint"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# https://www.pyimagesearch.com/2018/12/31/keras-conv2d-and-convolutional-layers/\n",
    "# Conv2D(batch_size, (input_shape), activation, padding) # more batch size in deeper layer\n",
    "# MaxPooling2D((input_shape), padding) # padding 'valid' = no padding, padding 'same' = preserve dimension + padding\n",
    "\n",
    "# Autoencoder Architecture\n",
    "input_img = Input(shape=(28, 28, 1)) # change this to other than 28*28 may show \"multiple\" shape\n",
    "x = Conv2D(16, (3, 3), activation='relu', padding='same')(input_img)\n",
    "x = MaxPooling2D((2, 2), padding='same')(x)\n",
    "x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)\n",
    "x = MaxPooling2D((2, 2), padding='same')(x)\n",
    "x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)\n",
    "encoded = MaxPooling2D((2, 2), padding='same')(x) # at this point the representation is (4, 4, 8) i.e. 128-dimensional\n",
    "\n",
    "x = Conv2D(8, (3, 3), activation='relu', padding='same')(encoded)\n",
    "x = UpSampling2D((2, 2))(x)\n",
    "x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)\n",
    "x = UpSampling2D((2, 2))(x)\n",
    "x = Conv2D(16, (3, 3), activation='relu')(x) # may want to intentionally drop padding here for other shape\n",
    "x = UpSampling2D((2, 2))(x)\n",
    "decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)\n",
    "\n",
    "autoencoder = Model(input_img, decoded)\n",
    "autoencoder.compile(optimizer='adam', loss='binary_crossentropy') # use binary_crossentropy, mean_squared_error doesnt work\n",
    "\n",
    "# Encoder Model\n",
    "encoder = Model(input_img, encoded)\n",
    "\n",
    "# Decoder Architecture\n",
    "Dlayer_input = Input(shape=(4,4,8)) # use model.summary() to check\n",
    "x = autoencoder.layers[-7](Dlayer_input)\n",
    "x = autoencoder.layers[-6](x)\n",
    "x = autoencoder.layers[-5](x)\n",
    "x = autoencoder.layers[-4](x)\n",
    "x = autoencoder.layers[-3](x)\n",
    "x = autoencoder.layers[-2](x)\n",
    "Dlayer_output = autoencoder.layers[-1](x) # extract last layer\n",
    "\n",
    "# Decoder Model\n",
    "decoder = Model(Dlayer_input, Dlayer_output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         [(None, 28, 28, 1)]       0         \n",
      "_________________________________________________________________\n",
      "conv2d (Conv2D)              (None, 28, 28, 16)        160       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 14, 14, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 14, 14, 8)         1160      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 7, 7, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 7, 7, 8)           584       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 4, 4, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 4, 4, 8)           584       \n",
      "_________________________________________________________________\n",
      "up_sampling2d (UpSampling2D) (None, 8, 8, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 8, 8, 8)           584       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_1 (UpSampling2 (None, 16, 16, 8)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_5 (Conv2D)            (None, 14, 14, 16)        1168      \n",
      "_________________________________________________________________\n",
      "up_sampling2d_2 (UpSampling2 (None, 28, 28, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_6 (Conv2D)            (None, 28, 28, 1)         145       \n",
      "=================================================================\n",
      "Total params: 4,385\n",
      "Trainable params: 4,385\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Model: \"model_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         [(None, 28, 28, 1)]       0         \n",
      "_________________________________________________________________\n",
      "conv2d (Conv2D)              (None, 28, 28, 16)        160       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 14, 14, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 14, 14, 8)         1160      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 7, 7, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 7, 7, 8)           584       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 4, 4, 8)           0         \n",
      "=================================================================\n",
      "Total params: 1,904\n",
      "Trainable params: 1,904\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Model: \"model_2\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         [(None, 4, 4, 8)]         0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 4, 4, 8)           584       \n",
      "_________________________________________________________________\n",
      "up_sampling2d (UpSampling2D) (None, 8, 8, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 8, 8, 8)           584       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_1 (UpSampling2 (None, 16, 16, 8)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_5 (Conv2D)            (None, 14, 14, 16)        1168      \n",
      "_________________________________________________________________\n",
      "up_sampling2d_2 (UpSampling2 (None, 28, 28, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_6 (Conv2D)            (None, 28, 28, 1)         145       \n",
      "=================================================================\n",
      "Total params: 2,481\n",
      "Trainable params: 2,481\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "autoencoder.summary()\n",
    "encoder.summary()\n",
    "decoder.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "60000\n",
      "10000\n",
      "x_train shape = (60000, 28, 28, 1)\n",
      "x_test shape = (10000, 28, 28, 1)\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "# Load mnist dataset\n",
    "(x_train, y_train),(x_test, y_test) = mnist.load_data()\n",
    "\n",
    "print(len(x_train))\n",
    "print(len(x_test))\n",
    "\n",
    "# Preprocess\n",
    "x_train = x_train.astype('float32')/255 # normalise to 0 and 1\n",
    "x_test  = x_test.astype('float32')/255\n",
    "#x_train = x_train.reshape(len(x_train), (np.prod(x_train.shape[1:]))) # 28x28 flatten into 1x784\n",
    "#x_test  = x_test.reshape(len(x_test), (np.prod(x_test.shape[1:])))\n",
    "x_train = np.reshape(x_train, (len(x_train), 28, 28, 1))  # adapt this if using `channels_first` image data format\n",
    "x_test = np.reshape(x_test, (len(x_test), 28, 28, 1))  # adapt this if using `channels_first` image data format\n",
    "\n",
    "print(f'x_train shape = {x_train.shape}') # 60000 images 784 vectors = 60000 row 784 col\n",
    "print(f'x_test shape = {x_test.shape}') # 10000 images 784 vectors = 10000 row 784 col"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "  2/469 [..............................] - ETA: 1:41 - loss: 0.6869WARNING:tensorflow:Method (on_train_batch_end) is slow compared to the batch update (0.146662). Check your callbacks.\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.2102\n",
      "Epoch 00001: val_loss improved from inf to 0.14225, saving model to autoencoder-bc/checkpoint/model.01-0.14.h5\n",
      "469/469 [==============================] - 85s 182ms/step - loss: 0.2102 - val_loss: 0.1423\n",
      "Epoch 2/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1335\n",
      "Epoch 00002: val_loss improved from 0.14225 to 0.12504, saving model to autoencoder-bc/checkpoint/model.02-0.13.h5\n",
      "469/469 [==============================] - 81s 173ms/step - loss: 0.1335 - val_loss: 0.1250\n",
      "Epoch 3/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1220\n",
      "Epoch 00003: val_loss improved from 0.12504 to 0.11719, saving model to autoencoder-bc/checkpoint/model.03-0.12.h5\n",
      "469/469 [==============================] - 86s 183ms/step - loss: 0.1220 - val_loss: 0.1172\n",
      "Epoch 4/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1159\n",
      "Epoch 00004: val_loss improved from 0.11719 to 0.11270, saving model to autoencoder-bc/checkpoint/model.04-0.11.h5\n",
      "469/469 [==============================] - 95s 203ms/step - loss: 0.1159 - val_loss: 0.1127\n",
      "Epoch 5/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1117\n",
      "Epoch 00005: val_loss improved from 0.11270 to 0.10909, saving model to autoencoder-bc/checkpoint/model.05-0.11.h5\n",
      "469/469 [==============================] - 87s 185ms/step - loss: 0.1117 - val_loss: 0.1091\n",
      "Epoch 6/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1083\n",
      "Epoch 00006: val_loss improved from 0.10909 to 0.10601, saving model to autoencoder-bc/checkpoint/model.06-0.11.h5\n",
      "469/469 [==============================] - 83s 177ms/step - loss: 0.1083 - val_loss: 0.1060\n",
      "Epoch 7/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1059\n",
      "Epoch 00007: val_loss improved from 0.10601 to 0.10372, saving model to autoencoder-bc/checkpoint/model.07-0.10.h5\n",
      "469/469 [==============================] - 93s 197ms/step - loss: 0.1059 - val_loss: 0.1037\n",
      "Epoch 8/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1041\n",
      "Epoch 00008: val_loss improved from 0.10372 to 0.10216, saving model to autoencoder-bc/checkpoint/model.08-0.10.h5\n",
      "469/469 [==============================] - 83s 177ms/step - loss: 0.1041 - val_loss: 0.1022\n",
      "Epoch 9/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1026\n",
      "Epoch 00009: val_loss improved from 0.10216 to 0.10091, saving model to autoencoder-bc/checkpoint/model.09-0.10.h5\n",
      "469/469 [==============================] - 104s 221ms/step - loss: 0.1026 - val_loss: 0.1009\n",
      "Epoch 10/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1013\n",
      "Epoch 00010: val_loss improved from 0.10091 to 0.10007, saving model to autoencoder-bc/checkpoint/model.10-0.10.h5\n",
      "469/469 [==============================] - 149s 318ms/step - loss: 0.1013 - val_loss: 0.1001\n",
      "Epoch 11/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.1003\n",
      "Epoch 00011: val_loss improved from 0.10007 to 0.09902, saving model to autoencoder-bc/checkpoint/model.11-0.10.h5\n",
      "469/469 [==============================] - 125s 266ms/step - loss: 0.1003 - val_loss: 0.0990\n",
      "Epoch 12/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0993\n",
      "Epoch 00012: val_loss improved from 0.09902 to 0.09780, saving model to autoencoder-bc/checkpoint/model.12-0.10.h5\n",
      "469/469 [==============================] - 146s 311ms/step - loss: 0.0993 - val_loss: 0.0978\n",
      "Epoch 13/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0986\n",
      "Epoch 00013: val_loss improved from 0.09780 to 0.09715, saving model to autoencoder-bc/checkpoint/model.13-0.10.h5\n",
      "469/469 [==============================] - 145s 309ms/step - loss: 0.0986 - val_loss: 0.0972\n",
      "Epoch 14/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0979\n",
      "Epoch 00014: val_loss improved from 0.09715 to 0.09635, saving model to autoencoder-bc/checkpoint/model.14-0.10.h5\n",
      "469/469 [==============================] - 152s 324ms/step - loss: 0.0979 - val_loss: 0.0963\n",
      "Epoch 15/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0972\n",
      "Epoch 00015: val_loss did not improve from 0.09635\n",
      "469/469 [==============================] - 131s 279ms/step - loss: 0.0972 - val_loss: 0.0966\n",
      "Epoch 16/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0966\n",
      "Epoch 00016: val_loss improved from 0.09635 to 0.09513, saving model to autoencoder-bc/checkpoint/model.16-0.10.h5\n",
      "469/469 [==============================] - 134s 285ms/step - loss: 0.0966 - val_loss: 0.0951\n",
      "Epoch 17/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0961\n",
      "Epoch 00017: val_loss did not improve from 0.09513\n",
      "469/469 [==============================] - 129s 276ms/step - loss: 0.0961 - val_loss: 0.0954\n",
      "Epoch 18/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0956\n",
      "Epoch 00018: val_loss improved from 0.09513 to 0.09429, saving model to autoencoder-bc/checkpoint/model.18-0.09.h5\n",
      "469/469 [==============================] - 128s 274ms/step - loss: 0.0956 - val_loss: 0.0943\n",
      "Epoch 19/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0951\n",
      "Epoch 00019: val_loss improved from 0.09429 to 0.09382, saving model to autoencoder-bc/checkpoint/model.19-0.09.h5\n",
      "469/469 [==============================] - 113s 240ms/step - loss: 0.0951 - val_loss: 0.0938\n",
      "Epoch 20/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0947- ETA: 2s -\n",
      "Epoch 00020: val_loss did not improve from 0.09382\n",
      "469/469 [==============================] - 105s 224ms/step - loss: 0.0947 - val_loss: 0.0949\n",
      "Epoch 21/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0944\n",
      "Epoch 00021: val_loss improved from 0.09382 to 0.09304, saving model to autoencoder-bc/checkpoint/model.21-0.09.h5\n",
      "469/469 [==============================] - 117s 251ms/step - loss: 0.0944 - val_loss: 0.0930\n",
      "Epoch 22/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0941\n",
      "Epoch 00022: val_loss improved from 0.09304 to 0.09270, saving model to autoencoder-bc/checkpoint/model.22-0.09.h5\n",
      "469/469 [==============================] - 159s 338ms/step - loss: 0.0941 - val_loss: 0.0927\n",
      "Epoch 23/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0937\n",
      "Epoch 00023: val_loss did not improve from 0.09270\n",
      "469/469 [==============================] - 159s 340ms/step - loss: 0.0937 - val_loss: 0.0928\n",
      "Epoch 24/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0935\n",
      "Epoch 00024: val_loss improved from 0.09270 to 0.09228, saving model to autoencoder-bc/checkpoint/model.24-0.09.h5\n",
      "469/469 [==============================] - 129s 275ms/step - loss: 0.0935 - val_loss: 0.0923\n",
      "Epoch 25/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0932\n",
      "Epoch 00025: val_loss improved from 0.09228 to 0.09212, saving model to autoencoder-bc/checkpoint/model.25-0.09.h5\n",
      "469/469 [==============================] - 137s 293ms/step - loss: 0.0932 - val_loss: 0.0921\n",
      "Epoch 26/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0931\n",
      "Epoch 00026: val_loss improved from 0.09212 to 0.09193, saving model to autoencoder-bc/checkpoint/model.26-0.09.h5\n",
      "469/469 [==============================] - 142s 304ms/step - loss: 0.0931 - val_loss: 0.0919\n",
      "Epoch 27/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0928\n",
      "Epoch 00027: val_loss did not improve from 0.09193\n",
      "469/469 [==============================] - 156s 333ms/step - loss: 0.0928 - val_loss: 0.0921\n",
      "Epoch 28/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0926\n",
      "Epoch 00028: val_loss improved from 0.09193 to 0.09175, saving model to autoencoder-bc/checkpoint/model.28-0.09.h5\n",
      "469/469 [==============================] - 125s 266ms/step - loss: 0.0926 - val_loss: 0.0918\n",
      "Epoch 29/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0925\n",
      "Epoch 00029: val_loss improved from 0.09175 to 0.09124, saving model to autoencoder-bc/checkpoint/model.29-0.09.h5\n",
      "469/469 [==============================] - 116s 248ms/step - loss: 0.0925 - val_loss: 0.0912\n",
      "Epoch 30/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0923\n",
      "Epoch 00030: val_loss did not improve from 0.09124\n",
      "469/469 [==============================] - 135s 288ms/step - loss: 0.0923 - val_loss: 0.0913\n",
      "Epoch 31/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0922\n",
      "Epoch 00031: val_loss improved from 0.09124 to 0.09108, saving model to autoencoder-bc/checkpoint/model.31-0.09.h5\n",
      "469/469 [==============================] - 114s 242ms/step - loss: 0.0922 - val_loss: 0.0911\n",
      "Epoch 32/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0920\n",
      "Epoch 00032: val_loss improved from 0.09108 to 0.09099, saving model to autoencoder-bc/checkpoint/model.32-0.09.h5\n",
      "469/469 [==============================] - 141s 300ms/step - loss: 0.0920 - val_loss: 0.0910\n",
      "Epoch 33/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0919\n",
      "Epoch 00033: val_loss improved from 0.09099 to 0.09085, saving model to autoencoder-bc/checkpoint/model.33-0.09.h5\n",
      "469/469 [==============================] - 140s 299ms/step - loss: 0.0919 - val_loss: 0.0908\n",
      "Epoch 34/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0919\n",
      "Epoch 00034: val_loss did not improve from 0.09085\n",
      "469/469 [==============================] - 121s 258ms/step - loss: 0.0919 - val_loss: 0.0909\n",
      "Epoch 35/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0918\n",
      "Epoch 00035: val_loss improved from 0.09085 to 0.09061, saving model to autoencoder-bc/checkpoint/model.35-0.09.h5\n",
      "469/469 [==============================] - 113s 241ms/step - loss: 0.0918 - val_loss: 0.0906\n",
      "Epoch 36/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0916\n",
      "Epoch 00036: val_loss did not improve from 0.09061\n",
      "469/469 [==============================] - 110s 235ms/step - loss: 0.0916 - val_loss: 0.0912\n",
      "Epoch 37/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0915\n",
      "Epoch 00037: val_loss improved from 0.09061 to 0.09035, saving model to autoencoder-bc/checkpoint/model.37-0.09.h5\n",
      "469/469 [==============================] - 96s 205ms/step - loss: 0.0915 - val_loss: 0.0904\n",
      "Epoch 38/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0914\n",
      "Epoch 00038: val_loss improved from 0.09035 to 0.09031, saving model to autoencoder-bc/checkpoint/model.38-0.09.h5\n",
      "469/469 [==============================] - 117s 249ms/step - loss: 0.0914 - val_loss: 0.0903\n",
      "Epoch 39/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0913\n",
      "Epoch 00039: val_loss did not improve from 0.09031\n",
      "469/469 [==============================] - 102s 218ms/step - loss: 0.0913 - val_loss: 0.0905\n",
      "Epoch 40/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0914\n",
      "Epoch 00040: val_loss did not improve from 0.09031\n",
      "469/469 [==============================] - 128s 273ms/step - loss: 0.0914 - val_loss: 0.0903\n",
      "Epoch 41/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0912\n",
      "Epoch 00041: val_loss improved from 0.09031 to 0.09005, saving model to autoencoder-bc/checkpoint/model.41-0.09.h5\n",
      "469/469 [==============================] - 113s 242ms/step - loss: 0.0912 - val_loss: 0.0900\n",
      "Epoch 42/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0911\n",
      "Epoch 00042: val_loss did not improve from 0.09005\n",
      "469/469 [==============================] - 105s 225ms/step - loss: 0.0911 - val_loss: 0.0905\n",
      "Epoch 43/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0910\n",
      "Epoch 00043: val_loss did not improve from 0.09005\n",
      "469/469 [==============================] - 119s 254ms/step - loss: 0.0910 - val_loss: 0.0903\n",
      "Epoch 44/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0909\n",
      "Epoch 00044: val_loss did not improve from 0.09005\n",
      "469/469 [==============================] - 123s 263ms/step - loss: 0.0909 - val_loss: 0.0901\n",
      "Epoch 45/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0909\n",
      "Epoch 00045: val_loss improved from 0.09005 to 0.08981, saving model to autoencoder-bc/checkpoint/model.45-0.09.h5\n",
      "469/469 [==============================] - 113s 240ms/step - loss: 0.0909 - val_loss: 0.0898\n",
      "Epoch 46/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0907\n",
      "Epoch 00046: val_loss improved from 0.08981 to 0.08971, saving model to autoencoder-bc/checkpoint/model.46-0.09.h5\n",
      "469/469 [==============================] - 131s 280ms/step - loss: 0.0907 - val_loss: 0.0897\n",
      "Epoch 47/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0907\n",
      "Epoch 00047: val_loss did not improve from 0.08971\n",
      "469/469 [==============================] - 111s 238ms/step - loss: 0.0907 - val_loss: 0.0900\n",
      "Epoch 48/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0907\n",
      "Epoch 00048: val_loss improved from 0.08971 to 0.08955, saving model to autoencoder-bc/checkpoint/model.48-0.09.h5\n",
      "469/469 [==============================] - 100s 213ms/step - loss: 0.0907 - val_loss: 0.0895\n",
      "Epoch 49/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0906\n",
      "Epoch 00049: val_loss did not improve from 0.08955\n",
      "469/469 [==============================] - 98s 209ms/step - loss: 0.0906 - val_loss: 0.0898\n",
      "Epoch 50/50\n",
      "469/469 [==============================] - ETA: 0s - loss: 0.0905- ETA: 2s - l\n",
      "Epoch 00050: val_loss improved from 0.08955 to 0.08934, saving model to autoencoder-bc/checkpoint/model.50-0.09.h5\n",
      "469/469 [==============================] - 100s 213ms/step - loss: 0.0905 - val_loss: 0.0893\n",
      "Wall time: 1h 38min 59s\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x2b0015b2e88>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "%%time\n",
    "import os\n",
    "\n",
    "try:\n",
    "    os.makedirs(\"autoencoder-bc/logs\")\n",
    "except Exception:\n",
    "    pass\n",
    "\n",
    "try:\n",
    "    os.makedirs(\"autoencoder-bc/checkpoint\")\n",
    "except Exception:\n",
    "    pass\n",
    "\n",
    "# Run this in a terminal\n",
    "#tensorboard --logdir=autoencoder-bc/logs\n",
    "\n",
    "callbacks=[\n",
    "    TensorBoard(log_dir='autoencoder-bc/logs'),\n",
    "    ModelCheckpoint(filepath='autoencoder-bc/checkpoint/model.{epoch:02d}-{val_loss:.2f}.h5',\n",
    "                    verbose=1, save_best_only=True, mode='min')\n",
    "]\n",
    "\n",
    "# Fit data\n",
    "autoencoder.fit(x_train, x_train,\n",
    "                epochs=50,\n",
    "                batch_size=128,\n",
    "                shuffle=True,\n",
    "                validation_data=(x_test, x_test),\n",
    "                callbacks=callbacks)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "encoded_imgs = encoder.predict(x_test)\n",
    "decoded_imgs = decoder.predict(encoded_imgs)\n",
    "#decoded_imgs = autoencoder.predict(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(28, 28, 1)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(4, 4, 8)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(28, 28, 1)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# test single image input and check output\n",
    "from matplotlib import pyplot as plt\n",
    "plt.gray()\n",
    "\n",
    "# image number from dataset\n",
    "i = 100\n",
    "\n",
    "inputs = x_test\n",
    "encoded = encoded_imgs\n",
    "decoded = decoded_imgs\n",
    "\n",
    "# inputs\n",
    "print(inputs[i].shape)\n",
    "plt.figure(figsize=(20, 4))\n",
    "plt.imshow(inputs[i].reshape(28, 28), interpolation='nearest')\n",
    "plt.show()\n",
    "\n",
    "# test transpose (for RGB inputs)\n",
    "#plt.figure(figsize=(20, 4))\n",
    "#plt.imshow(inputs[i].T, interpolation='nearest') # TypeError: Invalid shape (3, 128, 128) for image data\n",
    "#plt.show()\n",
    "#plt.imshow(inputs[i].T.reshape(28*1,28).T, interpolation='nearest') # (width * channel, length)\n",
    "#plt.show()\n",
    "\n",
    "# encoded\n",
    "print(encoded[i].shape)\n",
    "plt.figure(figsize=(20, 4))\n",
    "plt.imshow(encoded[i].T.reshape(4*8, 4).T, interpolation='nearest') # for not flatten\n",
    "#plt.imshow(encoded[i].reshape(8*1, 8), interpolation='nearest') # for flatten\n",
    "plt.show()\n",
    "\n",
    "# decoded\n",
    "print(decoded[i].shape)\n",
    "plt.figure(figsize=(20, 4))\n",
    "plt.imshow(decoded[i].reshape(28, 28), interpolation='nearest')\n",
    "plt.show()\n",
    "\n",
    "del inputs, encoded, decoded"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 10 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x576 with 10 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 10 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 1.61 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20, 4))\n",
    "for i in range(n):\n",
    "    ax = plt.subplot(1, n, i+1)\n",
    "    plt.imshow(x_test[i].reshape(28, 28))\n",
    "    plt.gray()\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "\n",
    "plt.show()\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20, 8))\n",
    "for i in range(n):\n",
    "    # What does encoded images looks like?\n",
    "    ax = plt.subplot(1, n, i+1)\n",
    "    plt.imshow(encoded_imgs[i].T.reshape(4*8,4).T) # encoded 64 # (8,8) # (4, 4*4)\n",
    "    #plt.imshow(decoded_imgs[i].reshape(4, 14*14).T) # input or decoded 784 # (28,28) # (4, 14*14)\n",
    "    plt.gray()\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "\n",
    "plt.show()\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20, 4))\n",
    "for i in range(n):\n",
    "    ax = plt.subplot(1, n, i+1)\n",
    "    plt.imshow(decoded_imgs[i].reshape(28,28))\n",
    "    plt.gray()\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MSE = 0.0034, PSNR (dB) = 24.67\n",
      "MSE = 0.0092, PSNR (dB) = 20.36\n",
      "MSE = 0.0013, PSNR (dB) = 28.92\n",
      "MSE = 0.0076, PSNR (dB) = 21.17\n",
      "MSE = 0.0090, PSNR (dB) = 20.45\n",
      "MSE = 0.0009, PSNR (dB) = 30.54\n",
      "MSE = 0.0121, PSNR (dB) = 19.18\n",
      "MSE = 0.0114, PSNR (dB) = 19.44\n",
      "MSE = 0.0145, PSNR (dB) = 18.39\n",
      "MSE = 0.0063, PSNR (dB) = 22.03\n",
      "Wall time: 11 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# https://dsp.stackexchange.com/questions/38065/peak-signal-to-noise-ratio-psnr-in-python-for-an-image\n",
    "import math\n",
    "import numpy as np\n",
    "\n",
    "def mse(img1, img2):\n",
    "    return np.mean( (img1 - img2) ** 2 )\n",
    "\n",
    "def psnr(img1, img2):\n",
    "    mserr = mse(img1, img2)\n",
    "    PIXEL_MAX = 1.0\n",
    "    try:\n",
    "        return 20 * math.log10(PIXEL_MAX / math.sqrt(mserr))\n",
    "    except ZeroDivisionError:\n",
    "        return 'Same image'\n",
    "\n",
    "n = 10\n",
    "for i in range(n):\n",
    "    original = x_test[i]\n",
    "    contrast = decoded_imgs[i]\n",
    "    a = mse(original, contrast)\n",
    "    b = psnr(original, contrast) # can also use cv2.PSNR rather than defined function psnr\n",
    "    if b != 'Same image':\n",
    "        print(f'MSE = {a:.4f}, PSNR (dB) = {b:.2f}')\n",
    "    else:\n",
    "        print(f'MSE = {a:.4f}, PSNR too high')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9553\n",
      "0.9135\n",
      "0.9768\n",
      "0.9279\n",
      "0.8656\n",
      "0.9905\n",
      "0.8653\n",
      "0.8503\n",
      "0.8093\n",
      "0.9346\n",
      "Wall time: 145 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# https://github.com/imamun93/Image-Similarities-using-SSIM/blob/master/blog5_ssim.ipynb\n",
    "import numpy as np\n",
    "from tensorflow.image import ssim, rgb_to_yuv\n",
    "from tensorflow import convert_to_tensor\n",
    "\n",
    "n = 10\n",
    "for i in range(n):\n",
    "    original = x_test[i].reshape(1, 28, 28, 1)\n",
    "    contrast = decoded_imgs[i].reshape(1, 28, 28, 1)\n",
    "    a = np.asarray(ssim(convert_to_tensor(original), convert_to_tensor(contrast), max_val=1)) # <class 'tensorflow.python.framework.ops.EagerTensor'>\n",
    "    #a = np.asarray(ssim(rgb_to_yuv(convert_to_tensor(original)), rgb_to_yuv(convert_to_tensor(contrast)), max_val=1))\n",
    "    print(f'{a[0]:.4f}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9890\n",
      "0.9687\n",
      "0.9945\n",
      "0.9713\n",
      "0.9746\n",
      "0.9981\n",
      "0.9469\n",
      "0.9494\n",
      "0.9378\n",
      "0.9763\n",
      "Wall time: 110 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# https://github.com/tensorflow/models/blob/master/research/compression/image_encoder/msssim.py\n",
    "import numpy as np\n",
    "from scipy import signal\n",
    "from scipy.ndimage.filters import convolve\n",
    "\n",
    "def _FSpecialGauss(size, sigma):\n",
    "  \"\"\"Function to mimic the 'fspecial' gaussian MATLAB function.\"\"\"\n",
    "  radius = size // 2\n",
    "  offset = 0.0\n",
    "  start, stop = -radius, radius + 1\n",
    "  if size % 2 == 0:\n",
    "    offset = 0.5\n",
    "    stop -= 1\n",
    "  x, y = np.mgrid[offset + start:stop, offset + start:stop]\n",
    "  assert len(x) == size\n",
    "  g = np.exp(-((x**2 + y**2)/(2.0 * sigma**2)))\n",
    "  return g / g.sum()\n",
    "\n",
    "\n",
    "def _SSIMForMultiScale(img1, img2, max_val=255, filter_size=11,\n",
    "                       filter_sigma=1.5, k1=0.01, k2=0.03):\n",
    "  \"\"\"Return the Structural Similarity Map between `img1` and `img2`.\n",
    "  This function attempts to match the functionality of ssim_index_new.m by\n",
    "  Zhou Wang: http://www.cns.nyu.edu/~lcv/ssim/msssim.zip\n",
    "  Arguments:\n",
    "    img1: Numpy array holding the first RGB image batch.\n",
    "    img2: Numpy array holding the second RGB image batch.\n",
    "    max_val: the dynamic range of the images (i.e., the difference between the\n",
    "      maximum the and minimum allowed values).\n",
    "    filter_size: Size of blur kernel to use (will be reduced for small images).\n",
    "    filter_sigma: Standard deviation for Gaussian blur kernel (will be reduced\n",
    "      for small images).\n",
    "    k1: Constant used to maintain stability in the SSIM calculation (0.01 in\n",
    "      the original paper).\n",
    "    k2: Constant used to maintain stability in the SSIM calculation (0.03 in\n",
    "      the original paper).\n",
    "  Returns:\n",
    "    Pair containing the mean SSIM and contrast sensitivity between `img1` and\n",
    "    `img2`.\n",
    "  Raises:\n",
    "    RuntimeError: If input images don't have the same shape or don't have four\n",
    "      dimensions: [batch_size, height, width, depth].\n",
    "  \"\"\"\n",
    "  if img1.shape != img2.shape:\n",
    "    raise RuntimeError('Input images must have the same shape (%s vs. %s).',\n",
    "                       img1.shape, img2.shape)\n",
    "  if img1.ndim != 4:\n",
    "    raise RuntimeError('Input images must have four dimensions, not %d',\n",
    "                       img1.ndim)\n",
    "\n",
    "  img1 = img1.astype(np.float64)\n",
    "  img2 = img2.astype(np.float64)\n",
    "  _, height, width, _ = img1.shape\n",
    "\n",
    "  # Filter size can't be larger than height or width of images.\n",
    "  size = min(filter_size, height, width)\n",
    "\n",
    "  # Scale down sigma if a smaller filter size is used.\n",
    "  sigma = size * filter_sigma / filter_size if filter_size else 0\n",
    "\n",
    "  if filter_size:\n",
    "    window = np.reshape(_FSpecialGauss(size, sigma), (1, size, size, 1))\n",
    "    mu1 = signal.fftconvolve(img1, window, mode='valid')\n",
    "    mu2 = signal.fftconvolve(img2, window, mode='valid')\n",
    "    sigma11 = signal.fftconvolve(img1 * img1, window, mode='valid')\n",
    "    sigma22 = signal.fftconvolve(img2 * img2, window, mode='valid')\n",
    "    sigma12 = signal.fftconvolve(img1 * img2, window, mode='valid')\n",
    "  else:\n",
    "    # Empty blur kernel so no need to convolve.\n",
    "    mu1, mu2 = img1, img2\n",
    "    sigma11 = img1 * img1\n",
    "    sigma22 = img2 * img2\n",
    "    sigma12 = img1 * img2\n",
    "\n",
    "  mu11 = mu1 * mu1\n",
    "  mu22 = mu2 * mu2\n",
    "  mu12 = mu1 * mu2\n",
    "  sigma11 -= mu11\n",
    "  sigma22 -= mu22\n",
    "  sigma12 -= mu12\n",
    "\n",
    "  # Calculate intermediate values used by both ssim and cs_map.\n",
    "  c1 = (k1 * max_val) ** 2\n",
    "  c2 = (k2 * max_val) ** 2\n",
    "  v1 = 2.0 * sigma12 + c2\n",
    "  v2 = sigma11 + sigma22 + c2\n",
    "  ssim = np.mean((((2.0 * mu12 + c1) * v1) / ((mu11 + mu22 + c1) * v2)))\n",
    "  cs = np.mean(v1 / v2)\n",
    "  return ssim, cs\n",
    "\n",
    "\n",
    "def MultiScaleSSIM(img1, img2, max_val=255, filter_size=11, filter_sigma=1.5,\n",
    "                   k1=0.01, k2=0.03, weights=None):\n",
    "  \"\"\"Return the MS-SSIM score between `img1` and `img2`.\n",
    "  This function implements Multi-Scale Structural Similarity (MS-SSIM) Image\n",
    "  Quality Assessment according to Zhou Wang's paper, \"Multi-scale structural\n",
    "  similarity for image quality assessment\" (2003).\n",
    "  Link: https://ece.uwaterloo.ca/~z70wang/publications/msssim.pdf\n",
    "  Author's MATLAB implementation:\n",
    "  http://www.cns.nyu.edu/~lcv/ssim/msssim.zip\n",
    "  Arguments:\n",
    "    img1: Numpy array holding the first RGB image batch.\n",
    "    img2: Numpy array holding the second RGB image batch.\n",
    "    max_val: the dynamic range of the images (i.e., the difference between the\n",
    "      maximum the and minimum allowed values).\n",
    "    filter_size: Size of blur kernel to use (will be reduced for small images).\n",
    "    filter_sigma: Standard deviation for Gaussian blur kernel (will be reduced\n",
    "      for small images).\n",
    "    k1: Constant used to maintain stability in the SSIM calculation (0.01 in\n",
    "      the original paper).\n",
    "    k2: Constant used to maintain stability in the SSIM calculation (0.03 in\n",
    "      the original paper).\n",
    "    weights: List of weights for each level; if none, use five levels and the\n",
    "      weights from the original paper.\n",
    "  Returns:\n",
    "    MS-SSIM score between `img1` and `img2`.\n",
    "  Raises:\n",
    "    RuntimeError: If input images don't have the same shape or don't have four\n",
    "      dimensions: [batch_size, height, width, depth].\n",
    "  \"\"\"\n",
    "  if img1.shape != img2.shape:\n",
    "    raise RuntimeError('Input images must have the same shape (%s vs. %s).',\n",
    "                       img1.shape, img2.shape)\n",
    "  if img1.ndim != 4:\n",
    "    raise RuntimeError('Input images must have four dimensions, not %d',\n",
    "                       img1.ndim)\n",
    "\n",
    "  # Note: default weights don't sum to 1.0 but do match the paper / matlab code.\n",
    "  weights = np.array(weights if weights else\n",
    "                     [0.0448, 0.2856, 0.3001, 0.2363, 0.1333])\n",
    "  levels = weights.size\n",
    "  downsample_filter = np.ones((1, 2, 2, 1)) / 4.0\n",
    "  im1, im2 = [x.astype(np.float64) for x in [img1, img2]]\n",
    "  mssim = np.array([])\n",
    "  mcs = np.array([])\n",
    "  for _ in range(levels):\n",
    "    ssim, cs = _SSIMForMultiScale(\n",
    "        im1, im2, max_val=max_val, filter_size=filter_size,\n",
    "        filter_sigma=filter_sigma, k1=k1, k2=k2)\n",
    "    mssim = np.append(mssim, ssim)\n",
    "    mcs = np.append(mcs, cs)\n",
    "    filtered = [convolve(im, downsample_filter, mode='reflect')\n",
    "                for im in [im1, im2]]\n",
    "    im1, im2 = [x[:, ::2, ::2, :] for x in filtered]\n",
    "  return (np.prod(mcs[0:levels-1] ** weights[0:levels-1]) *\n",
    "          (mssim[levels-1] ** weights[levels-1]))\n",
    "\n",
    "n = 10\n",
    "for i in range(n):\n",
    "    original = x_test[i].reshape(1, 28, 28, 1)\n",
    "    contrast = decoded_imgs[i].reshape(1, 28, 28, 1)\n",
    "    d=MultiScaleSSIM(original, contrast, max_val=1.0)\n",
    "    print(f'{d:.4f}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 132 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "import os\n",
    "\n",
    "try:\n",
    "    os.makedirs('autoencoder-bc/model')\n",
    "except Exception:\n",
    "    pass\n",
    "\n",
    "save_model(autoencoder, 'autoencoder-bc/model/autoencoderv3-bc.h5')\n",
    "save_model(encoder, 'autoencoder-bc/model/encoderv3-bc.h5')\n",
    "save_model(decoder, 'autoencoder-bc/model/decoderv3-bc.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:No training configuration found in the save file, so the model was *not* compiled. Compile it manually.\n",
      "WARNING:tensorflow:No training configuration found in the save file, so the model was *not* compiled. Compile it manually.\n",
      "Model: \"model\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         [(None, 28, 28, 1)]       0         \n",
      "_________________________________________________________________\n",
      "conv2d (Conv2D)              (None, 28, 28, 16)        160       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 14, 14, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 14, 14, 8)         1160      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 7, 7, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 7, 7, 8)           584       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 4, 4, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 4, 4, 8)           584       \n",
      "_________________________________________________________________\n",
      "up_sampling2d (UpSampling2D) (None, 8, 8, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 8, 8, 8)           584       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_1 (UpSampling2 (None, 16, 16, 8)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_5 (Conv2D)            (None, 14, 14, 16)        1168      \n",
      "_________________________________________________________________\n",
      "up_sampling2d_2 (UpSampling2 (None, 28, 28, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_6 (Conv2D)            (None, 28, 28, 1)         145       \n",
      "=================================================================\n",
      "Total params: 4,385\n",
      "Trainable params: 4,385\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Model: \"model_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         [(None, 28, 28, 1)]       0         \n",
      "_________________________________________________________________\n",
      "conv2d (Conv2D)              (None, 28, 28, 16)        160       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 14, 14, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 14, 14, 8)         1160      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 7, 7, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 7, 7, 8)           584       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 4, 4, 8)           0         \n",
      "=================================================================\n",
      "Total params: 1,904\n",
      "Trainable params: 1,904\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Model: \"model_2\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         [(None, 4, 4, 8)]         0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 4, 4, 8)           584       \n",
      "_________________________________________________________________\n",
      "up_sampling2d (UpSampling2D) (None, 8, 8, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 8, 8, 8)           584       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_1 (UpSampling2 (None, 16, 16, 8)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_5 (Conv2D)            (None, 14, 14, 16)        1168      \n",
      "_________________________________________________________________\n",
      "up_sampling2d_2 (UpSampling2 (None, 28, 28, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_6 (Conv2D)            (None, 28, 28, 1)         145       \n",
      "=================================================================\n",
      "Total params: 2,481\n",
      "Trainable params: 2,481\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Wall time: 3.73 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "import os\n",
    "\n",
    "autoencoder = load_model('autoencoder-bc/model/autoencoderv3-bc.h5')\n",
    "encoder = load_model('autoencoder-bc/model/encoderv3-bc.h5')\n",
    "decoder = load_model('autoencoder-bc/model/decoderv3-bc.h5')\n",
    "\n",
    "autoencoder.summary()\n",
    "encoder.summary()\n",
    "decoder.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}