TF2_WebScraping_TransferLearning_Augmentation.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "TF2-WebScraping-TransferLearning-Augmentation.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8T2taV61R1_f",
        "colab_type": "text"
      },
      "source": [
        "# Image Recognition & Transfer Learning\n",
        "\n",
        "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/googlecolab/colabtools/blob/master/notebooks/colab-github-demo.ipynb)\n",
        "\n",
        "Notebook for an example end-to-end Deep Learning project:\n",
        "- Creating a new Dataset by scraping images from Google Images\n",
        "- Building an efficient datapipeline for images with tf.data (with data augmentation)\n",
        "- Training using Transfer Learning (by gradually unfreezing more and more layers) "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "3gGC8SVb2JXS",
        "outputId": "38714d9b-72a6-45be-c4f0-c6a998cb5128",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 70
        }
      },
      "source": [
        "try:\n",
        "    # %tensorflow_version only exists in Colab.\n",
        "    %tensorflow_version 2.x\n",
        "    %load_ext tensorboard\n",
        "except Exception:\n",
        "    pass\n",
        "\n",
        "import tensorflow as tf\n",
        "keras = tf.keras\n",
        "AUTOTUNE = tf.data.experimental.AUTOTUNE\n",
        "import os\n",
        "print(tf.__version__)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "The tensorboard extension is already loaded. To reload it, use:\n",
            "  %reload_ext tensorboard\n",
            "2.0.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wKeYqw3FV11E",
        "colab_type": "text"
      },
      "source": [
        "# Part 1: Create a Dataset (Web Scraping)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uhHBpwxBZUJK",
        "colab_type": "text"
      },
      "source": [
        "We will create a Dataset with images from 3 different categories \"smartphone\", \"laptop\", \"blender\" using images from Google Images. To download the images in an automated fashion, we will use the python tool <code>google_images_download</code>. \n",
        "\n",
        "To download more than 100 images with <code>google_images_download</code> you need to chrome installed on your system, and download chromedriver.\n",
        "As we can't do this on Colab, we will do it on our local machine:\n",
        "1. Install Chrome\n",
        "2. Download chromedriver.exe from https://sites.google.com/a/chromium.org/chromedriver/downloads\n",
        "3. <code>> pip install google_images_download</code>\n",
        "4. Download images using keywords \"smartphone\", \"laptop\", \"blender\"\n",
        "```\n",
        "googleimagesdownload -k \"smartphone\" -l 200 --chromedriver C:\\Windows\\chromedriver.exe\n",
        "googleimagesdownload -k \"laptop\" -l 200 --chromedriver C:\\Windows\\chromedriver.exe\n",
        "googleimagesdownload -k \"blender\" -l 200 --chromedriver C:\\Windows\\chromedriver.exe\n",
        "```\n",
        "5. Upload the images to a cloud storage (e.g Google Drive, Google Cloud Storage, AWS S3)\n",
        "6. Download the images to your Colab Instance \n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8JCnivpa8Kk1",
        "colab_type": "text"
      },
      "source": [
        "### Load Data\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2deXydZ88MYa",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import os\n",
        "URL = 'https://storage.googleapis.com/dl-workshop/datasets/laptop_smartphone_blender.zip'\n",
        "\n",
        "path_to_zip = tf.keras.utils.get_file('laptop_smartphone_blender.zip', origin=URL, extract=True)\n",
        "dataset_path = os.path.join(os.path.dirname(path_to_zip), 'laptop_smartphone_blender')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "06yFF0fv9EHw",
        "colab_type": "code",
        "outputId": "ff05973a-882e-41b0-b4c9-70af33e9baad",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "import pathlib\n",
        "import os\n",
        "import random\n",
        "\n",
        "data_root = pathlib.Path(dataset_path)\n",
        "\n",
        "class_names = sorted([e.name for e in (data_root).iterdir() if e.is_dir()])\n",
        "class_names_to_label = {name: label for label, name in enumerate(class_names)}\n",
        "label_to_class_names = {label: name for label, name in enumerate(class_names)}\n",
        "\n",
        "label_to_class_names"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{0: 'blender', 1: 'laptop', 2: 'smartphone'}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 54
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VSQLfFnisXte",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def remove_invalid_images(image_paths):\n",
        "    valid_image_paths = []\n",
        "\n",
        "    for path in image_paths:\n",
        "        try:\n",
        "            image = tf.io.read_file(str(path))\n",
        "            image = tf.image.decode_jpeg(image, channels=3)\n",
        "            valid_image_paths.append(path)\n",
        "        except:\n",
        "            # print(f'invalid image file found: {path}')\n",
        "            pass\n",
        "\n",
        "    print(f'\\n{len(image_paths) - len(valid_image_paths)} out of {len(image_paths)} files will be ignored')\n",
        "    \n",
        "    return valid_image_paths"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uSr6BCMG9T7u",
        "colab_type": "code",
        "outputId": "bba5c90c-2110-453d-eb2e-cdec6faa4bc3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "supported_filetypes = {'png', 'jpg', 'jpeg'}\n",
        "\n",
        "all_image_paths = []\n",
        "\n",
        "for f_type in supported_filetypes:\n",
        "    all_image_paths += list(data_root.glob(f'*/*.{f_type}'))\n",
        "\n",
        "random.seed(0)\n",
        "random.shuffle(all_image_paths)\n",
        "\n",
        "all_image_paths = remove_invalid_images(all_image_paths)\n",
        "\n",
        "all_image_labels = []\n",
        "\n",
        "for i, path in enumerate(all_image_paths):\n",
        "    all_image_labels.append(class_names_to_label[path.parent.name])\n",
        "    all_image_paths[i] = str(all_image_paths[i])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "43 out of 594 files will be ignored\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zqndykq6kVCs",
        "colab_type": "code",
        "outputId": "ee96d941-d59a-455e-a0ea-5962704e0736",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        }
      },
      "source": [
        "nr_train_images = int(0.7*len(all_image_paths))\n",
        "\n",
        "train_image_paths = all_image_paths[:nr_train_images]\n",
        "test_image_paths = all_image_paths[nr_train_images:]\n",
        "\n",
        "train_image_labels = all_image_labels[:nr_train_images]\n",
        "test_image_labels = all_image_labels[nr_train_images:]\n",
        "\n",
        "len(test_image_labels)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "166"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 57
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YeUvXzYm9uIb",
        "colab_type": "text"
      },
      "source": [
        "# Part 2: Create a Data Pipeline (with Data Augmentation)\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_AH6457O6QZT",
        "colab_type": "text"
      },
      "source": [
        "https://www.tensorflow.org/guide/data <br>\n",
        "For performance reasons, we encourage you to use TensorFlow operations for preprocessing your data whenever possible. However, it is sometimes useful to call external Python libraries when parsing your input data. You can use the tf.py_function() operation in a Dataset.map() transformation."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BotBEfYX6PsS",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import scipy.ndimage as ndimage\n",
        "import numpy as np\n",
        "\n",
        "\n",
        "def random_rotate_image(image):\n",
        "    image =  ndimage.rotate(image, np.random.uniform(-30,30), reshape=False)\n",
        "    return image\n",
        "\n",
        "def tf_random_rotate_image(image):\n",
        "    img_shape = image.shape\n",
        "    [image,] = tf.py_function(random_rotate_image, [image], [tf.float32])\n",
        "    image.set_shape(img_shape)\n",
        "    return image\n",
        "\n",
        "def random_crop(image, random=True):\n",
        "    if random:\n",
        "        shape = tf.shape(image)\n",
        "        min_dim = tf.reduce_min([shape[0], shape[1]]) * 90 // 100\n",
        "        image = tf.image.random_crop(image, [min_dim, min_dim, 3])\n",
        "    else:\n",
        "        image = tf.image.central_crop(image, 0.5)\n",
        "\n",
        "    return image\n",
        "\n",
        "def random_color(image):\n",
        "    image = tf.image.random_hue(image, 0.08)\n",
        "    image = tf.image.random_saturation(image, 0.6, 1.6)\n",
        "    image = tf.image.random_brightness(image, 0.05)\n",
        "    image = tf.image.random_contrast(image, 0.7, 1.3)\n",
        "    \n",
        "    return image"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "SwZgecHCr5R-",
        "colab": {}
      },
      "source": [
        "from functools import partial\n",
        "\n",
        "IMG_SHAPE = (64, 64)\n",
        "\n",
        "def random_augs(image):\n",
        "    choice1 = tf.random.uniform(shape=[], minval=0., maxval=1., dtype=tf.float32)\n",
        "    image = tf.cond(choice1 > 0.7, lambda: image, lambda: random_crop(image))\n",
        "\n",
        "    choice2 = tf.random.uniform(shape=[], minval=0., maxval=1., dtype=tf.float32)\n",
        "    image = tf.cond(choice2 > 0.7, lambda: image, lambda: random_color(image))\n",
        "\n",
        "    choice3 = tf.random.uniform(shape=[], minval=0., maxval=1., dtype=tf.float32)\n",
        "    image = tf.cond(choice3 > 0.7, lambda: image, lambda: tf_random_rotate_image(image))\n",
        "\n",
        "    return image\n",
        "\n",
        "\n",
        "def augmentations(image):\n",
        "    image = tf.image.random_flip_left_right(image)\n",
        "    image = tf.image.random_flip_up_down(image)\n",
        "\n",
        "    return random_augs(image)\n",
        "\n",
        "\n",
        "def load_and_preprocess_image(image_path, label, img_shape, augmentation):\n",
        "    image = tf.io.read_file(image_path)\n",
        "    image = tf.image.decode_jpeg(image, channels=3)\n",
        "    image = keras.applications.resnet_v2.preprocess_input(tf.cast(image, tf.float32))\n",
        "\n",
        "    if augmentation:\n",
        "        image = augmentations(image)\n",
        "        image = tf.clip_by_value(image, -1, 1)\n",
        "\n",
        "    image = tf.image.resize(image, img_shape)\n",
        "    label = tf.cast(label, tf.float32)    \n",
        "\n",
        "    return image, label\n",
        "\n",
        "\n",
        "def create_dataset(file_paths, labels, epochs=1, batch_size=64, buffer_size=10000, train=True, augmentation=False):\n",
        "    dataset = tf.data.Dataset.from_tensor_slices((file_paths, labels))\n",
        "    dataset = dataset.map(partial(load_and_preprocess_image, img_shape=IMG_SHAPE, augmentation=augmentation), num_parallel_calls=AUTOTUNE)\n",
        "    if train:\n",
        "        dataset = dataset.cache()\n",
        "        dataset = dataset.shuffle(buffer_size=buffer_size)\n",
        "    dataset = dataset.repeat(epochs)\n",
        "    dataset = dataset.batch(batch_size)\n",
        "    if train:\n",
        "        dataset = dataset.prefetch(AUTOTUNE)\n",
        "    \n",
        "    return dataset"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5dTouMsi8o3e",
        "colab_type": "code",
        "outputId": "27f573ec-cfa0-44d1-d744-5803fe0dfac8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 195
        }
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "ds = create_dataset(train_image_paths, train_image_labels, batch_size=5, train=False, augmentation=False)\n",
        "\n",
        "for batch, labels in ds.take(1):\n",
        "    fig, ax = plt.subplots(1, len(batch), figsize=(15,20))\n",
        "    for i, (img, label) in enumerate(zip(batch, labels)):\n",
        "        fig.axes[i].imshow((img+1)/2)\n",
        "        fig.axes[i].set_title(label_to_class_names[labels[i].numpy()])\n",
        "        fig.axes[i].set_xticks([])\n",
        "        fig.axes[i].set_yticks([])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA1YAAACyCAYAAABBYV5iAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nE0DfgYEAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 1080x1440 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YpkTtO8SXSgK",
        "colab_type": "code",
        "outputId": "decb69e9-fa62-4d88-a335-26612a05303c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 195
        }
      },
      "source": [
        "ds = create_dataset(train_image_paths, train_image_labels, batch_size=5, train=False, augmentation=True)\n",
        "\n",
        "for batch, labels in ds.take(1):\n",
        "    fig, ax = plt.subplots(1, len(batch), figsize=(15,20))\n",
        "    for i, (img, label) in enumerate(zip(batch, labels)):\n",
        "        fig.axes[i].imshow((img+1)/2)\n",
        "        fig.axes[i].set_title(label_to_class_names[labels[i].numpy()])\n",
        "        fig.axes[i].set_xticks([])\n",
        "        fig.axes[i].set_yticks([])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA1YAAACyCAYAAABBYV5iAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 1080x1440 with 5 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dA_jZ6lkokBQ",
        "colab_type": "text"
      },
      "source": [
        "# Part 3: Modelling"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jIsF7ocUo0A2",
        "colab_type": "text"
      },
      "source": [
        "## Baseline Model (LeNet-5)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OscimM-Hojbq",
        "colab_type": "code",
        "outputId": "882db74c-b731-4820-f1ee-8b8875aad97e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 444
        }
      },
      "source": [
        "from tensorflow.keras import layers, models\n",
        "\n",
        "def create_baseline_model(nr_classes, input_shape):\n",
        "    return models.Sequential([\n",
        "        layers.Conv2D(32, (3, 3), activation='relu', input_shape=input_shape),\n",
        "        layers.MaxPooling2D((2, 2)),\n",
        "        layers.Conv2D(64, (3, 3), activation='relu'),\n",
        "        layers.MaxPooling2D((2, 2)),\n",
        "        layers.Conv2D(64, (3, 3), activation='relu'),\n",
        "        layers.Flatten(),\n",
        "        layers.Dense(64, activation='relu'),\n",
        "        layers.Dense(nr_classes, activation='softmax')\n",
        "    ])\n",
        "\n",
        "\n",
        "model = create_baseline_model(nr_classes=len(class_names), input_shape=IMG_SHAPE + (3,))\n",
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential_3\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d_9 (Conv2D)            (None, 62, 62, 32)        896       \n",
            "_________________________________________________________________\n",
            "max_pooling2d_57 (MaxPooling (None, 31, 31, 32)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_10 (Conv2D)           (None, 29, 29, 64)        18496     \n",
            "_________________________________________________________________\n",
            "max_pooling2d_58 (MaxPooling (None, 14, 14, 64)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_11 (Conv2D)           (None, 12, 12, 64)        36928     \n",
            "_________________________________________________________________\n",
            "flatten_3 (Flatten)          (None, 9216)              0         \n",
            "_________________________________________________________________\n",
            "dense_40 (Dense)             (None, 64)                589888    \n",
            "_________________________________________________________________\n",
            "dense_41 (Dense)             (None, 3)                 195       \n",
            "=================================================================\n",
            "Total params: 646,403\n",
            "Trainable params: 646,403\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VrCzLzdBplXJ",
        "colab_type": "code",
        "outputId": "2e8b1f6e-9945-4d3d-d6f7-5b5c55929b5e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "model = create_baseline_model(nr_classes=len(class_names), input_shape=IMG_SHAPE + (3,))\n",
        "model.summary()\n",
        "\n",
        "nr_epochs = 30\n",
        "batch_size = 32\n",
        "\n",
        "train_ds = create_dataset(train_image_paths, train_image_labels, epochs=nr_epochs, batch_size=batch_size, train=True, augmentation=False)\n",
        "test_ds = create_dataset(test_image_paths, test_image_labels, epochs=nr_epochs, batch_size=batch_size, train=False, augmentation=False)\n",
        "\n",
        "steps_per_epoch = len(train_image_paths) // batch_size\n",
        "validation_steps = len(test_image_paths) // batch_size\n",
        "\n",
        "model.compile(optimizer=tf.keras.optimizers.Adam(lr=3e-4),\n",
        "              loss='sparse_categorical_crossentropy',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "history = model.fit(train_ds,\n",
        "                    epochs = nr_epochs,\n",
        "                    steps_per_epoch = steps_per_epoch,\n",
        "                    validation_data = test_ds, \n",
        "                    validation_steps = validation_steps,\n",
        "                    callbacks = [])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train for 12 steps, validate for 5 steps\n",
            "Epoch 1/30\n",
            "12/12 [==============================] - 10s 850ms/step - loss: 0.9299 - accuracy: 0.5130 - val_loss: 0.7423 - val_accuracy: 0.7125\n",
            "Epoch 2/30\n",
            "12/12 [==============================] - 3s 284ms/step - loss: 0.7374 - accuracy: 0.7214 - val_loss: 0.6500 - val_accuracy: 0.7937\n",
            "Epoch 3/30\n",
            "12/12 [==============================] - 3s 283ms/step - loss: 0.5810 - accuracy: 0.8203 - val_loss: 0.5769 - val_accuracy: 0.7688\n",
            "Epoch 4/30\n",
            "12/12 [==============================] - 3s 287ms/step - loss: 0.4759 - accuracy: 0.7995 - val_loss: 0.6246 - val_accuracy: 0.7437\n",
            "Epoch 5/30\n",
            "12/12 [==============================] - 3s 280ms/step - loss: 0.3936 - accuracy: 0.8438 - val_loss: 0.4873 - val_accuracy: 0.7937\n",
            "Epoch 6/30\n",
            "12/12 [==============================] - 4s 297ms/step - loss: 0.2917 - accuracy: 0.9036 - val_loss: 0.4378 - val_accuracy: 0.8000\n",
            "Epoch 7/30\n",
            "12/12 [==============================] - 4s 301ms/step - loss: 0.2372 - accuracy: 0.9219 - val_loss: 0.4952 - val_accuracy: 0.8188\n",
            "Epoch 8/30\n",
            "12/12 [==============================] - 4s 298ms/step - loss: 0.1800 - accuracy: 0.9505 - val_loss: 0.4238 - val_accuracy: 0.8500\n",
            "Epoch 9/30\n",
            "12/12 [==============================] - 4s 293ms/step - loss: 0.1360 - accuracy: 0.9661 - val_loss: 0.4499 - val_accuracy: 0.8375\n",
            "Epoch 10/30\n",
            "12/12 [==============================] - 4s 302ms/step - loss: 0.1065 - accuracy: 0.9792 - val_loss: 0.4561 - val_accuracy: 0.8125\n",
            "Epoch 11/30\n",
            "12/12 [==============================] - 4s 295ms/step - loss: 0.0799 - accuracy: 0.9922 - val_loss: 0.4780 - val_accuracy: 0.8125\n",
            "Epoch 12/30\n",
            "12/12 [==============================] - 3s 291ms/step - loss: 0.0654 - accuracy: 0.9896 - val_loss: 0.4838 - val_accuracy: 0.8188\n",
            "Epoch 13/30\n",
            "12/12 [==============================] - 3s 289ms/step - loss: 0.0460 - accuracy: 0.9974 - val_loss: 0.5222 - val_accuracy: 0.8125\n",
            "Epoch 14/30\n",
            "12/12 [==============================] - 3s 284ms/step - loss: 0.0377 - accuracy: 0.9948 - val_loss: 0.5492 - val_accuracy: 0.8250\n",
            "Epoch 15/30\n",
            "12/12 [==============================] - 3s 291ms/step - loss: 0.0281 - accuracy: 0.9974 - val_loss: 0.5591 - val_accuracy: 0.8188\n",
            "Epoch 16/30\n",
            "12/12 [==============================] - 4s 292ms/step - loss: 0.0270 - accuracy: 0.9948 - val_loss: 0.6321 - val_accuracy: 0.8250\n",
            "Epoch 17/30\n",
            "12/12 [==============================] - 3s 288ms/step - loss: 0.0235 - accuracy: 1.0000 - val_loss: 0.6016 - val_accuracy: 0.8313\n",
            "Epoch 18/30\n",
            "12/12 [==============================] - 3s 289ms/step - loss: 0.0199 - accuracy: 0.9948 - val_loss: 0.6478 - val_accuracy: 0.8313\n",
            "Epoch 19/30\n",
            "12/12 [==============================] - 3s 288ms/step - loss: 0.0123 - accuracy: 1.0000 - val_loss: 0.6261 - val_accuracy: 0.8062\n",
            "Epoch 20/30\n",
            "12/12 [==============================] - 3s 283ms/step - loss: 0.0122 - accuracy: 1.0000 - val_loss: 0.6684 - val_accuracy: 0.8375\n",
            "Epoch 21/30\n",
            "12/12 [==============================] - 3s 284ms/step - loss: 0.0105 - accuracy: 1.0000 - val_loss: 0.6790 - val_accuracy: 0.8250\n",
            "Epoch 22/30\n",
            "12/12 [==============================] - 3s 286ms/step - loss: 0.0084 - accuracy: 1.0000 - val_loss: 0.7076 - val_accuracy: 0.8250\n",
            "Epoch 23/30\n",
            "12/12 [==============================] - 3s 278ms/step - loss: 0.0057 - accuracy: 1.0000 - val_loss: 0.7213 - val_accuracy: 0.8125\n",
            "Epoch 24/30\n",
            "12/12 [==============================] - 3s 277ms/step - loss: 0.0057 - accuracy: 1.0000 - val_loss: 0.7216 - val_accuracy: 0.8375\n",
            "Epoch 25/30\n",
            "12/12 [==============================] - 3s 278ms/step - loss: 0.0049 - accuracy: 1.0000 - val_loss: 0.7236 - val_accuracy: 0.8062\n",
            "Epoch 26/30\n",
            "12/12 [==============================] - 3s 274ms/step - loss: 0.0044 - accuracy: 1.0000 - val_loss: 0.7544 - val_accuracy: 0.8188\n",
            "Epoch 27/30\n",
            "12/12 [==============================] - 3s 284ms/step - loss: 0.0051 - accuracy: 1.0000 - val_loss: 0.7980 - val_accuracy: 0.8313\n",
            "Epoch 28/30\n",
            "12/12 [==============================] - 3s 278ms/step - loss: 0.0042 - accuracy: 1.0000 - val_loss: 0.7638 - val_accuracy: 0.8125\n",
            "Epoch 29/30\n",
            "12/12 [==============================] - 3s 276ms/step - loss: 0.0030 - accuracy: 1.0000 - val_loss: 0.7741 - val_accuracy: 0.8188\n",
            "Epoch 30/30\n",
            "12/12 [==============================] - 3s 273ms/step - loss: 0.0033 - accuracy: 1.0000 - val_loss: 0.8220 - val_accuracy: 0.8375\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2F-8THgSwAa8",
        "colab_type": "text"
      },
      "source": [
        "## Using Transfer Learning (ResNet-50)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ir0GNO_G9iRq",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def create_resnet50_model(nr_classes, learning_rate=3e-4, dropout=0):\n",
        "    base_model = keras.applications.resnet_v2.ResNet50V2(weights=\"imagenet\", include_top=False)\n",
        "    \n",
        "    for layer in base_model.layers:\n",
        "        if layer.__class__.__name__ != \"BatchNormalization\":\n",
        "            layer.trainable = False\n",
        "    \n",
        "    avg = keras.layers.GlobalAveragePooling2D()(base_model.output)\n",
        "    mx = keras.layers.GlobalMaxPooling2D()(base_model.output)\n",
        "    out = tf.keras.layers.Concatenate()([avg, mx])\n",
        "    out = keras.layers.BatchNormalization()(out)\n",
        "    out = keras.layers.Dropout(0)(out)\n",
        "    out = keras.layers.Dense(512, activation=\"relu\")(out)\n",
        "    out = keras.layers.BatchNormalization()(out)\n",
        "    out = keras.layers.Dropout(0)(out)\n",
        "    out = keras.layers.Dense(nr_classes, activation=\"softmax\")(out)\n",
        "    \n",
        "    model = keras.models.Model(inputs=base_model.input, outputs=out)\n",
        "    \n",
        "    optimizer = tf.keras.optimizers.Adam(lr=learning_rate)\n",
        "    model.compile(loss=\"sparse_categorical_crossentropy\", optimizer=optimizer, metrics=[\"accuracy\"])\n",
        "    \n",
        "    return model\n",
        "\n",
        "\n",
        "def unfreeze(model, nr_unfreezed_layers=None, lr=0.0001):\n",
        "    if nr_unfreezed_layers == None:\n",
        "        nr_freezed_layers = 0\n",
        "    else:\n",
        "        nr_freezed_layers = len(model.layers) - nr_unfreezed_layers\n",
        "\n",
        "    model.trainable = True\n",
        "    for i in range(len(model.layers)):\n",
        "        if i < nr_freezed_layers:\n",
        "            model.layers[i].trainable = False\n",
        "        else:\n",
        "            model.layers[i].trainable = True\n",
        "            if nr_unfreezed_layers:\n",
        "                print(f'Unfreezed Layer {i}: {model.layers[i]}')\n",
        "\n",
        "    optimizer = tf.keras.optimizers.Adam(lr=lr)\n",
        "    model.compile(loss=\"sparse_categorical_crossentropy\", optimizer=optimizer, metrics=[\"accuracy\"])\n",
        "    \n",
        "    return model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OLidnM0nQNAI",
        "colab_type": "text"
      },
      "source": [
        "## Training without augmentation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YHzX3kmSEjJB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "BATCH_SIZE = 32\n",
        "steps_per_epoch = len(train_image_paths) // BATCH_SIZE\n",
        "validation_steps = len(test_image_paths) // BATCH_SIZE"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SS7Rpv6u7njz",
        "colab_type": "text"
      },
      "source": [
        "### Step 1: Training Freezed Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "laxOWWe6NQo2",
        "colab_type": "code",
        "outputId": "37ecf80f-9810-4267-94ea-d4c70355d264",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 568
        }
      },
      "source": [
        "model = create_resnet50_model(len(class_names), learning_rate=2e-4)\n",
        "\n",
        "nr_epochs = 15\n",
        "\n",
        "train_ds = create_dataset(train_image_paths, train_image_labels, epochs=nr_epochs, batch_size=BATCH_SIZE, train=True)\n",
        "test_ds = create_dataset(test_image_paths, test_image_labels, epochs=nr_epochs, batch_size=BATCH_SIZE, train=False)\n",
        "\n",
        "history = model.fit(train_ds,\n",
        "                    epochs = nr_epochs,\n",
        "                    steps_per_epoch = steps_per_epoch,\n",
        "                    validation_data = test_ds, \n",
        "                    validation_steps = validation_steps,\n",
        "                    callbacks = [])\n",
        "\n",
        "model.save_weights('./step-1/checkpoint')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train for 12 steps, validate for 5 steps\n",
            "Epoch 1/15\n",
            "12/12 [==============================] - 22s 2s/step - loss: 1.2169 - accuracy: 0.5182 - val_loss: 1.2112 - val_accuracy: 0.5688\n",
            "Epoch 2/15\n",
            "12/12 [==============================] - 5s 399ms/step - loss: 0.5607 - accuracy: 0.7943 - val_loss: 0.9399 - val_accuracy: 0.6750\n",
            "Epoch 3/15\n",
            "12/12 [==============================] - 5s 386ms/step - loss: 0.3766 - accuracy: 0.8464 - val_loss: 0.9504 - val_accuracy: 0.6625\n",
            "Epoch 4/15\n",
            "12/12 [==============================] - 5s 390ms/step - loss: 0.3246 - accuracy: 0.8802 - val_loss: 0.8502 - val_accuracy: 0.7250\n",
            "Epoch 5/15\n",
            "12/12 [==============================] - 5s 386ms/step - loss: 0.2438 - accuracy: 0.9167 - val_loss: 0.7973 - val_accuracy: 0.7063\n",
            "Epoch 6/15\n",
            "12/12 [==============================] - 5s 388ms/step - loss: 0.1848 - accuracy: 0.9323 - val_loss: 0.7524 - val_accuracy: 0.7188\n",
            "Epoch 7/15\n",
            "12/12 [==============================] - 5s 386ms/step - loss: 0.1925 - accuracy: 0.9245 - val_loss: 0.6857 - val_accuracy: 0.7500\n",
            "Epoch 8/15\n",
            "12/12 [==============================] - 5s 390ms/step - loss: 0.1683 - accuracy: 0.9297 - val_loss: 0.6133 - val_accuracy: 0.7500\n",
            "Epoch 9/15\n",
            "12/12 [==============================] - 5s 394ms/step - loss: 0.1424 - accuracy: 0.9531 - val_loss: 0.5678 - val_accuracy: 0.7812\n",
            "Epoch 10/15\n",
            "12/12 [==============================] - 5s 398ms/step - loss: 0.1463 - accuracy: 0.9479 - val_loss: 0.5369 - val_accuracy: 0.7688\n",
            "Epoch 11/15\n",
            "12/12 [==============================] - 5s 389ms/step - loss: 0.1366 - accuracy: 0.9609 - val_loss: 0.5025 - val_accuracy: 0.7812\n",
            "Epoch 12/15\n",
            "12/12 [==============================] - 5s 419ms/step - loss: 0.1326 - accuracy: 0.9505 - val_loss: 0.4602 - val_accuracy: 0.8188\n",
            "Epoch 13/15\n",
            "12/12 [==============================] - 5s 410ms/step - loss: 0.1479 - accuracy: 0.9505 - val_loss: 0.4769 - val_accuracy: 0.8125\n",
            "Epoch 14/15\n",
            "12/12 [==============================] - 5s 396ms/step - loss: 0.1405 - accuracy: 0.9453 - val_loss: 0.5387 - val_accuracy: 0.7812\n",
            "Epoch 15/15\n",
            "12/12 [==============================] - 5s 405ms/step - loss: 0.1206 - accuracy: 0.9531 - val_loss: 0.5371 - val_accuracy: 0.7937\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XXNeCd9t78WE",
        "colab_type": "text"
      },
      "source": [
        "### Step 2: Unfreeze last 30 Layers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tv1wcCXpypC0",
        "colab_type": "code",
        "outputId": "b99804f6-35d1-4cd6-a836-5d4ae64cc8fc",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "# Create a new model instance\n",
        "model = create_resnet50_model(len(class_names))\n",
        "\n",
        "# Restore the weights\n",
        "model.load_weights('./step-1/checkpoint')\n",
        "\n",
        "model = unfreeze(model, nr_unfreezed_layers=30, lr=0.0001)\n",
        "\n",
        "nr_epochs = 25\n",
        "\n",
        "train_ds = create_dataset(train_image_paths, train_image_labels, epochs=nr_epochs, batch_size=BATCH_SIZE, train=True)\n",
        "test_ds = create_dataset(test_image_paths, test_image_labels, epochs=nr_epochs, batch_size=BATCH_SIZE, train=False)\n",
        "\n",
        "history = model.fit(train_ds,\n",
        "                    epochs = nr_epochs,\n",
        "                    steps_per_epoch = steps_per_epoch,\n",
        "                    validation_data = test_ds, \n",
        "                    validation_steps = validation_steps,\n",
        "                    callbacks = [])\n",
        "\n",
        "model.save_weights('./step-2/checkpoint')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Unfreezed Layer 169: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c01bd36a0>\n",
            "Unfreezed Layer 170: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c01bd3f60>\n",
            "Unfreezed Layer 171: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5c01bad400>\n",
            "Unfreezed Layer 172: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c01bdd080>\n",
            "Unfreezed Layer 173: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c01b47860>\n",
            "Unfreezed Layer 174: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c01b65160>\n",
            "Unfreezed Layer 175: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c01b2ec18>\n",
            "Unfreezed Layer 176: <tensorflow.python.keras.layers.merge.Add object at 0x7f5c01add400>\n",
            "Unfreezed Layer 177: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c01add470>\n",
            "Unfreezed Layer 178: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c01ae58d0>\n",
            "Unfreezed Layer 179: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c01aabf98>\n",
            "Unfreezed Layer 180: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c01a60860>\n",
            "Unfreezed Layer 181: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c01a6e160>\n",
            "Unfreezed Layer 182: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5c01939c50>\n",
            "Unfreezed Layer 183: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c01a6e240>\n",
            "Unfreezed Layer 184: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c018e6a20>\n",
            "Unfreezed Layer 185: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c018f2320>\n",
            "Unfreezed Layer 186: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c018b8dd8>\n",
            "Unfreezed Layer 187: <tensorflow.python.keras.layers.merge.Add object at 0x7f5c0186b5c0>\n",
            "Unfreezed Layer 188: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c0186b630>\n",
            "Unfreezed Layer 189: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c01878a90>\n",
            "Unfreezed Layer 190: <tensorflow.python.keras.layers.pooling.GlobalAveragePooling2D object at 0x7f5c0179b668>\n",
            "Unfreezed Layer 191: <tensorflow.python.keras.layers.pooling.GlobalMaxPooling2D object at 0x7f5c017b6828>\n",
            "Unfreezed Layer 192: <tensorflow.python.keras.layers.merge.Concatenate object at 0x7f5c017b6a20>\n",
            "Unfreezed Layer 193: <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7f5c017b6780>\n",
            "Unfreezed Layer 194: <tensorflow.python.keras.layers.core.Dropout object at 0x7f5c01746a20>\n",
            "Unfreezed Layer 195: <tensorflow.python.keras.layers.core.Dense object at 0x7f5c017204e0>\n",
            "Unfreezed Layer 196: <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7f5c0173d860>\n",
            "Unfreezed Layer 197: <tensorflow.python.keras.layers.core.Dropout object at 0x7f5c016d70f0>\n",
            "Unfreezed Layer 198: <tensorflow.python.keras.layers.core.Dense object at 0x7f5c016cef60>\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate\n",
            "WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/alpha/guide/checkpoints#loading_mechanics for details.\n",
            "Train for 12 steps, validate for 5 steps\n",
            "Epoch 1/25\n",
            "12/12 [==============================] - 17s 1s/step - loss: 0.1198 - accuracy: 0.9557 - val_loss: 0.5573 - val_accuracy: 0.7688\n",
            "Epoch 2/25\n",
            "12/12 [==============================] - 4s 343ms/step - loss: 0.0959 - accuracy: 0.9714 - val_loss: 0.5911 - val_accuracy: 0.7812\n",
            "Epoch 3/25\n",
            "12/12 [==============================] - 4s 344ms/step - loss: 0.0489 - accuracy: 0.9896 - val_loss: 0.5694 - val_accuracy: 0.8125\n",
            "Epoch 4/25\n",
            "12/12 [==============================] - 4s 341ms/step - loss: 0.0460 - accuracy: 0.9844 - val_loss: 0.5444 - val_accuracy: 0.8188\n",
            "Epoch 5/25\n",
            "12/12 [==============================] - 4s 334ms/step - loss: 0.0333 - accuracy: 0.9870 - val_loss: 0.5459 - val_accuracy: 0.8000\n",
            "Epoch 6/25\n",
            "12/12 [==============================] - 4s 332ms/step - loss: 0.0436 - accuracy: 0.9896 - val_loss: 0.5743 - val_accuracy: 0.7812\n",
            "Epoch 7/25\n",
            "12/12 [==============================] - 4s 336ms/step - loss: 0.0227 - accuracy: 1.0000 - val_loss: 0.5887 - val_accuracy: 0.7875\n",
            "Epoch 8/25\n",
            "12/12 [==============================] - 4s 341ms/step - loss: 0.0502 - accuracy: 0.9792 - val_loss: 0.5973 - val_accuracy: 0.7937\n",
            "Epoch 9/25\n",
            "12/12 [==============================] - 4s 354ms/step - loss: 0.0213 - accuracy: 0.9948 - val_loss: 0.5782 - val_accuracy: 0.7750\n",
            "Epoch 10/25\n",
            "12/12 [==============================] - 4s 337ms/step - loss: 0.0249 - accuracy: 0.9974 - val_loss: 0.5470 - val_accuracy: 0.8000\n",
            "Epoch 11/25\n",
            "12/12 [==============================] - 4s 339ms/step - loss: 0.0274 - accuracy: 0.9896 - val_loss: 0.5245 - val_accuracy: 0.8250\n",
            "Epoch 12/25\n",
            "12/12 [==============================] - 4s 335ms/step - loss: 0.0321 - accuracy: 0.9896 - val_loss: 0.5093 - val_accuracy: 0.8375\n",
            "Epoch 13/25\n",
            "12/12 [==============================] - 4s 340ms/step - loss: 0.0091 - accuracy: 0.9974 - val_loss: 0.5261 - val_accuracy: 0.8313\n",
            "Epoch 14/25\n",
            "12/12 [==============================] - 4s 338ms/step - loss: 0.0193 - accuracy: 0.9948 - val_loss: 0.5269 - val_accuracy: 0.8313\n",
            "Epoch 15/25\n",
            "12/12 [==============================] - 4s 340ms/step - loss: 0.0110 - accuracy: 1.0000 - val_loss: 0.5321 - val_accuracy: 0.8250\n",
            "Epoch 16/25\n",
            "12/12 [==============================] - 4s 338ms/step - loss: 0.0190 - accuracy: 0.9974 - val_loss: 0.5014 - val_accuracy: 0.8188\n",
            "Epoch 17/25\n",
            "12/12 [==============================] - 4s 331ms/step - loss: 0.0295 - accuracy: 0.9922 - val_loss: 0.5066 - val_accuracy: 0.8313\n",
            "Epoch 18/25\n",
            "12/12 [==============================] - 4s 332ms/step - loss: 0.0152 - accuracy: 0.9948 - val_loss: 0.5230 - val_accuracy: 0.8375\n",
            "Epoch 19/25\n",
            "12/12 [==============================] - 4s 335ms/step - loss: 0.0126 - accuracy: 0.9974 - val_loss: 0.5417 - val_accuracy: 0.8062\n",
            "Epoch 20/25\n",
            "12/12 [==============================] - 4s 351ms/step - loss: 0.0264 - accuracy: 0.9922 - val_loss: 0.5911 - val_accuracy: 0.8125\n",
            "Epoch 21/25\n",
            "12/12 [==============================] - 4s 346ms/step - loss: 0.0421 - accuracy: 0.9896 - val_loss: 0.5822 - val_accuracy: 0.8250\n",
            "Epoch 22/25\n",
            "12/12 [==============================] - 4s 345ms/step - loss: 0.0211 - accuracy: 0.9922 - val_loss: 0.5851 - val_accuracy: 0.8250\n",
            "Epoch 23/25\n",
            "12/12 [==============================] - 4s 347ms/step - loss: 0.0080 - accuracy: 1.0000 - val_loss: 0.5581 - val_accuracy: 0.8500\n",
            "Epoch 24/25\n",
            "12/12 [==============================] - 4s 348ms/step - loss: 0.0195 - accuracy: 0.9896 - val_loss: 0.5130 - val_accuracy: 0.8438\n",
            "Epoch 25/25\n",
            "12/12 [==============================] - 4s 349ms/step - loss: 0.0317 - accuracy: 0.9948 - val_loss: 0.5022 - val_accuracy: 0.8313\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1EIyicel8DWC",
        "colab_type": "text"
      },
      "source": [
        "### Step 3: Unfreeze last 50 Layers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XhtmpdixzPQr",
        "colab_type": "code",
        "outputId": "95b102dd-49dc-49c6-912e-8c4be229cbe2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "# Create a new model instance\n",
        "model = create_resnet50_model(len(class_names))\n",
        "\n",
        "# Restore the weights\n",
        "model.load_weights('./step-2/checkpoint')\n",
        "\n",
        "model = unfreeze(model, nr_unfreezed_layers=50, lr=0.00004)\n",
        "\n",
        "nr_epochs = 30\n",
        "\n",
        "train_ds = create_dataset(train_image_paths, train_image_labels, epochs=nr_epochs, batch_size=BATCH_SIZE, train=True)\n",
        "test_ds = create_dataset(test_image_paths, test_image_labels, epochs=nr_epochs, batch_size=BATCH_SIZE, train=False)\n",
        "\n",
        "history = model.fit(train_ds,\n",
        "                    epochs = nr_epochs,\n",
        "                    steps_per_epoch = steps_per_epoch,\n",
        "                    validation_data = test_ds, \n",
        "                    validation_steps = validation_steps,\n",
        "                    callbacks = [])\n",
        "\n",
        "model.save_weights('./step-3/checkpoint')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Unfreezed Layer 149: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf51a5f28>\n",
            "Unfreezed Layer 150: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf5138978>\n",
            "Unfreezed Layer 151: <tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x7f5bf52aa940>\n",
            "Unfreezed Layer 152: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf5172f98>\n",
            "Unfreezed Layer 153: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bf5128c18>\n",
            "Unfreezed Layer 154: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf5128d30>\n",
            "Unfreezed Layer 155: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf50b9f98>\n",
            "Unfreezed Layer 156: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf50aef28>\n",
            "Unfreezed Layer 157: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf5040fd0>\n",
            "Unfreezed Layer 158: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf5052b70>\n",
            "Unfreezed Layer 159: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bf5052b38>\n",
            "Unfreezed Layer 160: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf5026f98>\n",
            "Unfreezed Layer 161: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf4fd6470>\n",
            "Unfreezed Layer 162: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf50ae860>\n",
            "Unfreezed Layer 163: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf50cd0b8>\n",
            "Unfreezed Layer 164: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf4fe1160>\n",
            "Unfreezed Layer 165: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bf4f41fd0>\n",
            "Unfreezed Layer 166: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf4f41ef0>\n",
            "Unfreezed Layer 167: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf4f6d4a8>\n",
            "Unfreezed Layer 168: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf4f345f8>\n",
            "Unfreezed Layer 169: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf4ede470>\n",
            "Unfreezed Layer 170: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf4ecac18>\n",
            "Unfreezed Layer 171: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bf4eeb160>\n",
            "Unfreezed Layer 172: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf4edee10>\n",
            "Unfreezed Layer 173: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf4e62630>\n",
            "Unfreezed Layer 174: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf4e62e80>\n",
            "Unfreezed Layer 175: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf4e62f98>\n",
            "Unfreezed Layer 176: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bf4dd1fd0>\n",
            "Unfreezed Layer 177: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf4de5240>\n",
            "Unfreezed Layer 178: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf4df46a0>\n",
            "Unfreezed Layer 179: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf4d3be48>\n",
            "Unfreezed Layer 180: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf4d6a630>\n",
            "Unfreezed Layer 181: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf4d6ae80>\n",
            "Unfreezed Layer 182: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bf4d6af98>\n",
            "Unfreezed Layer 183: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf4d740f0>\n",
            "Unfreezed Layer 184: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf4cef7f0>\n",
            "Unfreezed Layer 185: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf4cf6208>\n",
            "Unfreezed Layer 186: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf4c47ba8>\n",
            "Unfreezed Layer 187: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bf4c74390>\n",
            "Unfreezed Layer 188: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf4c74400>\n",
            "Unfreezed Layer 189: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf4bfe860>\n",
            "Unfreezed Layer 190: <tensorflow.python.keras.layers.pooling.GlobalAveragePooling2D object at 0x7f5bf4bfe7b8>\n",
            "Unfreezed Layer 191: <tensorflow.python.keras.layers.pooling.GlobalMaxPooling2D object at 0x7f5bf4b3f828>\n",
            "Unfreezed Layer 192: <tensorflow.python.keras.layers.merge.Concatenate object at 0x7f5bf4b3f978>\n",
            "Unfreezed Layer 193: <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7f5bf4b3f0f0>\n",
            "Unfreezed Layer 194: <tensorflow.python.keras.layers.core.Dropout object at 0x7f5bf4b4c940>\n",
            "Unfreezed Layer 195: <tensorflow.python.keras.layers.core.Dense object at 0x7f5bf4b4c898>\n",
            "Unfreezed Layer 196: <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7f5bf4ab9e10>\n",
            "Unfreezed Layer 197: <tensorflow.python.keras.layers.core.Dropout object at 0x7f5bf4ade048>\n",
            "Unfreezed Layer 198: <tensorflow.python.keras.layers.core.Dense object at 0x7f5bf4acef60>\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate\n",
            "WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/alpha/guide/checkpoints#loading_mechanics for details.\n",
            "Train for 12 steps, validate for 5 steps\n",
            "Epoch 1/30\n",
            "12/12 [==============================] - 18s 1s/step - loss: 0.0101 - accuracy: 0.9948 - val_loss: 0.5408 - val_accuracy: 0.8313\n",
            "Epoch 2/30\n",
            "12/12 [==============================] - 4s 375ms/step - loss: 0.0071 - accuracy: 0.9974 - val_loss: 0.5607 - val_accuracy: 0.8375\n",
            "Epoch 3/30\n",
            "12/12 [==============================] - 4s 373ms/step - loss: 0.0080 - accuracy: 0.9948 - val_loss: 0.5800 - val_accuracy: 0.8438\n",
            "Epoch 4/30\n",
            "12/12 [==============================] - 4s 356ms/step - loss: 0.0212 - accuracy: 0.9896 - val_loss: 0.5659 - val_accuracy: 0.8250\n",
            "Epoch 5/30\n",
            "12/12 [==============================] - 4s 355ms/step - loss: 0.0061 - accuracy: 0.9974 - val_loss: 0.5547 - val_accuracy: 0.8250\n",
            "Epoch 6/30\n",
            "12/12 [==============================] - 4s 359ms/step - loss: 0.0166 - accuracy: 0.9948 - val_loss: 0.5878 - val_accuracy: 0.8250\n",
            "Epoch 7/30\n",
            "12/12 [==============================] - 4s 356ms/step - loss: 0.0073 - accuracy: 0.9974 - val_loss: 0.6168 - val_accuracy: 0.8188\n",
            "Epoch 8/30\n",
            "12/12 [==============================] - 4s 359ms/step - loss: 0.0038 - accuracy: 1.0000 - val_loss: 0.6178 - val_accuracy: 0.8188\n",
            "Epoch 9/30\n",
            "12/12 [==============================] - 4s 371ms/step - loss: 0.0094 - accuracy: 0.9948 - val_loss: 0.5926 - val_accuracy: 0.8188\n",
            "Epoch 10/30\n",
            "12/12 [==============================] - 4s 371ms/step - loss: 0.0105 - accuracy: 0.9948 - val_loss: 0.5634 - val_accuracy: 0.8188\n",
            "Epoch 11/30\n",
            "12/12 [==============================] - 4s 362ms/step - loss: 0.0092 - accuracy: 0.9974 - val_loss: 0.5334 - val_accuracy: 0.8062\n",
            "Epoch 12/30\n",
            "12/12 [==============================] - 4s 362ms/step - loss: 0.0024 - accuracy: 1.0000 - val_loss: 0.5145 - val_accuracy: 0.8125\n",
            "Epoch 13/30\n",
            "12/12 [==============================] - 4s 362ms/step - loss: 0.0026 - accuracy: 1.0000 - val_loss: 0.5025 - val_accuracy: 0.8188\n",
            "Epoch 14/30\n",
            "12/12 [==============================] - 4s 362ms/step - loss: 0.0095 - accuracy: 0.9974 - val_loss: 0.5211 - val_accuracy: 0.8125\n",
            "Epoch 15/30\n",
            "12/12 [==============================] - 4s 368ms/step - loss: 0.0069 - accuracy: 1.0000 - val_loss: 0.5174 - val_accuracy: 0.8188\n",
            "Epoch 16/30\n",
            "12/12 [==============================] - 4s 366ms/step - loss: 0.0082 - accuracy: 0.9948 - val_loss: 0.5218 - val_accuracy: 0.8313\n",
            "Epoch 17/30\n",
            "12/12 [==============================] - 4s 368ms/step - loss: 0.0082 - accuracy: 0.9974 - val_loss: 0.5357 - val_accuracy: 0.8375\n",
            "Epoch 18/30\n",
            "12/12 [==============================] - 4s 366ms/step - loss: 0.0049 - accuracy: 0.9974 - val_loss: 0.5291 - val_accuracy: 0.8438\n",
            "Epoch 19/30\n",
            "12/12 [==============================] - 4s 357ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 0.5303 - val_accuracy: 0.8438\n",
            "Epoch 20/30\n",
            "12/12 [==============================] - 4s 358ms/step - loss: 0.0028 - accuracy: 1.0000 - val_loss: 0.5325 - val_accuracy: 0.8438\n",
            "Epoch 21/30\n",
            "12/12 [==============================] - 4s 364ms/step - loss: 0.0042 - accuracy: 1.0000 - val_loss: 0.5370 - val_accuracy: 0.8438\n",
            "Epoch 22/30\n",
            "12/12 [==============================] - 4s 347ms/step - loss: 9.3324e-04 - accuracy: 1.0000 - val_loss: 0.5365 - val_accuracy: 0.8375\n",
            "Epoch 23/30\n",
            "12/12 [==============================] - 4s 353ms/step - loss: 0.0093 - accuracy: 0.9974 - val_loss: 0.5220 - val_accuracy: 0.8375\n",
            "Epoch 24/30\n",
            "12/12 [==============================] - 4s 352ms/step - loss: 0.0036 - accuracy: 1.0000 - val_loss: 0.5199 - val_accuracy: 0.8375\n",
            "Epoch 25/30\n",
            "12/12 [==============================] - 4s 356ms/step - loss: 0.0043 - accuracy: 0.9974 - val_loss: 0.4923 - val_accuracy: 0.8562\n",
            "Epoch 26/30\n",
            "12/12 [==============================] - 4s 359ms/step - loss: 8.9625e-04 - accuracy: 1.0000 - val_loss: 0.4872 - val_accuracy: 0.8500\n",
            "Epoch 27/30\n",
            "12/12 [==============================] - 4s 354ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 0.4867 - val_accuracy: 0.8500\n",
            "Epoch 28/30\n",
            "12/12 [==============================] - 4s 356ms/step - loss: 0.0046 - accuracy: 0.9974 - val_loss: 0.4939 - val_accuracy: 0.8500\n",
            "Epoch 29/30\n",
            "12/12 [==============================] - 4s 348ms/step - loss: 0.0027 - accuracy: 0.9974 - val_loss: 0.4964 - val_accuracy: 0.8500\n",
            "Epoch 30/30\n",
            "12/12 [==============================] - 4s 356ms/step - loss: 0.0010 - accuracy: 1.0000 - val_loss: 0.4915 - val_accuracy: 0.8625\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IjGzpBf08Jb4",
        "colab_type": "text"
      },
      "source": [
        "### Step 4: Unfreeze all Layers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N8Xs0l-Hz9eF",
        "colab_type": "code",
        "outputId": "45ac68dd-507c-41ca-b6dc-a32411747bbe",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "# Create a new model instance\n",
        "model = create_resnet50_model(len(class_names))\n",
        "\n",
        "# Restore the weights\n",
        "model.load_weights('./step-3/checkpoint')\n",
        "\n",
        "model = unfreeze(model, lr=0.00003)\n",
        "\n",
        "nr_epochs = 150\n",
        "batch_size = 32\n",
        "\n",
        "train_ds = create_dataset(train_image_paths, train_image_labels, epochs=nr_epochs, batch_size=batch_size, train=True)\n",
        "test_ds = create_dataset(test_image_paths, test_image_labels, epochs=nr_epochs, batch_size=batch_size, train=False)\n",
        "\n",
        "history = model.fit(train_ds,\n",
        "                    epochs = nr_epochs,\n",
        "                    steps_per_epoch = steps_per_epoch,\n",
        "                    validation_data = test_ds, \n",
        "                    validation_steps = validation_steps,\n",
        "                    callbacks = [])\n",
        "\n",
        "model.save_weights('./step-4/checkpoint')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Unfreezed Layer 0: <tensorflow.python.keras.engine.input_layer.InputLayer object at 0x7f5bffff1048>\n",
            "Unfreezed Layer 1: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5c06a0d518>\n",
            "Unfreezed Layer 2: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5befaa3a20>\n",
            "Unfreezed Layer 3: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bffff1f98>\n",
            "Unfreezed Layer 4: <tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x7f5bee23f588>\n",
            "Unfreezed Layer 5: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c0e168860>\n",
            "Unfreezed Layer 6: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffff8588>\n",
            "Unfreezed Layer 7: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c00068518>\n",
            "Unfreezed Layer 8: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c00070710>\n",
            "Unfreezed Layer 9: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c00070f98>\n",
            "Unfreezed Layer 10: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5c00009b00>\n",
            "Unfreezed Layer 11: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c000490f0>\n",
            "Unfreezed Layer 12: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c000358d0>\n",
            "Unfreezed Layer 13: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bfff821d0>\n",
            "Unfreezed Layer 14: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bee26a6a0>\n",
            "Unfreezed Layer 15: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bfff4d5f8>\n",
            "Unfreezed Layer 16: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bfff79470>\n",
            "Unfreezed Layer 17: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bfff794e0>\n",
            "Unfreezed Layer 18: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bfff02940>\n",
            "Unfreezed Layer 19: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bfff028d0>\n",
            "Unfreezed Layer 20: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffefb8d0>\n",
            "Unfreezed Layer 21: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffe8b1d0>\n",
            "Unfreezed Layer 22: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bffe52630>\n",
            "Unfreezed Layer 23: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffe8b2b0>\n",
            "Unfreezed Layer 24: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffdffa90>\n",
            "Unfreezed Layer 25: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffe0d390>\n",
            "Unfreezed Layer 26: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffdd1e48>\n",
            "Unfreezed Layer 27: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bffd83630>\n",
            "Unfreezed Layer 28: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffd836a0>\n",
            "Unfreezed Layer 29: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffd89b00>\n",
            "Unfreezed Layer 30: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffd06860>\n",
            "Unfreezed Layer 31: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffd06d68>\n",
            "Unfreezed Layer 32: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffd16908>\n",
            "Unfreezed Layer 33: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bffccef60>\n",
            "Unfreezed Layer 34: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffd169e8>\n",
            "Unfreezed Layer 35: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffc88f28>\n",
            "Unfreezed Layer 36: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffc9aac8>\n",
            "Unfreezed Layer 37: <tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x7f5bffd4df98>\n",
            "Unfreezed Layer 38: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffc9aa90>\n",
            "Unfreezed Layer 39: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bffc0ad68>\n",
            "Unfreezed Layer 40: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffc0ae80>\n",
            "Unfreezed Layer 41: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffc30278>\n",
            "Unfreezed Layer 42: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffba2208>\n",
            "Unfreezed Layer 43: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffbb1400>\n",
            "Unfreezed Layer 44: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffb8e9b0>\n",
            "Unfreezed Layer 45: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bffb3f0f0>\n",
            "Unfreezed Layer 46: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffbb1da0>\n",
            "Unfreezed Layer 47: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffb365c0>\n",
            "Unfreezed Layer 48: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffac0080>\n",
            "Unfreezed Layer 49: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffbf9a20>\n",
            "Unfreezed Layer 50: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffb36f60>\n",
            "Unfreezed Layer 51: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bffaa4fd0>\n",
            "Unfreezed Layer 52: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffab91d0>\n",
            "Unfreezed Layer 53: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bffa47630>\n",
            "Unfreezed Layer 54: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bffa0ddd8>\n",
            "Unfreezed Layer 55: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bffa23d68>\n",
            "Unfreezed Layer 56: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bff9cb080>\n",
            "Unfreezed Layer 57: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bff9bfe80>\n",
            "Unfreezed Layer 58: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bff997320>\n",
            "Unfreezed Layer 59: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bff942780>\n",
            "Unfreezed Layer 60: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bff942fd0>\n",
            "Unfreezed Layer 61: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bff91ab38>\n",
            "Unfreezed Layer 62: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bee1fd320>\n",
            "Unfreezed Layer 63: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bee1fd390>\n",
            "Unfreezed Layer 64: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bee2047f0>\n",
            "Unfreezed Layer 65: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bee1cb940>\n",
            "Unfreezed Layer 66: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bee182780>\n",
            "Unfreezed Layer 67: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bee182fd0>\n",
            "Unfreezed Layer 68: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bee157b70>\n",
            "Unfreezed Layer 69: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bee18e1d0>\n",
            "Unfreezed Layer 70: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bee106940>\n",
            "Unfreezed Layer 71: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bee111240>\n",
            "Unfreezed Layer 72: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bee0d9cf8>\n",
            "Unfreezed Layer 73: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bee0894e0>\n",
            "Unfreezed Layer 74: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bee089550>\n",
            "Unfreezed Layer 75: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bee08e9b0>\n",
            "Unfreezed Layer 76: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bee00c710>\n",
            "Unfreezed Layer 77: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bee00ceb8>\n",
            "Unfreezed Layer 78: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bee01b7b8>\n",
            "Unfreezed Layer 79: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bedfd8f98>\n",
            "Unfreezed Layer 80: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bee01b898>\n",
            "Unfreezed Layer 81: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bedf8ff28>\n",
            "Unfreezed Layer 82: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bedfa1978>\n",
            "Unfreezed Layer 83: <tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x7f5bee08e940>\n",
            "Unfreezed Layer 84: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bedf5cf98>\n",
            "Unfreezed Layer 85: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bedf13c18>\n",
            "Unfreezed Layer 86: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bedf13d30>\n",
            "Unfreezed Layer 87: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bedf20f98>\n",
            "Unfreezed Layer 88: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bede96f28>\n",
            "Unfreezed Layer 89: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bedeabfd0>\n",
            "Unfreezed Layer 90: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bede3cb70>\n",
            "Unfreezed Layer 91: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bede3cb38>\n",
            "Unfreezed Layer 92: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bede11f98>\n",
            "Unfreezed Layer 93: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5beddbf470>\n",
            "Unfreezed Layer 94: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bede96860>\n",
            "Unfreezed Layer 95: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bedf23048>\n",
            "Unfreezed Layer 96: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5beddca160>\n",
            "Unfreezed Layer 97: <tensorflow.python.keras.layers.merge.Add object at 0x7f5beddadfd0>\n",
            "Unfreezed Layer 98: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5beddadef0>\n",
            "Unfreezed Layer 99: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bedd494e0>\n",
            "Unfreezed Layer 100: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bedd1bc88>\n",
            "Unfreezed Layer 101: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bedcc9470>\n",
            "Unfreezed Layer 102: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bedcd4160>\n",
            "Unfreezed Layer 103: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bf60b9d30>\n",
            "Unfreezed Layer 104: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bedcc9e10>\n",
            "Unfreezed Layer 105: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5befaa4d68>\n",
            "Unfreezed Layer 106: <tensorflow.python.keras.layers.core.Activation object at 0x7f5befaa3898>\n",
            "Unfreezed Layer 107: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5befab6e80>\n",
            "Unfreezed Layer 108: <tensorflow.python.keras.layers.merge.Add object at 0x7f5befab46a0>\n",
            "Unfreezed Layer 109: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5befab48d0>\n",
            "Unfreezed Layer 110: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf5397f98>\n",
            "Unfreezed Layer 111: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf4a4b748>\n",
            "Unfreezed Layer 112: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c012c1550>\n",
            "Unfreezed Layer 113: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0a7ef438>\n",
            "Unfreezed Layer 114: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5c0797df98>\n",
            "Unfreezed Layer 115: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c0a7efbe0>\n",
            "Unfreezed Layer 116: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c07bb7be0>\n",
            "Unfreezed Layer 117: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf9c7c7b8>\n",
            "Unfreezed Layer 118: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c085ef8d0>\n",
            "Unfreezed Layer 119: <tensorflow.python.keras.layers.merge.Add object at 0x7f5c07beef28>\n",
            "Unfreezed Layer 120: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c07beeac8>\n",
            "Unfreezed Layer 121: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c05dc0898>\n",
            "Unfreezed Layer 122: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c06233358>\n",
            "Unfreezed Layer 123: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c05ba8518>\n",
            "Unfreezed Layer 124: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c05b9fef0>\n",
            "Unfreezed Layer 125: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5c05ba9780>\n",
            "Unfreezed Layer 126: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c05b7fdd8>\n",
            "Unfreezed Layer 127: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c05dd3978>\n",
            "Unfreezed Layer 128: <tensorflow.python.keras.layers.core.Activation object at 0x7f5cd6b285f8>\n",
            "Unfreezed Layer 129: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c05a8a080>\n",
            "Unfreezed Layer 130: <tensorflow.python.keras.layers.merge.Add object at 0x7f5c0e161eb8>\n",
            "Unfreezed Layer 131: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c045cca20>\n",
            "Unfreezed Layer 132: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0e1685f8>\n",
            "Unfreezed Layer 133: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bfe5adbe0>\n",
            "Unfreezed Layer 134: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bfe6494a8>\n",
            "Unfreezed Layer 135: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c01395fd0>\n",
            "Unfreezed Layer 136: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bfef48ac8>\n",
            "Unfreezed Layer 137: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c01395d68>\n",
            "Unfreezed Layer 138: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c013e9d30>\n",
            "Unfreezed Layer 139: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0146ae48>\n",
            "Unfreezed Layer 140: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c026f8940>\n",
            "Unfreezed Layer 141: <tensorflow.python.keras.layers.merge.Add object at 0x7f5c005b5c18>\n",
            "Unfreezed Layer 142: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c005b52b0>\n",
            "Unfreezed Layer 143: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0045a240>\n",
            "Unfreezed Layer 144: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c053bdb70>\n",
            "Unfreezed Layer 145: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c053bdcf8>\n",
            "Unfreezed Layer 146: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0784d7f0>\n",
            "Unfreezed Layer 147: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bfdcf9f98>\n",
            "Unfreezed Layer 148: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c0784d630>\n",
            "Unfreezed Layer 149: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bfdce71d0>\n",
            "Unfreezed Layer 150: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf5d9bba8>\n",
            "Unfreezed Layer 151: <tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x7f5c0a76d550>\n",
            "Unfreezed Layer 152: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf5d91ac8>\n",
            "Unfreezed Layer 153: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bf5d81550>\n",
            "Unfreezed Layer 154: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf5d81358>\n",
            "Unfreezed Layer 155: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf5d7c3c8>\n",
            "Unfreezed Layer 156: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf2e47240>\n",
            "Unfreezed Layer 157: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c0022d780>\n",
            "Unfreezed Layer 158: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0022d160>\n",
            "Unfreezed Layer 159: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5c00219080>\n",
            "Unfreezed Layer 160: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf48e8320>\n",
            "Unfreezed Layer 161: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf48e0160>\n",
            "Unfreezed Layer 162: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf4964e10>\n",
            "Unfreezed Layer 163: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bf5d7ceb8>\n",
            "Unfreezed Layer 164: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c026e4fd0>\n",
            "Unfreezed Layer 165: <tensorflow.python.keras.layers.merge.Add object at 0x7f5c027deb70>\n",
            "Unfreezed Layer 166: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c027dec88>\n",
            "Unfreezed Layer 167: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c026cdef0>\n",
            "Unfreezed Layer 168: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c027e0f98>\n",
            "Unfreezed Layer 169: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c07811fd0>\n",
            "Unfreezed Layer 170: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0292f8d0>\n",
            "Unfreezed Layer 171: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5c0293df28>\n",
            "Unfreezed Layer 172: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c0292f9b0>\n",
            "Unfreezed Layer 173: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c00240f98>\n",
            "Unfreezed Layer 174: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0026da58>\n",
            "Unfreezed Layer 175: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c0026da20>\n",
            "Unfreezed Layer 176: <tensorflow.python.keras.layers.merge.Add object at 0x7f5c03281d30>\n",
            "Unfreezed Layer 177: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5c03281e48>\n",
            "Unfreezed Layer 178: <tensorflow.python.keras.layers.core.Activation object at 0x7f5c0327f240>\n",
            "Unfreezed Layer 179: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5c0327f1d0>\n",
            "Unfreezed Layer 180: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bf3dcefd0>\n",
            "Unfreezed Layer 181: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bf3dd0a58>\n",
            "Unfreezed Layer 182: <tensorflow.python.keras.layers.convolutional.ZeroPadding2D object at 0x7f5bf3dd0a20>\n",
            "Unfreezed Layer 183: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bfefa4eb8>\n",
            "Unfreezed Layer 184: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bfe5fe390>\n",
            "Unfreezed Layer 185: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bfefa8cc0>\n",
            "Unfreezed Layer 186: <tensorflow.python.keras.layers.convolutional.Conv2D object at 0x7f5bfe62f080>\n",
            "Unfreezed Layer 187: <tensorflow.python.keras.layers.merge.Add object at 0x7f5bff5bbef0>\n",
            "Unfreezed Layer 188: <tensorflow.python.keras.layers.normalization.BatchNormalization object at 0x7f5bff5bbd68>\n",
            "Unfreezed Layer 189: <tensorflow.python.keras.layers.core.Activation object at 0x7f5bff587400>\n",
            "Unfreezed Layer 190: <tensorflow.python.keras.layers.pooling.GlobalAveragePooling2D object at 0x7f5bff587358>\n",
            "Unfreezed Layer 191: <tensorflow.python.keras.layers.pooling.GlobalMaxPooling2D object at 0x7f5c001204e0>\n",
            "Unfreezed Layer 192: <tensorflow.python.keras.layers.merge.Concatenate object at 0x7f5bf28ed390>\n",
            "Unfreezed Layer 193: <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7f5bf28ed080>\n",
            "Unfreezed Layer 194: <tensorflow.python.keras.layers.core.Dropout object at 0x7f5bf28e7358>\n",
            "Unfreezed Layer 195: <tensorflow.python.keras.layers.core.Dense object at 0x7f5bf28e72e8>\n",
            "Unfreezed Layer 196: <tensorflow.python.keras.layers.normalization_v2.BatchNormalization object at 0x7f5c0a7e5ac8>\n",
            "Unfreezed Layer 197: <tensorflow.python.keras.layers.core.Dropout object at 0x7f5c0a7faa58>\n",
            "Unfreezed Layer 198: <tensorflow.python.keras.layers.core.Dense object at 0x7f5c0266f2b0>\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay\n",
            "WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate\n",
            "WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/alpha/guide/checkpoints#loading_mechanics for details.\n",
            "Train for 12 steps, validate for 5 steps\n",
            "Epoch 1/150\n",
            "12/12 [==============================] - 24s 2s/step - loss: 0.0112 - accuracy: 0.9974 - val_loss: 0.5710 - val_accuracy: 0.8375\n",
            "Epoch 2/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0024 - accuracy: 1.0000 - val_loss: 0.6236 - val_accuracy: 0.8375\n",
            "Epoch 3/150\n",
            "12/12 [==============================] - 5s 455ms/step - loss: 0.0104 - accuracy: 0.9974 - val_loss: 0.6543 - val_accuracy: 0.8562\n",
            "Epoch 4/150\n",
            "12/12 [==============================] - 5s 452ms/step - loss: 0.0205 - accuracy: 0.9922 - val_loss: 0.6781 - val_accuracy: 0.8375\n",
            "Epoch 5/150\n",
            "12/12 [==============================] - 5s 436ms/step - loss: 0.0038 - accuracy: 0.9974 - val_loss: 0.7903 - val_accuracy: 0.8313\n",
            "Epoch 6/150\n",
            "12/12 [==============================] - 5s 442ms/step - loss: 0.0115 - accuracy: 0.9948 - val_loss: 0.7158 - val_accuracy: 0.8250\n",
            "Epoch 7/150\n",
            "12/12 [==============================] - 5s 436ms/step - loss: 0.0099 - accuracy: 0.9948 - val_loss: 0.6891 - val_accuracy: 0.8125\n",
            "Epoch 8/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 0.0066 - accuracy: 0.9974 - val_loss: 0.7275 - val_accuracy: 0.8125\n",
            "Epoch 9/150\n",
            "12/12 [==============================] - 5s 444ms/step - loss: 0.0104 - accuracy: 0.9948 - val_loss: 0.6723 - val_accuracy: 0.8125\n",
            "Epoch 10/150\n",
            "12/12 [==============================] - 5s 437ms/step - loss: 0.0045 - accuracy: 1.0000 - val_loss: 0.6869 - val_accuracy: 0.8250\n",
            "Epoch 11/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 0.0189 - accuracy: 0.9948 - val_loss: 0.7263 - val_accuracy: 0.8250\n",
            "Epoch 12/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0134 - accuracy: 0.9974 - val_loss: 0.7118 - val_accuracy: 0.8313\n",
            "Epoch 13/150\n",
            "12/12 [==============================] - 5s 443ms/step - loss: 0.0064 - accuracy: 1.0000 - val_loss: 0.7076 - val_accuracy: 0.8375\n",
            "Epoch 14/150\n",
            "12/12 [==============================] - 5s 445ms/step - loss: 0.0041 - accuracy: 1.0000 - val_loss: 0.7219 - val_accuracy: 0.8438\n",
            "Epoch 15/150\n",
            "12/12 [==============================] - 5s 443ms/step - loss: 9.8906e-04 - accuracy: 1.0000 - val_loss: 0.7105 - val_accuracy: 0.8687\n",
            "Epoch 16/150\n",
            "12/12 [==============================] - 5s 443ms/step - loss: 0.0075 - accuracy: 0.9974 - val_loss: 0.6825 - val_accuracy: 0.8687\n",
            "Epoch 17/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 0.6684 - val_accuracy: 0.8687\n",
            "Epoch 18/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 0.0136 - accuracy: 0.9948 - val_loss: 0.6839 - val_accuracy: 0.8687\n",
            "Epoch 19/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 0.0017 - accuracy: 1.0000 - val_loss: 0.6804 - val_accuracy: 0.8625\n",
            "Epoch 20/150\n",
            "12/12 [==============================] - 5s 433ms/step - loss: 0.0017 - accuracy: 1.0000 - val_loss: 0.6730 - val_accuracy: 0.8562\n",
            "Epoch 21/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 0.0027 - accuracy: 1.0000 - val_loss: 0.6537 - val_accuracy: 0.8500\n",
            "Epoch 22/150\n",
            "12/12 [==============================] - 5s 436ms/step - loss: 0.0017 - accuracy: 1.0000 - val_loss: 0.6652 - val_accuracy: 0.8562\n",
            "Epoch 23/150\n",
            "12/12 [==============================] - 5s 432ms/step - loss: 0.0190 - accuracy: 0.9948 - val_loss: 0.6750 - val_accuracy: 0.8625\n",
            "Epoch 24/150\n",
            "12/12 [==============================] - 5s 430ms/step - loss: 0.0060 - accuracy: 1.0000 - val_loss: 0.6754 - val_accuracy: 0.8562\n",
            "Epoch 25/150\n",
            "12/12 [==============================] - 5s 431ms/step - loss: 0.0122 - accuracy: 0.9974 - val_loss: 0.7087 - val_accuracy: 0.8562\n",
            "Epoch 26/150\n",
            "12/12 [==============================] - 5s 436ms/step - loss: 0.0071 - accuracy: 0.9974 - val_loss: 0.7260 - val_accuracy: 0.8438\n",
            "Epoch 27/150\n",
            "12/12 [==============================] - 5s 440ms/step - loss: 0.0031 - accuracy: 1.0000 - val_loss: 0.7262 - val_accuracy: 0.8250\n",
            "Epoch 28/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 0.7179 - val_accuracy: 0.8313\n",
            "Epoch 29/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 0.0037 - accuracy: 1.0000 - val_loss: 0.7010 - val_accuracy: 0.8562\n",
            "Epoch 30/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0051 - accuracy: 0.9974 - val_loss: 0.7164 - val_accuracy: 0.8562\n",
            "Epoch 31/150\n",
            "12/12 [==============================] - 5s 432ms/step - loss: 0.0027 - accuracy: 1.0000 - val_loss: 0.7417 - val_accuracy: 0.8438\n",
            "Epoch 32/150\n",
            "12/12 [==============================] - 5s 432ms/step - loss: 6.7898e-04 - accuracy: 1.0000 - val_loss: 0.7463 - val_accuracy: 0.8438\n",
            "Epoch 33/150\n",
            "12/12 [==============================] - 5s 428ms/step - loss: 0.0025 - accuracy: 1.0000 - val_loss: 0.7231 - val_accuracy: 0.8500\n",
            "Epoch 34/150\n",
            "12/12 [==============================] - 5s 426ms/step - loss: 0.0046 - accuracy: 0.9974 - val_loss: 0.7037 - val_accuracy: 0.8500\n",
            "Epoch 35/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0082 - accuracy: 0.9974 - val_loss: 0.6944 - val_accuracy: 0.8625\n",
            "Epoch 36/150\n",
            "12/12 [==============================] - 5s 445ms/step - loss: 0.0018 - accuracy: 1.0000 - val_loss: 0.6832 - val_accuracy: 0.8625\n",
            "Epoch 37/150\n",
            "12/12 [==============================] - 5s 432ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 0.6989 - val_accuracy: 0.8562\n",
            "Epoch 38/150\n",
            "12/12 [==============================] - 5s 447ms/step - loss: 0.0055 - accuracy: 0.9974 - val_loss: 0.7093 - val_accuracy: 0.8562\n",
            "Epoch 39/150\n",
            "12/12 [==============================] - 6s 465ms/step - loss: 0.0079 - accuracy: 0.9974 - val_loss: 0.6717 - val_accuracy: 0.8500\n",
            "Epoch 40/150\n",
            "12/12 [==============================] - 5s 454ms/step - loss: 0.0059 - accuracy: 0.9974 - val_loss: 0.6318 - val_accuracy: 0.8687\n",
            "Epoch 41/150\n",
            "12/12 [==============================] - 5s 455ms/step - loss: 9.4442e-04 - accuracy: 1.0000 - val_loss: 0.6367 - val_accuracy: 0.8625\n",
            "Epoch 42/150\n",
            "12/12 [==============================] - 5s 452ms/step - loss: 0.0042 - accuracy: 0.9974 - val_loss: 0.6151 - val_accuracy: 0.8500\n",
            "Epoch 43/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 0.0017 - accuracy: 1.0000 - val_loss: 0.6065 - val_accuracy: 0.8500\n",
            "Epoch 44/150\n",
            "12/12 [==============================] - 5s 436ms/step - loss: 0.0027 - accuracy: 1.0000 - val_loss: 0.6270 - val_accuracy: 0.8313\n",
            "Epoch 45/150\n",
            "12/12 [==============================] - 5s 440ms/step - loss: 0.0020 - accuracy: 1.0000 - val_loss: 0.6281 - val_accuracy: 0.8438\n",
            "Epoch 46/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 0.0019 - accuracy: 1.0000 - val_loss: 0.6264 - val_accuracy: 0.8500\n",
            "Epoch 47/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 5.1365e-04 - accuracy: 1.0000 - val_loss: 0.6202 - val_accuracy: 0.8500\n",
            "Epoch 48/150\n",
            "12/12 [==============================] - 5s 433ms/step - loss: 6.0006e-04 - accuracy: 1.0000 - val_loss: 0.6258 - val_accuracy: 0.8562\n",
            "Epoch 49/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0035 - accuracy: 0.9974 - val_loss: 0.6197 - val_accuracy: 0.8562\n",
            "Epoch 50/150\n",
            "12/12 [==============================] - 5s 437ms/step - loss: 0.0010 - accuracy: 1.0000 - val_loss: 0.5673 - val_accuracy: 0.8625\n",
            "Epoch 51/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0014 - accuracy: 1.0000 - val_loss: 0.5464 - val_accuracy: 0.8500\n",
            "Epoch 52/150\n",
            "12/12 [==============================] - 5s 451ms/step - loss: 0.0019 - accuracy: 1.0000 - val_loss: 0.5521 - val_accuracy: 0.8500\n",
            "Epoch 53/150\n",
            "12/12 [==============================] - 5s 441ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 0.5673 - val_accuracy: 0.8500\n",
            "Epoch 54/150\n",
            "12/12 [==============================] - 5s 443ms/step - loss: 9.4635e-04 - accuracy: 1.0000 - val_loss: 0.5816 - val_accuracy: 0.8625\n",
            "Epoch 55/150\n",
            "12/12 [==============================] - 5s 433ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 0.5827 - val_accuracy: 0.8562\n",
            "Epoch 56/150\n",
            "12/12 [==============================] - 5s 431ms/step - loss: 0.0026 - accuracy: 1.0000 - val_loss: 0.5779 - val_accuracy: 0.8562\n",
            "Epoch 57/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 0.0054 - accuracy: 0.9974 - val_loss: 0.5728 - val_accuracy: 0.8500\n",
            "Epoch 58/150\n",
            "12/12 [==============================] - 5s 431ms/step - loss: 0.0052 - accuracy: 1.0000 - val_loss: 0.5420 - val_accuracy: 0.8438\n",
            "Epoch 59/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 3.5778e-04 - accuracy: 1.0000 - val_loss: 0.5449 - val_accuracy: 0.8375\n",
            "Epoch 60/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 9.1883e-04 - accuracy: 1.0000 - val_loss: 0.5434 - val_accuracy: 0.8375\n",
            "Epoch 61/150\n",
            "12/12 [==============================] - 5s 430ms/step - loss: 0.0028 - accuracy: 1.0000 - val_loss: 0.5455 - val_accuracy: 0.8375\n",
            "Epoch 62/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 9.4373e-04 - accuracy: 1.0000 - val_loss: 0.5676 - val_accuracy: 0.8313\n",
            "Epoch 63/150\n",
            "12/12 [==============================] - 5s 436ms/step - loss: 0.0023 - accuracy: 1.0000 - val_loss: 0.5570 - val_accuracy: 0.8438\n",
            "Epoch 64/150\n",
            "12/12 [==============================] - 5s 432ms/step - loss: 0.0026 - accuracy: 1.0000 - val_loss: 0.5562 - val_accuracy: 0.8500\n",
            "Epoch 65/150\n",
            "12/12 [==============================] - 5s 431ms/step - loss: 6.3932e-04 - accuracy: 1.0000 - val_loss: 0.5454 - val_accuracy: 0.8625\n",
            "Epoch 66/150\n",
            "12/12 [==============================] - 5s 437ms/step - loss: 0.0027 - accuracy: 0.9974 - val_loss: 0.5473 - val_accuracy: 0.8562\n",
            "Epoch 67/150\n",
            "12/12 [==============================] - 5s 437ms/step - loss: 0.0065 - accuracy: 0.9974 - val_loss: 0.5435 - val_accuracy: 0.8687\n",
            "Epoch 68/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 0.5557 - val_accuracy: 0.8625\n",
            "Epoch 69/150\n",
            "12/12 [==============================] - 5s 440ms/step - loss: 0.0042 - accuracy: 0.9974 - val_loss: 0.5970 - val_accuracy: 0.8750\n",
            "Epoch 70/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 0.0011 - accuracy: 1.0000 - val_loss: 0.6209 - val_accuracy: 0.8750\n",
            "Epoch 71/150\n",
            "12/12 [==============================] - 5s 440ms/step - loss: 0.0166 - accuracy: 0.9974 - val_loss: 0.5784 - val_accuracy: 0.8500\n",
            "Epoch 72/150\n",
            "12/12 [==============================] - 5s 450ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 0.5578 - val_accuracy: 0.8687\n",
            "Epoch 73/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0194 - accuracy: 0.9922 - val_loss: 0.5347 - val_accuracy: 0.8687\n",
            "Epoch 74/150\n",
            "12/12 [==============================] - 5s 449ms/step - loss: 5.3014e-04 - accuracy: 1.0000 - val_loss: 0.4959 - val_accuracy: 0.8875\n",
            "Epoch 75/150\n",
            "12/12 [==============================] - 5s 452ms/step - loss: 0.0070 - accuracy: 0.9974 - val_loss: 0.5028 - val_accuracy: 0.8687\n",
            "Epoch 76/150\n",
            "12/12 [==============================] - 5s 442ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 0.5222 - val_accuracy: 0.8625\n",
            "Epoch 77/150\n",
            "12/12 [==============================] - 5s 449ms/step - loss: 7.7342e-04 - accuracy: 1.0000 - val_loss: 0.5218 - val_accuracy: 0.8687\n",
            "Epoch 78/150\n",
            "12/12 [==============================] - 5s 448ms/step - loss: 0.0146 - accuracy: 0.9974 - val_loss: 0.5307 - val_accuracy: 0.8625\n",
            "Epoch 79/150\n",
            "12/12 [==============================] - 5s 440ms/step - loss: 4.8768e-04 - accuracy: 1.0000 - val_loss: 0.5441 - val_accuracy: 0.8687\n",
            "Epoch 80/150\n",
            "12/12 [==============================] - 5s 456ms/step - loss: 0.0037 - accuracy: 1.0000 - val_loss: 0.5503 - val_accuracy: 0.8813\n",
            "Epoch 81/150\n",
            "12/12 [==============================] - 5s 456ms/step - loss: 0.0018 - accuracy: 1.0000 - val_loss: 0.5554 - val_accuracy: 0.8625\n",
            "Epoch 82/150\n",
            "12/12 [==============================] - 5s 453ms/step - loss: 0.0090 - accuracy: 1.0000 - val_loss: 0.5622 - val_accuracy: 0.8687\n",
            "Epoch 83/150\n",
            "12/12 [==============================] - 5s 457ms/step - loss: 0.0080 - accuracy: 0.9974 - val_loss: 0.6475 - val_accuracy: 0.8625\n",
            "Epoch 84/150\n",
            "12/12 [==============================] - 5s 454ms/step - loss: 0.0290 - accuracy: 0.9896 - val_loss: 0.6107 - val_accuracy: 0.8562\n",
            "Epoch 85/150\n",
            "12/12 [==============================] - 5s 453ms/step - loss: 0.0062 - accuracy: 1.0000 - val_loss: 0.6171 - val_accuracy: 0.8500\n",
            "Epoch 86/150\n",
            "12/12 [==============================] - 5s 452ms/step - loss: 0.0121 - accuracy: 0.9974 - val_loss: 0.6492 - val_accuracy: 0.8625\n",
            "Epoch 87/150\n",
            "12/12 [==============================] - 5s 451ms/step - loss: 0.0084 - accuracy: 0.9974 - val_loss: 0.6327 - val_accuracy: 0.8562\n",
            "Epoch 88/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0075 - accuracy: 0.9922 - val_loss: 0.6818 - val_accuracy: 0.8500\n",
            "Epoch 89/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 0.0010 - accuracy: 1.0000 - val_loss: 0.6764 - val_accuracy: 0.8375\n",
            "Epoch 90/150\n",
            "12/12 [==============================] - 5s 443ms/step - loss: 0.0136 - accuracy: 0.9948 - val_loss: 0.6512 - val_accuracy: 0.8500\n",
            "Epoch 91/150\n",
            "12/12 [==============================] - 5s 430ms/step - loss: 0.0040 - accuracy: 0.9974 - val_loss: 0.6011 - val_accuracy: 0.8625\n",
            "Epoch 92/150\n",
            "12/12 [==============================] - 5s 432ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 0.5892 - val_accuracy: 0.8562\n",
            "Epoch 93/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 0.0047 - accuracy: 0.9974 - val_loss: 0.6016 - val_accuracy: 0.8625\n",
            "Epoch 94/150\n",
            "12/12 [==============================] - 5s 433ms/step - loss: 0.0055 - accuracy: 0.9974 - val_loss: 0.5892 - val_accuracy: 0.8375\n",
            "Epoch 95/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 0.0067 - accuracy: 0.9974 - val_loss: 0.6427 - val_accuracy: 0.8438\n",
            "Epoch 96/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 0.0196 - accuracy: 0.9922 - val_loss: 0.6499 - val_accuracy: 0.8500\n",
            "Epoch 97/150\n",
            "12/12 [==============================] - 5s 437ms/step - loss: 0.0148 - accuracy: 0.9974 - val_loss: 0.6679 - val_accuracy: 0.8500\n",
            "Epoch 98/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 0.0060 - accuracy: 1.0000 - val_loss: 0.6696 - val_accuracy: 0.8562\n",
            "Epoch 99/150\n",
            "12/12 [==============================] - 5s 446ms/step - loss: 0.0038 - accuracy: 1.0000 - val_loss: 0.6442 - val_accuracy: 0.8500\n",
            "Epoch 100/150\n",
            "12/12 [==============================] - 5s 432ms/step - loss: 0.0055 - accuracy: 0.9974 - val_loss: 0.6203 - val_accuracy: 0.8500\n",
            "Epoch 101/150\n",
            "12/12 [==============================] - 5s 432ms/step - loss: 0.0040 - accuracy: 0.9974 - val_loss: 0.6053 - val_accuracy: 0.8562\n",
            "Epoch 102/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 0.0062 - accuracy: 0.9974 - val_loss: 0.6010 - val_accuracy: 0.8687\n",
            "Epoch 103/150\n",
            "12/12 [==============================] - 5s 445ms/step - loss: 8.7725e-04 - accuracy: 1.0000 - val_loss: 0.5924 - val_accuracy: 0.8750\n",
            "Epoch 104/150\n",
            "12/12 [==============================] - 5s 441ms/step - loss: 0.0034 - accuracy: 0.9974 - val_loss: 0.5900 - val_accuracy: 0.8625\n",
            "Epoch 105/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 0.0020 - accuracy: 1.0000 - val_loss: 0.5805 - val_accuracy: 0.8687\n",
            "Epoch 106/150\n",
            "12/12 [==============================] - 5s 429ms/step - loss: 0.0162 - accuracy: 0.9948 - val_loss: 0.5807 - val_accuracy: 0.8750\n",
            "Epoch 107/150\n",
            "12/12 [==============================] - 5s 431ms/step - loss: 0.0079 - accuracy: 0.9948 - val_loss: 0.5389 - val_accuracy: 0.8562\n",
            "Epoch 108/150\n",
            "12/12 [==============================] - 5s 445ms/step - loss: 0.0020 - accuracy: 1.0000 - val_loss: 0.5436 - val_accuracy: 0.8500\n",
            "Epoch 109/150\n",
            "12/12 [==============================] - 5s 446ms/step - loss: 6.4375e-04 - accuracy: 1.0000 - val_loss: 0.5315 - val_accuracy: 0.8562\n",
            "Epoch 110/150\n",
            "12/12 [==============================] - 6s 462ms/step - loss: 0.0054 - accuracy: 0.9974 - val_loss: 0.5438 - val_accuracy: 0.8625\n",
            "Epoch 111/150\n",
            "12/12 [==============================] - 5s 441ms/step - loss: 0.0028 - accuracy: 1.0000 - val_loss: 0.5467 - val_accuracy: 0.8625\n",
            "Epoch 112/150\n",
            "12/12 [==============================] - 5s 447ms/step - loss: 0.0173 - accuracy: 0.9974 - val_loss: 0.4824 - val_accuracy: 0.8875\n",
            "Epoch 113/150\n",
            "12/12 [==============================] - 5s 449ms/step - loss: 8.4834e-04 - accuracy: 1.0000 - val_loss: 0.4720 - val_accuracy: 0.8875\n",
            "Epoch 114/150\n",
            "12/12 [==============================] - 5s 436ms/step - loss: 0.0019 - accuracy: 1.0000 - val_loss: 0.4649 - val_accuracy: 0.8938\n",
            "Epoch 115/150\n",
            "12/12 [==============================] - 5s 427ms/step - loss: 7.0565e-04 - accuracy: 1.0000 - val_loss: 0.4751 - val_accuracy: 0.8938\n",
            "Epoch 116/150\n",
            "12/12 [==============================] - 5s 447ms/step - loss: 0.0091 - accuracy: 0.9974 - val_loss: 0.4694 - val_accuracy: 0.8938\n",
            "Epoch 117/150\n",
            "12/12 [==============================] - 6s 462ms/step - loss: 0.0050 - accuracy: 0.9974 - val_loss: 0.4666 - val_accuracy: 0.8750\n",
            "Epoch 118/150\n",
            "12/12 [==============================] - 5s 451ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 0.4473 - val_accuracy: 0.8875\n",
            "Epoch 119/150\n",
            "12/12 [==============================] - 5s 447ms/step - loss: 7.0070e-04 - accuracy: 1.0000 - val_loss: 0.4409 - val_accuracy: 0.8875\n",
            "Epoch 120/150\n",
            "12/12 [==============================] - 5s 456ms/step - loss: 3.7396e-04 - accuracy: 1.0000 - val_loss: 0.4413 - val_accuracy: 0.8750\n",
            "Epoch 121/150\n",
            "12/12 [==============================] - 5s 447ms/step - loss: 6.7444e-04 - accuracy: 1.0000 - val_loss: 0.4356 - val_accuracy: 0.8938\n",
            "Epoch 122/150\n",
            "12/12 [==============================] - 5s 450ms/step - loss: 9.0728e-04 - accuracy: 1.0000 - val_loss: 0.4390 - val_accuracy: 0.8938\n",
            "Epoch 123/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 0.4398 - val_accuracy: 0.8938\n",
            "Epoch 124/150\n",
            "12/12 [==============================] - 5s 435ms/step - loss: 2.2633e-04 - accuracy: 1.0000 - val_loss: 0.4416 - val_accuracy: 0.9000\n",
            "Epoch 125/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 6.9257e-04 - accuracy: 1.0000 - val_loss: 0.4413 - val_accuracy: 0.8938\n",
            "Epoch 126/150\n",
            "12/12 [==============================] - 5s 430ms/step - loss: 7.5912e-04 - accuracy: 1.0000 - val_loss: 0.4365 - val_accuracy: 0.8938\n",
            "Epoch 127/150\n",
            "12/12 [==============================] - 5s 431ms/step - loss: 0.0018 - accuracy: 1.0000 - val_loss: 0.4397 - val_accuracy: 0.8938\n",
            "Epoch 128/150\n",
            "12/12 [==============================] - 5s 427ms/step - loss: 4.0779e-04 - accuracy: 1.0000 - val_loss: 0.4461 - val_accuracy: 0.8875\n",
            "Epoch 129/150\n",
            "12/12 [==============================] - 5s 424ms/step - loss: 3.4099e-04 - accuracy: 1.0000 - val_loss: 0.4500 - val_accuracy: 0.8813\n",
            "Epoch 130/150\n",
            "12/12 [==============================] - 5s 444ms/step - loss: 9.8217e-04 - accuracy: 1.0000 - val_loss: 0.4565 - val_accuracy: 0.8875\n",
            "Epoch 131/150\n",
            "12/12 [==============================] - 5s 448ms/step - loss: 4.0392e-04 - accuracy: 1.0000 - val_loss: 0.4590 - val_accuracy: 0.8875\n",
            "Epoch 132/150\n",
            "12/12 [==============================] - 5s 445ms/step - loss: 2.1642e-04 - accuracy: 1.0000 - val_loss: 0.4586 - val_accuracy: 0.8875\n",
            "Epoch 133/150\n",
            "12/12 [==============================] - 5s 440ms/step - loss: 1.0914e-04 - accuracy: 1.0000 - val_loss: 0.4563 - val_accuracy: 0.8938\n",
            "Epoch 134/150\n",
            "12/12 [==============================] - 5s 451ms/step - loss: 0.0013 - accuracy: 1.0000 - val_loss: 0.4621 - val_accuracy: 0.9000\n",
            "Epoch 135/150\n",
            "12/12 [==============================] - 5s 448ms/step - loss: 0.0037 - accuracy: 0.9974 - val_loss: 0.5304 - val_accuracy: 0.8875\n",
            "Epoch 136/150\n",
            "12/12 [==============================] - 5s 443ms/step - loss: 0.0010 - accuracy: 1.0000 - val_loss: 0.5693 - val_accuracy: 0.8750\n",
            "Epoch 137/150\n",
            "12/12 [==============================] - 5s 434ms/step - loss: 0.0108 - accuracy: 0.9948 - val_loss: 0.5515 - val_accuracy: 0.8687\n",
            "Epoch 138/150\n",
            "12/12 [==============================] - 5s 444ms/step - loss: 9.6343e-04 - accuracy: 1.0000 - val_loss: 0.5378 - val_accuracy: 0.8687\n",
            "Epoch 139/150\n",
            "12/12 [==============================] - 5s 440ms/step - loss: 0.0183 - accuracy: 0.9948 - val_loss: 0.5107 - val_accuracy: 0.8625\n",
            "Epoch 140/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 5.9575e-04 - accuracy: 1.0000 - val_loss: 0.4992 - val_accuracy: 0.8687\n",
            "Epoch 141/150\n",
            "12/12 [==============================] - 5s 442ms/step - loss: 0.0043 - accuracy: 1.0000 - val_loss: 0.5116 - val_accuracy: 0.8500\n",
            "Epoch 142/150\n",
            "12/12 [==============================] - 5s 444ms/step - loss: 0.0176 - accuracy: 0.9974 - val_loss: 0.4945 - val_accuracy: 0.8562\n",
            "Epoch 143/150\n",
            "12/12 [==============================] - 5s 438ms/step - loss: 1.9707e-04 - accuracy: 1.0000 - val_loss: 0.4793 - val_accuracy: 0.8500\n",
            "Epoch 144/150\n",
            "12/12 [==============================] - 5s 439ms/step - loss: 0.0012 - accuracy: 1.0000 - val_loss: 0.4673 - val_accuracy: 0.8500\n",
            "Epoch 145/150\n",
            "12/12 [==============================] - 6s 469ms/step - loss: 3.1469e-04 - accuracy: 1.0000 - val_loss: 0.4512 - val_accuracy: 0.8625\n",
            "Epoch 146/150\n",
            "12/12 [==============================] - 5s 456ms/step - loss: 0.0038 - accuracy: 1.0000 - val_loss: 0.4364 - val_accuracy: 0.8625\n",
            "Epoch 147/150\n",
            "12/12 [==============================] - 6s 458ms/step - loss: 4.8207e-04 - accuracy: 1.0000 - val_loss: 0.4219 - val_accuracy: 0.8687\n",
            "Epoch 148/150\n",
            "12/12 [==============================] - 5s 456ms/step - loss: 0.0045 - accuracy: 1.0000 - val_loss: 0.4205 - val_accuracy: 0.8813\n",
            "Epoch 149/150\n",
            "12/12 [==============================] - 6s 466ms/step - loss: 5.5519e-04 - accuracy: 1.0000 - val_loss: 0.4144 - val_accuracy: 0.8813\n",
            "Epoch 150/150\n",
            "12/12 [==============================] - 5s 437ms/step - loss: 5.2036e-04 - accuracy: 1.0000 - val_loss: 0.4127 - val_accuracy: 0.8813\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}