08_custom_dataset_ryazhskikh.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "08_custom_dataset_ryazhskikh.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [
        "a0anzotCI-II",
        "g5czXYi0WT3s"
      ],
      "toc_visible": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/darja/NeuralUni/blob/master/08_custom_dataset_ryazhskikh.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "metadata": {
        "id": "kLK-YQAPIfH_",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Классификация типов кораблей"
      ]
    },
    {
      "metadata": {
        "id": "hmvS5GnTYrTQ",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "Датасет состоит из изображений 400х400 со следующими типами кораблей:\n",
        "* Контейнеровоз (Cargo)\n",
        "* Судно береговой охраны (Coast Guard)\n",
        "* Паром (Ferry)\n",
        "* Рыболовецкое судно (Fishing)\n",
        "* Буксир (Tug)\n",
        "\n",
        "[Ссылка на датасет](https://drive.google.com/file/d/1zTKU96uN_P-YJhq3xZ8-uOP1uMe_nYcd/view?usp=sharing)"
      ]
    },
    {
      "metadata": {
        "id": "Wckh9qyvIk9C",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "from tensorflow.keras.applications.inception_v3 import InceptionV3\n",
        "\n",
        "from tensorflow.keras.preprocessing import image\n",
        "from tensorflow.keras.models import Model\n",
        "from tensorflow.keras.layers import Dense, GlobalAveragePooling2D, Dropout\n",
        "from tensorflow.python.keras.optimizers import Adam, SGD\n",
        "from tensorflow.python.keras.preprocessing.image import ImageDataGenerator\n",
        "from tensorflow.keras.callbacks import ModelCheckpoint, ReduceLROnPlateau, CSVLogger\n",
        "\n",
        "from google.colab import files\n",
        "import numpy as np\n",
        "import glob\n",
        "import imageio\n",
        "import datetime\n",
        "\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline \n",
        "\n",
        "from keras import backend as K"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "15NZ-K7lIocF",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Подготовка данных"
      ]
    },
    {
      "metadata": {
        "id": "a0anzotCI-II",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Загрузка данных с Google Drive"
      ]
    },
    {
      "metadata": {
        "id": "VLivDHF0HbRw",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 122
        },
        "outputId": "7f8acb0d-6a31-47fc-a675-af01fec1af72"
      },
      "cell_type": "code",
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/gdrive', force_remount=True)"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Go to this URL in a browser: https://accounts.google.com/o/oauth2/auth?client_id=947318989803-6bn6qk8qdgf4n4g3pfee6491hc0brc4i.apps.googleusercontent.com&redirect_uri=urn%3Aietf%3Awg%3Aoauth%3A2.0%3Aoob&scope=email%20https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fdocs.test%20https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fdrive%20https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fdrive.photos.readonly%20https%3A%2F%2Fwww.googleapis.com%2Fauth%2Fpeopleapi.readonly&response_type=code\n",
            "\n",
            "Enter your authorization code:\n",
            "··········\n",
            "Mounted at /content/gdrive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "fzVknhpwJIIs",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "!cp /content/gdrive/'My Drive'/neural/boatset.zip boatset.zip\n",
        "!unzip boatset.zip\n",
        "# !rm -rf __MACOSX\n",
        "# !rm -rf .DS_Store"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "OnPYXym_KyOD",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Входные параметры"
      ]
    },
    {
      "metadata": {
        "id": "v8YxrqlBK8qx",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# K.clear_session()\n",
        "\n",
        "# Пути к изображениям\n",
        "root_folder = \"boatset\"\n",
        "train_folder = root_folder + '/train'\n",
        "val_folder = root_folder + '/val'\n",
        "test_folder = root_folder + \"/test\"\n",
        "\n",
        "\n",
        "# Размеры изображений\n",
        "image_height, image_width = 400, 400\n",
        "target_size = (image_height, image_width)\n",
        "target_shape = (image_height, image_width, 3)\n",
        "\n",
        "# Классы изображений\n",
        "num_classes = 5\n",
        "classes = {0: \"Контейнеровоз\", 1: \"Береговая охрана\", 2: \"Паром\", 3: \"Рыболовецкое судно\", 4: \"Буксир\"}\n",
        "\n",
        "# Другое\n",
        "batch_size = 8\n",
        "lr = 2e-4\n",
        "optimizer = Adam(lr)\n",
        "# optimizer = RMSprop(lr)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "Dtk4swBjKhL_",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Генератор изображений"
      ]
    },
    {
      "metadata": {
        "id": "FhbWZicrKmLl",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        },
        "outputId": "782fecc7-eafe-4072-9987-3a5affce0577"
      },
      "cell_type": "code",
      "source": [
        "# Train generator\n",
        "train_datagen = ImageDataGenerator(rescale=1. / 255,\n",
        "                                  horizontal_flip=True,\n",
        "                                  fill_mode='nearest')\n",
        "\n",
        "train_generator = train_datagen.flow_from_directory(\n",
        "    train_folder,\n",
        "    target_size = target_size,\n",
        "    batch_size = batch_size,\n",
        "    class_mode = 'categorical')\n",
        "\n",
        "print(\"Train classes\", train_generator.class_indices)\n",
        "\n",
        "# Validation generator\n",
        "val_datagen = ImageDataGenerator(rescale=1. / 255)\n",
        "\n",
        "val_generator = val_datagen.flow_from_directory(\n",
        "    val_folder,\n",
        "    target_size = target_size,\n",
        "    batch_size = batch_size,\n",
        "    class_mode = 'categorical'\n",
        ")\n",
        "print(\"Val classes\", train_generator.class_indices)"
      ],
      "execution_count": 121,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Found 995 images belonging to 5 classes.\n",
            "Train classes {'cargo': 0, 'coast guard': 1, 'ferry': 2, 'fishing': 3, 'tug': 4}\n",
            "Found 160 images belonging to 5 classes.\n",
            "Val classes {'cargo': 0, 'coast guard': 1, 'ferry': 2, 'fishing': 3, 'tug': 4}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "EKPOV5p4NN2x",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Подготовка нейросети"
      ]
    },
    {
      "metadata": {
        "id": "Zk3a7qiNNU3k",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Создание нейросети"
      ]
    },
    {
      "metadata": {
        "id": "SwKdEwGdNdmY",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Нейросеть на основе Inception"
      ]
    },
    {
      "metadata": {
        "id": "MwKkufIZNXKA",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def create_model():\n",
        "    print(\"Loading Inception\")\n",
        "\n",
        "    base_model = InceptionV3(\n",
        "      weights='imagenet', \n",
        "      include_top=False, \n",
        "      input_shape=target_shape)\n",
        "\n",
        "    for layer in base_model.layers:\n",
        "      layer.trainable = False\n",
        "\n",
        "    print(\"Inception loaded\")\n",
        "\n",
        "    x = base_model.output\n",
        "    x = GlobalAveragePooling2D()(x)\n",
        "    x = Dense(1024, activation='relu')(x)\n",
        "    predictions = Dense(num_classes, activation='softmax')(x)\n",
        "\n",
        "    return Model(inputs=base_model.input, outputs=predictions)\n",
        "\n",
        "\n",
        "    return model\n",
        "\n",
        "def load_trained_model(weights_path):\n",
        "    model = create_model()\n",
        "    model.load_weights(weights_path)\n",
        "    return model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "qzK7kavBrGqR",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Callbacks"
      ]
    },
    {
      "metadata": {
        "id": "Q_M8UPKrrJ0c",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "4defdb20-2f62-43f8-fa61-831622ad3a66"
      },
      "cell_type": "code",
      "source": [
        "opt_string = optimizer\n",
        "if (type(optimizer) == Adam):\n",
        "    opt_string = \"adam_%.0E\" % lr\n",
        "elif (type(optimizer) == RMSprop):\n",
        "    opt_string = \"rmsprop_%.0E\" % lr\n",
        "\n",
        "dt = datetime.datetime.now().strftime('%Y%m%d-%H%M')\n",
        "log_file = 'training.%s.batch_%s.%s.log' % (opt_string, batch_size, dt)\n",
        "weights_file = 'shipnet.%s.batch_%s.%s.h5' % (opt_string, batch_size, dt)\n",
        "\n",
        "print(\"Logging to [%s], saving network to [%s]\" % (log_file, weights_file))\n",
        "    \n",
        "checkpoint = ModelCheckpoint(\n",
        "    weights_file,                           \n",
        "    monitor='val_acc',\n",
        "    save_best_only=True,\n",
        "    verbose = 1)\n",
        "\n",
        "learning_rate = ReduceLROnPlateau(\n",
        "    monitor='val_loss', \n",
        "    factor=0.5, \n",
        "    patience=5, \n",
        "    verbose=1, \n",
        "    mode='auto',\n",
        "    min_lr=1e-6)\n",
        "\n",
        "csv_logger = CSVLogger(log_file)\n",
        "\n",
        "callbacks = [checkpoint,learning_rate,csv_logger]"
      ],
      "execution_count": 123,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Logging to [training.adam_2E-04.batch_8.20190323-2306.log], saving network to [shipnet.adam_2E-04.batch_8.20190323-2306.h5]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "g1fRIppaNiAr",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Обучение нейросети"
      ]
    },
    {
      "metadata": {
        "id": "jIO91x2TNkvz",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 2951
        },
        "outputId": "8095530b-30ea-4223-c8bc-a3490089bd43"
      },
      "cell_type": "code",
      "source": [
        "model = create_model()\n",
        "\n",
        "model.compile(loss='categorical_crossentropy',\n",
        "            optimizer=optimizer, \n",
        "            metrics=['accuracy'])\n",
        "\n",
        "\n",
        "history = model.fit_generator(\n",
        "    train_generator,\n",
        "    steps_per_epoch = train_generator.samples // batch_size,\n",
        "    validation_data = val_generator, \n",
        "    validation_steps = val_generator.samples // batch_size,\n",
        "    epochs = 30,\n",
        "    verbose = 1,\n",
        "    callbacks = callbacks\n",
        ")\n",
        "\n",
        "plt.plot(history.history['acc'], label='Train')\n",
        "plt.plot(history.history['val_acc'], label='Validation')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": 125,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loading Inception\n",
            "Inception loaded\n",
            "Epoch 1/30\n",
            "20/20 [==============================] - 9s 449ms/step - loss: 0.9861 - acc: 0.6187\n",
            "\n",
            "Epoch 00001: val_acc improved from -inf to 0.61875, saving model to shipnet.adam_2E-04.batch_8.20190323-2306.h5\n",
            "125/125 [==============================] - 734s 6s/step - loss: 1.2094 - acc: 0.5276 - val_loss: 0.9861 - val_acc: 0.6187\n",
            "Epoch 2/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.8102 - acc: 0.7063\n",
            "\n",
            "Epoch 00002: val_acc improved from 0.61875 to 0.70625, saving model to shipnet.adam_2E-04.batch_8.20190323-2306.h5\n",
            "125/125 [==============================] - 29s 231ms/step - loss: 0.8349 - acc: 0.7025 - val_loss: 0.8102 - val_acc: 0.7063\n",
            "Epoch 3/30\n",
            "20/20 [==============================] - 4s 184ms/step - loss: 0.8294 - acc: 0.7250\n",
            "\n",
            "Epoch 00003: val_acc improved from 0.70625 to 0.72500, saving model to shipnet.adam_2E-04.batch_8.20190323-2306.h5\n",
            "125/125 [==============================] - 29s 228ms/step - loss: 0.6756 - acc: 0.7608 - val_loss: 0.8294 - val_acc: 0.7250\n",
            "Epoch 4/30\n",
            "20/20 [==============================] - 4s 186ms/step - loss: 0.7595 - acc: 0.7250\n",
            "\n",
            "Epoch 00004: val_acc did not improve from 0.72500\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.7283 - acc: 0.7296 - val_loss: 0.7595 - val_acc: 0.7250\n",
            "Epoch 5/30\n",
            "20/20 [==============================] - 4s 186ms/step - loss: 0.8097 - acc: 0.6375\n",
            "\n",
            "Epoch 00005: val_acc did not improve from 0.72500\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.7113 - acc: 0.7447 - val_loss: 0.8097 - val_acc: 0.6375\n",
            "Epoch 6/30\n",
            "20/20 [==============================] - 4s 185ms/step - loss: 0.7037 - acc: 0.7563\n",
            "\n",
            "Epoch 00006: val_acc improved from 0.72500 to 0.75625, saving model to shipnet.adam_2E-04.batch_8.20190323-2306.h5\n",
            "125/125 [==============================] - 29s 231ms/step - loss: 0.5965 - acc: 0.7849 - val_loss: 0.7037 - val_acc: 0.7563\n",
            "Epoch 7/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.6240 - acc: 0.7875\n",
            "\n",
            "Epoch 00007: val_acc improved from 0.75625 to 0.78750, saving model to shipnet.adam_2E-04.batch_8.20190323-2306.h5\n",
            "125/125 [==============================] - 29s 230ms/step - loss: 0.5801 - acc: 0.7849 - val_loss: 0.6240 - val_acc: 0.7875\n",
            "Epoch 8/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.6292 - acc: 0.7500\n",
            "\n",
            "Epoch 00008: val_acc did not improve from 0.78750\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.5411 - acc: 0.8050 - val_loss: 0.6292 - val_acc: 0.7500\n",
            "Epoch 9/30\n",
            "20/20 [==============================] - 4s 186ms/step - loss: 0.6781 - acc: 0.7000\n",
            "\n",
            "Epoch 00009: val_acc did not improve from 0.78750\n",
            "125/125 [==============================] - 27s 217ms/step - loss: 0.5654 - acc: 0.7950 - val_loss: 0.6781 - val_acc: 0.7000\n",
            "Epoch 10/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.6060 - acc: 0.7812\n",
            "\n",
            "Epoch 00010: val_acc did not improve from 0.78750\n",
            "125/125 [==============================] - 27s 217ms/step - loss: 0.5593 - acc: 0.7990 - val_loss: 0.6060 - val_acc: 0.7812\n",
            "Epoch 11/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.5721 - acc: 0.8000\n",
            "\n",
            "Epoch 00011: val_acc improved from 0.78750 to 0.80000, saving model to shipnet.adam_2E-04.batch_8.20190323-2306.h5\n",
            "125/125 [==============================] - 29s 231ms/step - loss: 0.5054 - acc: 0.8181 - val_loss: 0.5721 - val_acc: 0.8000\n",
            "Epoch 12/30\n",
            "20/20 [==============================] - 4s 186ms/step - loss: 0.6557 - acc: 0.7250\n",
            "\n",
            "Epoch 00012: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 217ms/step - loss: 0.5232 - acc: 0.8040 - val_loss: 0.6557 - val_acc: 0.7250\n",
            "Epoch 13/30\n",
            "20/20 [==============================] - 4s 186ms/step - loss: 0.5905 - acc: 0.7625\n",
            "\n",
            "Epoch 00013: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.4602 - acc: 0.8271 - val_loss: 0.5905 - val_acc: 0.7625\n",
            "Epoch 14/30\n",
            "20/20 [==============================] - 4s 185ms/step - loss: 0.6020 - acc: 0.7500\n",
            "\n",
            "Epoch 00014: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 215ms/step - loss: 0.4347 - acc: 0.8382 - val_loss: 0.6020 - val_acc: 0.7500\n",
            "Epoch 15/30\n",
            "20/20 [==============================] - 4s 185ms/step - loss: 0.6210 - acc: 0.7500\n",
            "\n",
            "Epoch 00015: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.4500 - acc: 0.8462 - val_loss: 0.6210 - val_acc: 0.7500\n",
            "Epoch 16/30\n",
            "20/20 [==============================] - 4s 186ms/step - loss: 0.6997 - acc: 0.7500\n",
            "\n",
            "Epoch 00016: val_acc did not improve from 0.80000\n",
            "\n",
            "Epoch 00016: ReduceLROnPlateau reducing learning rate to 9.999999747378752e-05.\n",
            "125/125 [==============================] - 105s 842ms/step - loss: 0.4496 - acc: 0.8342 - val_loss: 0.6997 - val_acc: 0.7500\n",
            "Epoch 17/30\n",
            "20/20 [==============================] - 4s 184ms/step - loss: 0.5964 - acc: 0.7563\n",
            "\n",
            "Epoch 00017: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 217ms/step - loss: 0.3764 - acc: 0.8513 - val_loss: 0.5964 - val_acc: 0.7563\n",
            "Epoch 18/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.5912 - acc: 0.7875\n",
            "\n",
            "Epoch 00018: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3568 - acc: 0.8714 - val_loss: 0.5912 - val_acc: 0.7875\n",
            "Epoch 19/30\n",
            "20/20 [==============================] - 4s 184ms/step - loss: 0.6178 - acc: 0.7500\n",
            "\n",
            "Epoch 00019: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3420 - acc: 0.8935 - val_loss: 0.6178 - val_acc: 0.7500\n",
            "Epoch 20/30\n",
            "20/20 [==============================] - 4s 185ms/step - loss: 0.6580 - acc: 0.7688\n",
            "\n",
            "Epoch 00020: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3715 - acc: 0.8653 - val_loss: 0.6580 - val_acc: 0.7688\n",
            "Epoch 21/30\n",
            "20/20 [==============================] - 4s 185ms/step - loss: 0.6130 - acc: 0.7625\n",
            "\n",
            "Epoch 00021: val_acc did not improve from 0.80000\n",
            "\n",
            "Epoch 00021: ReduceLROnPlateau reducing learning rate to 4.999999873689376e-05.\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3370 - acc: 0.8724 - val_loss: 0.6130 - val_acc: 0.7625\n",
            "Epoch 22/30\n",
            "20/20 [==============================] - 4s 188ms/step - loss: 0.6043 - acc: 0.7625\n",
            "\n",
            "Epoch 00022: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3232 - acc: 0.8864 - val_loss: 0.6043 - val_acc: 0.7625\n",
            "Epoch 23/30\n",
            "20/20 [==============================] - 4s 188ms/step - loss: 0.6234 - acc: 0.7500\n",
            "\n",
            "Epoch 00023: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3192 - acc: 0.8915 - val_loss: 0.6234 - val_acc: 0.7500\n",
            "Epoch 24/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.6213 - acc: 0.7563\n",
            "\n",
            "Epoch 00024: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3426 - acc: 0.8784 - val_loss: 0.6213 - val_acc: 0.7563\n",
            "Epoch 25/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.6141 - acc: 0.7812\n",
            "\n",
            "Epoch 00025: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3464 - acc: 0.8774 - val_loss: 0.6141 - val_acc: 0.7812\n",
            "Epoch 26/30\n",
            "20/20 [==============================] - 4s 186ms/step - loss: 0.5878 - acc: 0.7875\n",
            "\n",
            "Epoch 00026: val_acc did not improve from 0.80000\n",
            "\n",
            "Epoch 00026: ReduceLROnPlateau reducing learning rate to 2.499999936844688e-05.\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3269 - acc: 0.8945 - val_loss: 0.5878 - val_acc: 0.7875\n",
            "Epoch 27/30\n",
            "20/20 [==============================] - 4s 185ms/step - loss: 0.6357 - acc: 0.7625\n",
            "\n",
            "Epoch 00027: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 216ms/step - loss: 0.3279 - acc: 0.8965 - val_loss: 0.6357 - val_acc: 0.7625\n",
            "Epoch 28/30\n",
            "20/20 [==============================] - 4s 186ms/step - loss: 0.6136 - acc: 0.7500\n",
            "\n",
            "Epoch 00028: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 217ms/step - loss: 0.3115 - acc: 0.8975 - val_loss: 0.6136 - val_acc: 0.7500\n",
            "Epoch 29/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.6157 - acc: 0.7625\n",
            "\n",
            "Epoch 00029: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 217ms/step - loss: 0.2924 - acc: 0.9005 - val_loss: 0.6157 - val_acc: 0.7625\n",
            "Epoch 30/30\n",
            "20/20 [==============================] - 4s 187ms/step - loss: 0.6462 - acc: 0.7437\n",
            "\n",
            "Epoch 00030: val_acc did not improve from 0.80000\n",
            "125/125 [==============================] - 27s 217ms/step - loss: 0.3030 - acc: 0.8955 - val_loss: 0.6462 - val_acc: 0.7437\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXgAAAD4CAYAAADmWv3KAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "gpu3F7-hOFAO",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Демонстрация распознавания"
      ]
    },
    {
      "metadata": {
        "id": "g5czXYi0WT3s",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Утилиты"
      ]
    },
    {
      "metadata": {
        "id": "pDzvLP9TOIAX",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def evaluate_folder(folder):\n",
        "  datagen = ImageDataGenerator(rescale=1./255)\n",
        "  generator = datagen.flow_from_directory(\n",
        "    folder,\n",
        "    target_size = target_size,\n",
        "    batch_size = batch_size,\n",
        "    shuffle = False,\n",
        "    class_mode = 'categorical')\n",
        "  generator.reset()\n",
        "  print(\"Train classes\", generator.class_indices)\n",
        "  return model.evaluate_generator(\n",
        "      generator, \n",
        "      steps = generator.samples // batch_size,\n",
        "      verbose = 1)\n",
        "\n",
        "def read_image(filename):\n",
        "    return imageio.imread(filename)\n",
        "\n",
        "def predict_single_image(content):\n",
        "    x = content\n",
        "    x = x / 255\n",
        "    x = np.expand_dims(x, axis=0)\n",
        "    \n",
        "    prediction = model.predict(x)\n",
        "    predicted = np.argmax(prediction, axis=-1)\n",
        "    return predicted"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "iywJ1pqB-Yyi",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Загрузка наилучшей версии сети"
      ]
    },
    {
      "metadata": {
        "id": "_8Co72Vf-ds2",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 68
        },
        "outputId": "5b3296ed-dcc1-4365-d8b4-58bb791e27a4"
      },
      "cell_type": "code",
      "source": [
        "print(\"Loading model from %s\" % weights_file)\n",
        "model = load_trained_model(weights_file)\n",
        "model.compile(loss='categorical_crossentropy',\n",
        "            optimizer=optimizer, \n",
        "            metrics=['accuracy'])"
      ],
      "execution_count": 128,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loading model from shipnet.adam_2E-04.batch_8.20190323-2306.h5\n",
            "Loading Inception\n",
            "Inception loaded\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "LvNllnQPQBXg",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Распознавание размеченной выборки"
      ]
    },
    {
      "metadata": {
        "id": "c4aH665FOLwq",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "outputId": "d304a39c-9c02-42b9-fea6-46042860194e"
      },
      "cell_type": "code",
      "source": [
        "print(\"---- Test ----\\n\")\n",
        "train_eval = evaluate_folder(train_folder)\n",
        "\n",
        "print(\"\\n---- Validation ----\\n\")\n",
        "val_eval = evaluate_folder(val_folder)\n"
      ],
      "execution_count": 129,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "---- Test ----\n",
            "\n",
            "Found 995 images belonging to 5 classes.\n",
            "Train classes {'cargo': 0, 'coast guard': 1, 'ferry': 2, 'fishing': 3, 'tug': 4}\n",
            "125/125 [==============================] - 27s 218ms/step - loss: 0.5254 - acc: 0.8101\n",
            "\n",
            "---- Validation ----\n",
            "\n",
            "Found 160 images belonging to 5 classes.\n",
            "Train classes {'cargo': 0, 'coast_guard': 1, 'ferry': 2, 'fishing': 3, 'tug': 4}\n",
            "20/20 [==============================] - 4s 181ms/step - loss: 0.5721 - acc: 0.8000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "bMh-1NvvQFfk",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Демонстрация"
      ]
    },
    {
      "metadata": {
        "id": "ZoHOR6_lQRZu",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 442
        },
        "outputId": "34ffebac-95fd-44c3-c08d-e0acc68b483c"
      },
      "cell_type": "code",
      "source": [
        "test_files = glob.glob(test_folder + \"/*.jpg\")\n",
        "test_files.sort()\n",
        "\n",
        "for filename in test_files:\n",
        "    content = read_image(filename)\n",
        "    cl = predict_single_image(content)[0]\n",
        "    print(\"%s -> %s\" % (filename, classes[cl]))\n"
      ],
      "execution_count": 130,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "boatset/test/cargo_test_1.jpg -> Контейнеровоз\n",
            "boatset/test/cargo_test_2.jpg -> Контейнеровоз\n",
            "boatset/test/cargo_test_3.jpg -> Контейнеровоз\n",
            "boatset/test/cargo_test_4.jpg -> Контейнеровоз\n",
            "boatset/test/cargo_test_5.jpg -> Контейнеровоз\n",
            "boatset/test/coast_guard_test_1.jpg -> Береговая охрана\n",
            "boatset/test/coast_guard_test_2.jpg -> Буксир\n",
            "boatset/test/coast_guard_test_3.jpg -> Береговая охрана\n",
            "boatset/test/coast_guard_test_4.jpg -> Береговая охрана\n",
            "boatset/test/coast_guard_test_5.jpg -> Буксир\n",
            "boatset/test/ferry_test_1.jpg -> Паром\n",
            "boatset/test/ferry_test_2.jpg -> Паром\n",
            "boatset/test/ferry_test_3.jpg -> Паром\n",
            "boatset/test/ferry_test_4.jpg -> Паром\n",
            "boatset/test/ferry_test_5.jpg -> Паром\n",
            "boatset/test/fishing_test_1.jpg -> Рыболовецкое судно\n",
            "boatset/test/fishing_test_2.jpg -> Рыболовецкое судно\n",
            "boatset/test/fishing_test_3.jpg -> Рыболовецкое судно\n",
            "boatset/test/fishing_test_4.jpg -> Буксир\n",
            "boatset/test/fishing_test_5.jpg -> Береговая охрана\n",
            "boatset/test/tug_test_1.jpg -> Буксир\n",
            "boatset/test/tug_test_2.jpg -> Буксир\n",
            "boatset/test/tug_test_3.jpg -> Буксир\n",
            "boatset/test/tug_test_4.jpg -> Буксир\n",
            "boatset/test/tug_test_5.jpg -> Буксир\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}