cnn_segmentation_connected_components_1.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "cnn-segmentation-connected-components.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "metadata": {
        "id": "sliy7m7L-M-D",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Approach\n",
        "\n",
        "* Firstly a convolutional neural network is used to segment the image, using the bounding boxes directly as a mask. \n",
        "* Secondly connected components is used to separate multiple nodules.\n",
        "* Finally a bounding box is simply drawn around every connected component.\n",
        "\n",
        "# Network\n",
        "\n",
        "* The network consists of a number of residual blocks with convolutions and downsampling blocks with max pooling.\n",
        "* At the end of the network a single upsampling layer converts the output to the same shape as the input.\n",
        "\n",
        "As the input to the network is 256 by 256 (instead of the original 1024 by 1024) and the network downsamples a number of times without any meaningful upsampling (the final upsampling is just to match in 256 by 256 mask) the final prediction is very crude. If the network downsamples 4 times the final bounding boxes can only change with at least 16 pixels."
      ]
    },
    {
      "metadata": {
        "id": "xjzfSdjV-Y3y",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# install dependencies not included by Colab\n",
        "# use pip3 to ensure compatibility w/ Google Deep Learning Images \n",
        "!pip3 install -q pydicom \n",
        "!pip3 install -q tqdm \n",
        "!pip3 install -q imgaug \n",
        "!pip3 install -q kaggle"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "pu1pks8K-M-G",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "f0e9fcf6-02d9-4a7e-fd0c-51f342cb3dc6"
      },
      "cell_type": "code",
      "source": [
        "import os\n",
        "import csv\n",
        "import random\n",
        "import pydicom\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "from skimage import measure\n",
        "from skimage.transform import resize\n",
        "from keras import layers\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "\n",
        "from matplotlib import pyplot as plt"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "metadata": {
        "id": "ZFNwjWWc_JgL",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "from google.colab import auth\n",
        "from googleapiclient.http import MediaFileUpload\n",
        "from googleapiclient.discovery import build\n",
        "\n",
        "auth.authenticate_user()\n",
        "\n",
        "drive_service = build('drive', 'v3')\n",
        "\n",
        "def save_file_to_drive(name, path):\n",
        "  file_metadata = {\n",
        "      'name': name,\n",
        "      'mimeType': 'application/octet-stream'\n",
        "     }\n",
        "\n",
        "  media = MediaFileUpload(path, \n",
        "                    mimetype='application/octet-stream',\n",
        "                    resumable=True)\n",
        "\n",
        "  created = drive_service.files().create(body=file_metadata,\n",
        "                                   media_body=media,\n",
        "                                   fields='id').execute()\n",
        "\n",
        "  print('File ID: {}'.format(created.get('id')))\n",
        "\n",
        "  return created"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "JUFKH6Pjn4WJ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "3d2c1b42-9fa2-44ac-b7a9-d7c97895bad0"
      },
      "cell_type": "code",
      "source": [
        "# upload checkpoint to GCS\n",
        "project_id = 'mammography-198911'\n",
        "bucket_name = 'pneumonia'\n",
        "\n",
        "!gcloud config set project {project_id}"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Updated property [core/project].\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "ejrE2hQw-fMG",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# enter your Kaggle credentionals here\n",
        "os.environ['KAGGLE_USERNAME']=\"skooch\"\n",
        "os.environ['KAGGLE_KEY']=\"42f8a02ee92cc773d1dbe66565673ad3\""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "BJskmI-Y-fwY",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# Root directory of the project\n",
        "ROOT_DIR = os.path.abspath('./')\n",
        "\n",
        "# Directory to save logs and trained model\n",
        "MODEL_DIR = os.path.join(ROOT_DIR, 'logs')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "NonqWakT-jTC",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# If you are unable to download the competition dataset, check to see if you have \n",
        "# accepted the user agreement on the competition website. \n",
        "if not os.path.exists(os.path.join(ROOT_DIR, \"stage_1_detailed_class_info.csv.zip\")):\n",
        "  !kaggle competitions download -c rsna-pneumonia-detection-challenge"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "YHvOAY1M-pzE",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# unzipping takes a few minutes\n",
        "if not os.path.exists(os.path.join(ROOT_DIR, \"stage_1_train_labels.csv\")):\n",
        "  !unzip -q -o stage_1_test_images.zip -d stage_1_test_images\n",
        "  !unzip -q -o stage_1_train_images.zip -d stage_1_train_images\n",
        "  !unzip -q -o stage_1_train_labels.csv.zip"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "EQG4ghkH-M-e",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Load nodule locations\n",
        "\n",
        "Table contains [filename : nodule location] pairs per row. \n",
        "* If a filename contains multiple nodules, the table contains multiple rows with the same filename but different nodule locations. \n",
        "* If a filename contains no nodules it contains a single row with an empty nodule location.\n",
        "\n",
        "The code below loads the table and transforms it into a dictionary. \n",
        "* The dictionary uses the filename as key and a list of nodule locations in that filename as value. \n",
        "* If a filename is not present in the dictionary it means that it contains no nodules."
      ]
    },
    {
      "metadata": {
        "id": "67McqwYt-t5W",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "6d1d509e-e279-4f2f-fe1d-c2ea9a4b8d84"
      },
      "cell_type": "code",
      "source": [
        "train_dicom_dir = os.path.join(ROOT_DIR, 'stage_1_train_images')\n",
        "test_dicom_dir = os.path.join(ROOT_DIR, 'stage_1_test_images')\n",
        "print(\"Train dir:\", train_dicom_dir)"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train dir: /content/stage_1_train_images\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "HIYblxy8-M-f",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# empty dictionary\n",
        "nodule_locations = {}\n",
        "# load table\n",
        "with open('./stage_1_train_labels.csv', mode='r') as infile:\n",
        "    # open reader\n",
        "    reader = csv.reader(infile)\n",
        "    # skip header\n",
        "    next(reader, None)\n",
        "    # loop through rows\n",
        "    for rows in reader:\n",
        "        # retrieve information\n",
        "        filename = rows[0]\n",
        "        location = rows[1:5]\n",
        "        nodule = rows[5]\n",
        "        # if row contains a nodule add label to dictionary\n",
        "        # which contains a list of nodule locations per filename\n",
        "        if nodule == '1':\n",
        "            # convert string to float to int\n",
        "            location = [int(float(i)) for i in location]\n",
        "            # save nodule location in dictionary\n",
        "            if filename in nodule_locations:\n",
        "                nodule_locations[filename].append(location)\n",
        "            else:\n",
        "                nodule_locations[filename] = [location]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "guL6FY9N-M-m",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Load filenames"
      ]
    },
    {
      "metadata": {
        "id": "ebHtuAkP-M-s",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "7587e39f-23f9-4a92-fb17-5b71b69f68c2"
      },
      "cell_type": "code",
      "source": [
        "# load and shuffle filenames\n",
        "folder = train_dicom_dir\n",
        "filenames = os.listdir(folder)\n",
        "random.shuffle(filenames)\n",
        "# split into train and validation filenames\n",
        "n_valid_samples = 2560\n",
        "train_filenames = filenames[n_valid_samples:]\n",
        "valid_filenames = filenames[:n_valid_samples]\n",
        "print('n train samples', len(train_filenames))\n",
        "print('n valid samples', len(valid_filenames))\n",
        "n_train_samples = len(filenames) - n_valid_samples"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "n train samples 23124\n",
            "n valid samples 2560\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "XRTP_RT6-M-7",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        " # Data generator\n",
        "\n",
        "The dataset is too large to fit into memory, so we need to create a generator that loads data on the fly.\n",
        "\n",
        "* The generator takes in some filenames, batch_size and other parameters.\n",
        "\n",
        "* The generator outputs a random batch of numpy images and numpy masks.\n",
        "    "
      ]
    },
    {
      "metadata": {
        "id": "mNHOs4vXGdwJ",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "BATCH_SIZE = 12\n",
        "IMAGE_SIZE = 448\n",
        "CHECKPOINT_PATH = \"model1_416.h5\""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "OrKvegIrmjsc",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "ac4d9f2d-7f4e-4d23-8f0e-98359c60fa54"
      },
      "cell_type": "code",
      "source": [
        "# try to download the checkpoint, if it doesn't exist pass\n",
        "try:\n",
        "  !gsutil cp gs://{bucket_name}/{CHECKPOINT_PATH}  ./\n",
        "except:\n",
        "  pass"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CommandException: No URLs matched: gs://pneumonia/model1_416.h5\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "aD51VT0I-M-9",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "class generator(keras.utils.Sequence):\n",
        "    \n",
        "    def __init__(self, folder, filenames, nodule_locations=None, batch_size=BATCH_SIZE, image_size=IMAGE_SIZE, shuffle=True, predict=False):\n",
        "        self.folder = folder\n",
        "        self.filenames = filenames\n",
        "        self.nodule_locations = nodule_locations\n",
        "        self.batch_size = batch_size\n",
        "        self.image_size = image_size\n",
        "        self.shuffle = shuffle\n",
        "        self.predict = predict\n",
        "        self.on_epoch_end()\n",
        "        \n",
        "    def __load__(self, filename):\n",
        "        # load dicom file as numpy array\n",
        "        img = pydicom.dcmread(os.path.join(self.folder, filename)).pixel_array\n",
        "        # create empty mask\n",
        "        msk = np.zeros(img.shape)\n",
        "        # get filename without extension\n",
        "        filename = filename.split('.')[0]\n",
        "        # if image contains nodules\n",
        "        if filename in nodule_locations:\n",
        "            # loop through nodules\n",
        "            for location in nodule_locations[filename]:\n",
        "                # add 1's at the location of the nodule\n",
        "                x, y, w, h = location\n",
        "                msk[y:y+h, x:x+w] = 1\n",
        "        # resize both image and mask\n",
        "        img = resize(img, (self.image_size, self.image_size), mode='reflect')\n",
        "        msk = resize(msk, (self.image_size, self.image_size), mode='reflect') > 0.5\n",
        "        # add trailing channel dimension\n",
        "        img = np.expand_dims(img, -1)\n",
        "        msk = np.expand_dims(msk, -1)\n",
        "        return img, msk\n",
        "    \n",
        "    def __loadpredict__(self, filename):\n",
        "        # load dicom file as numpy array\n",
        "        img = pydicom.dcmread(os.path.join(self.folder, filename)).pixel_array\n",
        "        # resize image\n",
        "        img = resize(img, (self.image_size, self.image_size), mode='reflect')\n",
        "        # add trailing channel dimension\n",
        "        img = np.expand_dims(img, -1)\n",
        "        return img\n",
        "        \n",
        "    def __getitem__(self, index):\n",
        "        # select batch\n",
        "        filenames = self.filenames[index*self.batch_size:(index+1)*self.batch_size]\n",
        "        # predict mode: return images and filenames\n",
        "        if self.predict:\n",
        "            # load files\n",
        "            imgs = [self.__loadpredict__(filename) for filename in filenames]\n",
        "            # create numpy batch\n",
        "            imgs = np.array(imgs)\n",
        "            return imgs, filenames\n",
        "        # train mode: return images and masks\n",
        "        else:\n",
        "            # load files\n",
        "            items = [self.__load__(filename) for filename in filenames]\n",
        "            # unzip images and masks\n",
        "            imgs, msks = zip(*items)\n",
        "            # create numpy batch\n",
        "            imgs = np.array(imgs)\n",
        "            msks = np.array(msks)\n",
        "            return imgs, msks\n",
        "        \n",
        "    def on_epoch_end(self):\n",
        "        if self.shuffle:\n",
        "            random.shuffle(self.filenames)\n",
        "        \n",
        "    def __len__(self):\n",
        "        if self.predict:\n",
        "            # return everything\n",
        "            return int(np.ceil(len(self.filenames) / self.batch_size))\n",
        "        else:\n",
        "            # return full batches only\n",
        "            return int(len(self.filenames) / self.batch_size)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "UjoLi1xYiOEp",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# upload last checkpoint\n",
        "# from google.colab import files\n",
        "\n",
        "# uploaded = files.upload()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "RbHHFLzb-M_F",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Network"
      ]
    },
    {
      "metadata": {
        "id": "rYHu1v6V-M_R",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def create_downsample(channels, inputs):\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(inputs)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "    x = keras.layers.Conv2D(channels, 1, padding='same', use_bias=False)(x)\n",
        "\n",
        "    x = keras.layers.MaxPool2D(2)(x)\n",
        "    return x\n",
        "\n",
        "def create_resblock(channels, inputs):\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(inputs)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "    x = keras.layers.Conv2D(channels, 3, padding='same', use_bias=False)(x)\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(x)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "    x = keras.layers.Conv2D(channels, 3, padding='same', use_bias=False)(x)\n",
        "    \n",
        "    try:\n",
        "      res_layer = keras.layers.add([x, inputs])\n",
        "    except:\n",
        "      res_layer = x\n",
        "      \n",
        "    return res_layer\n",
        "\n",
        "def create_network(input_size, channels, n_blocks=2, depth=4):\n",
        "    # input\n",
        "    inputs = keras.Input(shape=(input_size, input_size, 1))\n",
        "    \n",
        "    # initial layers before downsampling\n",
        "    x_1 = keras.layers.Conv2D(channels, 3, strides=(2,2), padding='same', use_bias=False)(inputs)\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(x_1)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "    \n",
        "    x = keras.layers.Conv2D(channels, 3, padding='same', use_bias=False)(x)\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(x)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "    \n",
        "    x = keras.layers.Conv2D(channels, 3, padding='same', use_bias=False)(x)\n",
        "    x = keras.layers.add([x, x_1])\n",
        "    \n",
        "    # residual blocks\n",
        "    for d in range(depth):\n",
        "        channels = channels * 2\n",
        "        if channels > 512:\n",
        "          channels = 512\n",
        "        \n",
        "        x = create_downsample(channels, x)\n",
        "        for b in range(n_blocks):\n",
        "            x = create_resblock(channels, x)\n",
        "            \n",
        "    # output\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(x)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "    \n",
        "    # downsample one last time\n",
        "    x = keras.layers.Conv2D(512, (3,3), strides=(1,1), dilation_rate=(2,2), padding='same', activation='relu')(x)\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(x)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "    \n",
        "    # reduce number of feature maps\n",
        "    x = keras.layers.Conv2D(256, (1,1), padding='same', activation='relu')(x)\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(x)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "\n",
        "    #     x = keras.layers.Dropout(0.125)(x)\n",
        "    \n",
        "    # deconv to 4x size\n",
        "    x = keras.layers.Conv2DTranspose(64, (8,8), (4,4), padding=\"same\", activation=\"relu\")(x)\n",
        "    x = keras.layers.BatchNormalization(momentum=0.9)(x)\n",
        "    x = keras.layers.LeakyReLU(0)(x)\n",
        "    \n",
        "    # upsample\n",
        "    x = keras.layers.Conv2D(1, 1, activation='sigmoid')(x)\n",
        "    \n",
        "    # 1 transpose conv to get back to original size\n",
        "    outputs = keras.layers.UpSampling2D(2**(depth - 1))(x)\n",
        "    model = keras.Model(inputs=inputs, outputs=outputs)\n",
        "    return model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "wsBfvg2Y-M_e",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Train network\n"
      ]
    },
    {
      "metadata": {
        "id": "qkxSFJ9w-M_f",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 3502
        },
        "outputId": "9cdef511-ed1a-4e92-f68c-95f1132ab144"
      },
      "cell_type": "code",
      "source": [
        "# define iou or jaccard loss function\n",
        "def iou_loss(y_true, y_pred):\n",
        "    y_true = tf.reshape(y_true, [-1])\n",
        "    y_pred = tf.reshape(y_pred, [-1])\n",
        "    intersection = tf.reduce_sum(y_true * y_pred)\n",
        "    score = (intersection + 1.) / (tf.reduce_sum(y_true) + tf.reduce_sum(y_pred) - intersection + 1.)\n",
        "    return 1 - score\n",
        "\n",
        "# combine bce loss and iou loss\n",
        "def iou_bce_loss(y_true, y_pred):\n",
        "    return 0.3 * keras.losses.binary_crossentropy(y_true, y_pred) + 0.7 * iou_loss(y_true, y_pred)\n",
        "\n",
        "# mean iou as a metric\n",
        "def mean_iou(y_true, y_pred):\n",
        "    y_pred = tf.round(y_pred)\n",
        "    intersect = tf.reduce_sum(y_true * y_pred, axis=[1, 2, 3])\n",
        "    union = tf.reduce_sum(y_true, axis=[1, 2, 3]) + tf.reduce_sum(y_pred, axis=[1, 2, 3])\n",
        "    smooth = tf.ones(tf.shape(intersect))\n",
        "    return tf.reduce_mean((intersect + smooth) / (union - intersect + smooth))\n",
        "\n",
        "# create network and compiler\n",
        "model = create_network(input_size=IMAGE_SIZE, channels=32, n_blocks=2, depth=4)\n",
        "print(model.summary())\n",
        "\n",
        "model.compile(optimizer='adam',\n",
        "              loss=iou_bce_loss,\n",
        "              metrics=['accuracy', mean_iou])\n",
        "\n",
        "# cosine learning rate annealing\n",
        "def cosine_annealing(x):\n",
        "    lr = 0.005\n",
        "    epochs = 25\n",
        "    return lr*(np.cos(np.pi*x/epochs)+1.)/2\n",
        "  \n",
        "learning_rate = tf.keras.callbacks.LearningRateScheduler(cosine_annealing)\n",
        "checkpoint = keras.callbacks.ModelCheckpoint(CHECKPOINT_PATH, monitor='val_loss', verbose=1, save_best_only=False, save_weights_only=False, mode='auto', period=1)\n",
        "\n",
        "\n",
        "# create train and validation generators\n",
        "folder = train_dicom_dir\n",
        "train_gen = generator(folder, train_filenames, nodule_locations, batch_size=BATCH_SIZE, image_size=IMAGE_SIZE, shuffle=True, predict=False)\n",
        "valid_gen = generator(folder, valid_filenames, nodule_locations, batch_size=BATCH_SIZE, image_size=IMAGE_SIZE, shuffle=False, predict=False)\n"
      ],
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_3 (InputLayer)            (None, 448, 448, 1)  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_52 (Conv2D)              (None, 224, 224, 32) 288         input_3[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_52 (BatchNo (None, 224, 224, 32) 128         conv2d_52[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_52 (LeakyReLU)      (None, 224, 224, 32) 0           batch_normalization_52[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_53 (Conv2D)              (None, 224, 224, 32) 9216        leaky_re_lu_52[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_53 (BatchNo (None, 224, 224, 32) 128         conv2d_53[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_53 (LeakyReLU)      (None, 224, 224, 32) 0           batch_normalization_53[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_54 (Conv2D)              (None, 224, 224, 32) 9216        leaky_re_lu_53[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_18 (Add)                    (None, 224, 224, 32) 0           conv2d_54[0][0]                  \n",
            "                                                                 conv2d_52[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_54 (BatchNo (None, 224, 224, 32) 128         add_18[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_54 (LeakyReLU)      (None, 224, 224, 32) 0           batch_normalization_54[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_55 (Conv2D)              (None, 224, 224, 64) 2048        leaky_re_lu_54[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "max_pooling2d_8 (MaxPooling2D)  (None, 112, 112, 64) 0           conv2d_55[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_55 (BatchNo (None, 112, 112, 64) 256         max_pooling2d_8[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_55 (LeakyReLU)      (None, 112, 112, 64) 0           batch_normalization_55[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_56 (Conv2D)              (None, 112, 112, 64) 36864       leaky_re_lu_55[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_56 (BatchNo (None, 112, 112, 64) 256         conv2d_56[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_56 (LeakyReLU)      (None, 112, 112, 64) 0           batch_normalization_56[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_57 (Conv2D)              (None, 112, 112, 64) 36864       leaky_re_lu_56[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_19 (Add)                    (None, 112, 112, 64) 0           conv2d_57[0][0]                  \n",
            "                                                                 max_pooling2d_8[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_57 (BatchNo (None, 112, 112, 64) 256         add_19[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_57 (LeakyReLU)      (None, 112, 112, 64) 0           batch_normalization_57[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_58 (Conv2D)              (None, 112, 112, 64) 36864       leaky_re_lu_57[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_58 (BatchNo (None, 112, 112, 64) 256         conv2d_58[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_58 (LeakyReLU)      (None, 112, 112, 64) 0           batch_normalization_58[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_59 (Conv2D)              (None, 112, 112, 64) 36864       leaky_re_lu_58[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_20 (Add)                    (None, 112, 112, 64) 0           conv2d_59[0][0]                  \n",
            "                                                                 add_19[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_59 (BatchNo (None, 112, 112, 64) 256         add_20[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_59 (LeakyReLU)      (None, 112, 112, 64) 0           batch_normalization_59[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_60 (Conv2D)              (None, 112, 112, 128 8192        leaky_re_lu_59[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "max_pooling2d_9 (MaxPooling2D)  (None, 56, 56, 128)  0           conv2d_60[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_60 (BatchNo (None, 56, 56, 128)  512         max_pooling2d_9[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_60 (LeakyReLU)      (None, 56, 56, 128)  0           batch_normalization_60[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_61 (Conv2D)              (None, 56, 56, 128)  147456      leaky_re_lu_60[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_61 (BatchNo (None, 56, 56, 128)  512         conv2d_61[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_61 (LeakyReLU)      (None, 56, 56, 128)  0           batch_normalization_61[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_62 (Conv2D)              (None, 56, 56, 128)  147456      leaky_re_lu_61[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_21 (Add)                    (None, 56, 56, 128)  0           conv2d_62[0][0]                  \n",
            "                                                                 max_pooling2d_9[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_62 (BatchNo (None, 56, 56, 128)  512         add_21[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_62 (LeakyReLU)      (None, 56, 56, 128)  0           batch_normalization_62[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_63 (Conv2D)              (None, 56, 56, 128)  147456      leaky_re_lu_62[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_63 (BatchNo (None, 56, 56, 128)  512         conv2d_63[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_63 (LeakyReLU)      (None, 56, 56, 128)  0           batch_normalization_63[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_64 (Conv2D)              (None, 56, 56, 128)  147456      leaky_re_lu_63[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_22 (Add)                    (None, 56, 56, 128)  0           conv2d_64[0][0]                  \n",
            "                                                                 add_21[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_64 (BatchNo (None, 56, 56, 128)  512         add_22[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_64 (LeakyReLU)      (None, 56, 56, 128)  0           batch_normalization_64[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_65 (Conv2D)              (None, 56, 56, 256)  32768       leaky_re_lu_64[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "max_pooling2d_10 (MaxPooling2D) (None, 28, 28, 256)  0           conv2d_65[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_65 (BatchNo (None, 28, 28, 256)  1024        max_pooling2d_10[0][0]           \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_65 (LeakyReLU)      (None, 28, 28, 256)  0           batch_normalization_65[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_66 (Conv2D)              (None, 28, 28, 256)  589824      leaky_re_lu_65[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_66 (BatchNo (None, 28, 28, 256)  1024        conv2d_66[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_66 (LeakyReLU)      (None, 28, 28, 256)  0           batch_normalization_66[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_67 (Conv2D)              (None, 28, 28, 256)  589824      leaky_re_lu_66[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_23 (Add)                    (None, 28, 28, 256)  0           conv2d_67[0][0]                  \n",
            "                                                                 max_pooling2d_10[0][0]           \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_67 (BatchNo (None, 28, 28, 256)  1024        add_23[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_67 (LeakyReLU)      (None, 28, 28, 256)  0           batch_normalization_67[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_68 (Conv2D)              (None, 28, 28, 256)  589824      leaky_re_lu_67[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_68 (BatchNo (None, 28, 28, 256)  1024        conv2d_68[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_68 (LeakyReLU)      (None, 28, 28, 256)  0           batch_normalization_68[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_69 (Conv2D)              (None, 28, 28, 256)  589824      leaky_re_lu_68[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_24 (Add)                    (None, 28, 28, 256)  0           conv2d_69[0][0]                  \n",
            "                                                                 add_23[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_69 (BatchNo (None, 28, 28, 256)  1024        add_24[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_69 (LeakyReLU)      (None, 28, 28, 256)  0           batch_normalization_69[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_70 (Conv2D)              (None, 28, 28, 512)  131072      leaky_re_lu_69[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "max_pooling2d_11 (MaxPooling2D) (None, 14, 14, 512)  0           conv2d_70[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_70 (BatchNo (None, 14, 14, 512)  2048        max_pooling2d_11[0][0]           \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_70 (LeakyReLU)      (None, 14, 14, 512)  0           batch_normalization_70[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_71 (Conv2D)              (None, 14, 14, 512)  2359296     leaky_re_lu_70[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_71 (BatchNo (None, 14, 14, 512)  2048        conv2d_71[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_71 (LeakyReLU)      (None, 14, 14, 512)  0           batch_normalization_71[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_72 (Conv2D)              (None, 14, 14, 512)  2359296     leaky_re_lu_71[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_25 (Add)                    (None, 14, 14, 512)  0           conv2d_72[0][0]                  \n",
            "                                                                 max_pooling2d_11[0][0]           \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_72 (BatchNo (None, 14, 14, 512)  2048        add_25[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_72 (LeakyReLU)      (None, 14, 14, 512)  0           batch_normalization_72[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_73 (Conv2D)              (None, 14, 14, 512)  2359296     leaky_re_lu_72[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_73 (BatchNo (None, 14, 14, 512)  2048        conv2d_73[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_73 (LeakyReLU)      (None, 14, 14, 512)  0           batch_normalization_73[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_74 (Conv2D)              (None, 14, 14, 512)  2359296     leaky_re_lu_73[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "add_26 (Add)                    (None, 14, 14, 512)  0           conv2d_74[0][0]                  \n",
            "                                                                 add_25[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_74 (BatchNo (None, 14, 14, 512)  2048        add_26[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_74 (LeakyReLU)      (None, 14, 14, 512)  0           batch_normalization_74[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_75 (Conv2D)              (None, 14, 14, 512)  2359808     leaky_re_lu_74[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_75 (BatchNo (None, 14, 14, 512)  2048        conv2d_75[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_75 (LeakyReLU)      (None, 14, 14, 512)  0           batch_normalization_75[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_76 (Conv2D)              (None, 14, 14, 256)  131328      leaky_re_lu_75[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_76 (BatchNo (None, 14, 14, 256)  1024        conv2d_76[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_76 (LeakyReLU)      (None, 14, 14, 256)  0           batch_normalization_76[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_transpose_2 (Conv2DTrans (None, 56, 56, 64)   1048640     leaky_re_lu_76[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "batch_normalization_77 (BatchNo (None, 56, 56, 64)   256         conv2d_transpose_2[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "leaky_re_lu_77 (LeakyReLU)      (None, 56, 56, 64)   0           batch_normalization_77[0][0]     \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_77 (Conv2D)              (None, 56, 56, 1)    65          leaky_re_lu_77[0][0]             \n",
            "__________________________________________________________________________________________________\n",
            "up_sampling2d_2 (UpSampling2D)  (None, 448, 448, 1)  0           conv2d_77[0][0]                  \n",
            "==================================================================================================\n",
            "Total params: 16,289,313\n",
            "Trainable params: 16,277,857\n",
            "Non-trainable params: 11,456\n",
            "__________________________________________________________________________________________________\n",
            "None\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "KKSw1v5GE5bX",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# try:\n",
        "#   model.load_weights(CHECKPOINT_PATH)\n",
        "# except:\n",
        "#   pass\n",
        "model.load_weights(CHECKPOINT_PATH)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "A3spdZG5cBsF",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 561
        },
        "outputId": "0ff325fb-4c47-4027-f16a-d91883c2e0cb"
      },
      "cell_type": "code",
      "source": [
        "history = model.fit_generator(train_gen, validation_data=valid_gen, callbacks=[learning_rate, checkpoint], epochs=8, shuffle=True, verbose=1)"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/8\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.6417 - acc: 0.9567 - mean_iou: 0.5544\n",
            "Epoch 00001: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1446s 750ms/step - loss: 0.6418 - acc: 0.9567 - mean_iou: 0.5543 - val_loss: 1.7417 - val_acc: 0.5257 - val_mean_iou: 0.0449\n",
            "Epoch 2/8\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.6051 - acc: 0.9608 - mean_iou: 0.6262\n",
            "Epoch 00002: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1432s 743ms/step - loss: 0.6051 - acc: 0.9608 - mean_iou: 0.6262 - val_loss: 0.7385 - val_acc: 0.9743 - val_mean_iou: 0.7829\n",
            "Epoch 3/8\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5948 - acc: 0.9626 - mean_iou: 0.6341\n",
            "Epoch 00003: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1418s 736ms/step - loss: 0.5948 - acc: 0.9626 - mean_iou: 0.6341 - val_loss: 0.6402 - val_acc: 0.9461 - val_mean_iou: 0.5584\n",
            "Epoch 4/8\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5798 - acc: 0.9646 - mean_iou: 0.6542\n",
            "Epoch 00004: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1423s 739ms/step - loss: 0.5797 - acc: 0.9646 - mean_iou: 0.6543 - val_loss: 0.6776 - val_acc: 0.9749 - val_mean_iou: 0.7476\n",
            "Epoch 5/8\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5739 - acc: 0.9655 - mean_iou: 0.6517\n",
            "Epoch 00005: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1424s 739ms/step - loss: 0.5738 - acc: 0.9655 - mean_iou: 0.6518 - val_loss: 0.6082 - val_acc: 0.9728 - val_mean_iou: 0.7043\n",
            "Epoch 6/8\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5676 - acc: 0.9663 - mean_iou: 0.6542\n",
            "Epoch 00006: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1404s 729ms/step - loss: 0.5675 - acc: 0.9663 - mean_iou: 0.6542 - val_loss: 0.5760 - val_acc: 0.9631 - val_mean_iou: 0.6328\n",
            "Epoch 7/8\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5600 - acc: 0.9670 - mean_iou: 0.6647\n",
            "Epoch 00007: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1418s 736ms/step - loss: 0.5601 - acc: 0.9670 - mean_iou: 0.6647 - val_loss: 0.6711 - val_acc: 0.9750 - val_mean_iou: 0.7383\n",
            "Epoch 8/8\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5580 - acc: 0.9679 - mean_iou: 0.6659\n",
            "Epoch 00008: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1431s 743ms/step - loss: 0.5580 - acc: 0.9679 - mean_iou: 0.6660 - val_loss: 0.5731 - val_acc: 0.9688 - val_mean_iou: 0.6461\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "5sg5Qj3JW1QI",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 357
        },
        "outputId": "3efa4544-40ac-4848-cb8b-71776bcff0cd"
      },
      "cell_type": "code",
      "source": [
        "history = model.fit_generator(train_gen, validation_data=valid_gen, callbacks=[learning_rate, checkpoint], epochs=5, shuffle=True, verbose=1)"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/5\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5534 - acc: 0.9680 - mean_iou: 0.6665\n",
            "Epoch 00001: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1433s 744ms/step - loss: 0.5535 - acc: 0.9680 - mean_iou: 0.6665 - val_loss: 0.6023 - val_acc: 0.9740 - val_mean_iou: 0.7278\n",
            "Epoch 2/5\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5495 - acc: 0.9684 - mean_iou: 0.6727\n",
            "Epoch 00002: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1414s 734ms/step - loss: 0.5496 - acc: 0.9684 - mean_iou: 0.6727 - val_loss: 0.5716 - val_acc: 0.9713 - val_mean_iou: 0.6977\n",
            "Epoch 3/5\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5473 - acc: 0.9689 - mean_iou: 0.6718\n",
            "Epoch 00003: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1422s 738ms/step - loss: 0.5473 - acc: 0.9689 - mean_iou: 0.6718 - val_loss: 0.5813 - val_acc: 0.9645 - val_mean_iou: 0.6721\n",
            "Epoch 4/5\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5414 - acc: 0.9699 - mean_iou: 0.6771\n",
            "Epoch 00004: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1420s 737ms/step - loss: 0.5414 - acc: 0.9699 - mean_iou: 0.6771 - val_loss: 0.5651 - val_acc: 0.9653 - val_mean_iou: 0.6643\n",
            "Epoch 5/5\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5408 - acc: 0.9698 - mean_iou: 0.6758\n",
            "Epoch 00005: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1414s 734ms/step - loss: 0.5407 - acc: 0.9698 - mean_iou: 0.6759 - val_loss: 0.5631 - val_acc: 0.9719 - val_mean_iou: 0.6905\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "CO9WGS29Nbnu",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 255
        },
        "outputId": "406f172a-83a4-4524-a608-9a39df35393f"
      },
      "cell_type": "code",
      "source": [
        "history = model.fit_generator(train_gen, validation_data=valid_gen, callbacks=[learning_rate, checkpoint], epochs=5, shuffle=True, verbose=1)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/5\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5382 - acc: 0.9704 - mean_iou: 0.6761\n",
            "Epoch 00001: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1392s 722ms/step - loss: 0.5381 - acc: 0.9704 - mean_iou: 0.6762 - val_loss: 0.5593 - val_acc: 0.9700 - val_mean_iou: 0.6839\n",
            "Epoch 2/5\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5398 - acc: 0.9701 - mean_iou: 0.6775\n",
            "Epoch 00002: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1390s 721ms/step - loss: 0.5399 - acc: 0.9701 - mean_iou: 0.6775 - val_loss: 0.6207 - val_acc: 0.9478 - val_mean_iou: 0.5260\n",
            "Epoch 3/5\n",
            "1926/1927 [============================>.] - ETA: 0s - loss: 0.5316 - acc: 0.9710 - mean_iou: 0.6849\n",
            "Epoch 00003: saving model to model1_416.h5\n",
            "1927/1927 [==============================] - 1391s 722ms/step - loss: 0.5317 - acc: 0.9710 - mean_iou: 0.6848 - val_loss: 0.6025 - val_acc: 0.9748 - val_mean_iou: 0.7145\n",
            "Epoch 4/5\n",
            "1061/1927 [===============>..............] - ETA: 9:28 - loss: 0.5281 - acc: 0.9714 - mean_iou: 0.6876"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "y0oa3cNPcNI5",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# history = model.fit_generator(train_gen, validation_data=valid_gen, callbacks=[learning_rate, checkpoint], epochs=3, shuffle=True, verbose=1)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "yMDSkSHm-M_r",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 266
        },
        "outputId": "11ac092c-3a11-445d-bf9c-d3799b0a77f2"
      },
      "cell_type": "code",
      "source": [
        "plt.figure(figsize=(12,4))\n",
        "plt.subplot(131)\n",
        "plt.plot(history.epoch, history.history[\"loss\"], label=\"Train loss\")\n",
        "plt.plot(history.epoch, history.history[\"val_loss\"], label=\"Valid loss\")\n",
        "plt.legend()\n",
        "plt.subplot(132)\n",
        "plt.plot(history.epoch, history.history[\"acc\"], label=\"Train accuracy\")\n",
        "plt.plot(history.epoch, history.history[\"val_acc\"], label=\"Valid accuracy\")\n",
        "plt.legend()\n",
        "plt.subplot(133)\n",
        "plt.plot(history.epoch, history.history[\"mean_iou\"], label=\"Train iou\")\n",
        "plt.plot(history.epoch, history.history[\"val_mean_iou\"], label=\"Valid iou\")\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAsYAAAD5CAYAAAAk2/PkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIQQA/z9JVDuzfEkR+gAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f354972a208>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "dz9at3ZH-NAE",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Predict test images"
      ]
    },
    {
      "metadata": {
        "id": "cCMcFaVN-NAF",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "375d971c-9edb-4a09-cdbf-bf952eaa9104"
      },
      "cell_type": "code",
      "source": [
        "# load and shuffle filenames\n",
        "folder = test_dicom_dir\n",
        "test_filenames = os.listdir(folder)\n",
        "print('n test samples:', len(test_filenames))\n",
        "\n",
        "# create test generator with predict flag set to True\n",
        "test_gen = generator(folder, test_filenames, None, batch_size=25, image_size=IMAGE_SIZE, shuffle=False, predict=True)\n",
        "\n",
        "# create submission dictionary\n",
        "submission_dict = {}\n",
        "# loop through testset\n",
        "for imgs, filenames in test_gen:\n",
        "    # predict batch of images\n",
        "    preds = model.predict(imgs)\n",
        "    # loop through batch\n",
        "    for pred, filename in zip(preds, filenames):\n",
        "        # resize predicted mask\n",
        "        pred = resize(pred, (1024, 1024), mode='reflect')\n",
        "        # threshold predicted mask\n",
        "        comp = pred[:, :, 0] > 0.5\n",
        "        # apply connected components\n",
        "        comp = measure.label(comp)\n",
        "        # apply bounding boxes\n",
        "        predictionString = ''\n",
        "        for region in measure.regionprops(comp):\n",
        "            # retrieve x, y, height and width\n",
        "            y, x, y2, x2 = region.bbox\n",
        "            height = y2 - y\n",
        "            width = x2 - x\n",
        "            # proxy for confidence score\n",
        "            conf = np.mean(pred[y:y+height, x:x+width])\n",
        "            # add to predictionString\n",
        "            predictionString += str(conf) + ' ' + str(x) + ' ' + str(y) + ' ' + str(width) + ' ' + str(height) + ' '\n",
        "        # add filename and predictionString to dictionary\n",
        "        filename = filename.split('.')[0]\n",
        "        submission_dict[filename] = predictionString\n",
        "    # stop if we've got them all\n",
        "    if len(submission_dict) >= len(test_filenames):\n",
        "        break\n",
        "\n",
        "print(\"Done predicting!\")        \n",
        "\n",
        "# save dictionary as csv file\n",
        "sub = pd.DataFrame.from_dict(submission_dict,orient='index')\n",
        "sub.index.names = ['patientId']\n",
        "sub.columns = ['PredictionString']\n",
        "sub.to_csv('submission.csv')"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "n test samples: 1000\n",
            "Done predicting!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "whgCcttS_SUa",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "f259565d-10f6-4906-ebda-95b4fbde6d51"
      },
      "cell_type": "code",
      "source": [
        "!kaggle competitions submit -c rsna-pneumonia-detection-challenge -f submission.csv -m \"Segmentation from colab size 448x448 13 epochs\""
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Successfully submitted to RSNA Pneumonia Detection Challenge"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "D0UDLKjA_TZ7",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "8134ed68-a130-4161-eb33-441932849f16"
      },
      "cell_type": "code",
      "source": [
        "save_file_to_drive(\"submission.csv\", \"submission.csv\")"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "File ID: 1ymS6qEowqIOHolFhYl1T4kHhkTFAeeOB\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'id': '1ymS6qEowqIOHolFhYl1T4kHhkTFAeeOB'}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 33
        }
      ]
    },
    {
      "metadata": {
        "id": "20Yb5s_0yRUS",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "f8e4ffcf-7f32-41be-f265-637589e1d552"
      },
      "cell_type": "code",
      "source": [
        "save_file_to_drive(CHECKPOINT_PATH, CHECKPOINT_PATH)"
      ],
      "execution_count": 34,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "File ID: 1R_sjQ2vbgqKaMApl8efJuZ7bm46iLDAH\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'id': '1R_sjQ2vbgqKaMApl8efJuZ7bm46iLDAH'}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 34
        }
      ]
    },
    {
      "metadata": {
        "id": "D1wetIsJdTZz",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 272
        },
        "outputId": "141254fe-9ae3-4acf-c335-0b790694f7dc"
      },
      "cell_type": "code",
      "source": [
        "!gsutil cp ./{CHECKPOINT_PATH} gs://{bucket_name}/"
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Copying file://./model1_416.h5 [Content-Type=application/octet-stream]...\n",
            "/ [0 files][    0.0 B/186.7 MiB]                                                \r==> NOTE: You are uploading one or more large file(s), which would run\n",
            "significantly faster if you enable parallel composite uploads. This\n",
            "feature can be enabled by editing the\n",
            "\"parallel_composite_upload_threshold\" value in your .boto\n",
            "configuration file. However, note that if you do this large files will\n",
            "be uploaded as `composite objects\n",
            "<https://cloud.google.com/storage/docs/composite-objects>`_,which\n",
            "means that any user who downloads such objects will need to have a\n",
            "compiled crcmod installed (see \"gsutil help crcmod\"). This is because\n",
            "without a compiled crcmod, computing checksums on composite objects is\n",
            "so slow that gsutil disables downloads of composite objects.\n",
            "\n",
            "-\n",
            "Operation completed over 1 objects/186.7 MiB.                                    \n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}