unet_train.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [],
   "source": [
    "!pip install -U segmentation-models --quiet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/opt/conda/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n",
      "Using TensorFlow backend.\n",
      "/opt/conda/lib/python3.6/site-packages/classification_models/resnext/__init__.py:4: UserWarning: Current ResNext models are deprecated, use keras.applications ResNeXt models\n",
      "  warnings.warn('Current ResNext models are deprecated, '\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import os\n",
    "import json\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from tqdm import tqdm_notebook\n",
    "import albumentations as albu\n",
    "from skimage.exposure import adjust_gamma\n",
    "import cv2\n",
    "import keras\n",
    "from keras import backend as K\n",
    "from keras.models import Model\n",
    "from keras.layers import Input\n",
    "from keras.layers.convolutional import Conv2D, Conv2DTranspose\n",
    "from keras.layers.pooling import MaxPooling2D\n",
    "from keras.layers.merge import concatenate\n",
    "from keras.losses import binary_crossentropy\n",
    "from keras.optimizers import Adam\n",
    "from keras.callbacks import Callback, ModelCheckpoint, EarlyStopping\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from tqdm import tqdm\n",
    "from sklearn.model_selection import train_test_split\n",
    "import segmentation_models as sm\n",
    "from segmentation_models.utils import set_trainable"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_dir = '../input/understanding_cloud_organization/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(22184, 5)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Image_Label</th>\n",
       "      <th>EncodedPixels</th>\n",
       "      <th>ImageId</th>\n",
       "      <th>ClassId</th>\n",
       "      <th>hasMask</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0011165.jpg_Fish</td>\n",
       "      <td>264918 937 266318 937 267718 937 269118 937 27...</td>\n",
       "      <td>0011165.jpg</td>\n",
       "      <td>Fish</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0011165.jpg_Flower</td>\n",
       "      <td>1355565 1002 1356965 1002 1358365 1002 1359765...</td>\n",
       "      <td>0011165.jpg</td>\n",
       "      <td>Flower</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0011165.jpg_Gravel</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0011165.jpg</td>\n",
       "      <td>Gravel</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0011165.jpg_Sugar</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0011165.jpg</td>\n",
       "      <td>Sugar</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>002be4f.jpg_Fish</td>\n",
       "      <td>233813 878 235213 878 236613 878 238010 881 23...</td>\n",
       "      <td>002be4f.jpg</td>\n",
       "      <td>Fish</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          Image_Label                                      EncodedPixels  \\\n",
       "0    0011165.jpg_Fish  264918 937 266318 937 267718 937 269118 937 27...   \n",
       "1  0011165.jpg_Flower  1355565 1002 1356965 1002 1358365 1002 1359765...   \n",
       "2  0011165.jpg_Gravel                                                NaN   \n",
       "3   0011165.jpg_Sugar                                                NaN   \n",
       "4    002be4f.jpg_Fish  233813 878 235213 878 236613 878 238010 881 23...   \n",
       "\n",
       "       ImageId ClassId  hasMask  \n",
       "0  0011165.jpg    Fish     True  \n",
       "1  0011165.jpg  Flower     True  \n",
       "2  0011165.jpg  Gravel    False  \n",
       "3  0011165.jpg   Sugar    False  \n",
       "4  002be4f.jpg    Fish     True  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df = pd.read_csv(f'{data_dir}train.csv')\n",
    "train_df['ImageId'] = train_df['Image_Label'].apply(lambda x: x.split('_')[0])\n",
    "train_df['ClassId'] = train_df['Image_Label'].apply(lambda x: x.split('_')[1])\n",
    "train_df['hasMask'] = ~ train_df['EncodedPixels'].isna()\n",
    "\n",
    "print(train_df.shape)\n",
    "train_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(5546, 2)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>ImageId</th>\n",
       "      <th>hasMask</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>821</th>\n",
       "      <td>24dd99c.jpg</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1885</th>\n",
       "      <td>562f80a.jpg</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3260</th>\n",
       "      <td>944f10b.jpg</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1872</th>\n",
       "      <td>55b539f.jpg</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4464</th>\n",
       "      <td>cdf7242.jpg</td>\n",
       "      <td>4.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          ImageId  hasMask\n",
       "821   24dd99c.jpg      4.0\n",
       "1885  562f80a.jpg      4.0\n",
       "3260  944f10b.jpg      4.0\n",
       "1872  55b539f.jpg      4.0\n",
       "4464  cdf7242.jpg      4.0"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mask_count_df = train_df.groupby('ImageId').agg(np.sum).reset_index()\n",
    "mask_count_df.sort_values('hasMask', ascending=False, inplace=True)\n",
    "print(mask_count_df.shape)\n",
    "mask_count_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "sub_df = pd.read_csv(f'{data_dir}/sample_submission.csv')\n",
    "sub_df['ImageId'] = sub_df['Image_Label'].apply(lambda x: x.split('_')[0])\n",
    "test_imgs = pd.DataFrame(sub_df['ImageId'].unique(), columns=['ImageId'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def np_resize(img, input_shape):\n",
    "    \"\"\"\n",
    "    Reshape a numpy array, which is input_shape=(height, width), \n",
    "    as opposed to input_shape=(width, height) for cv2\n",
    "    \"\"\"\n",
    "    height, width = input_shape\n",
    "    return cv2.resize(img, (width, height))\n",
    "    \n",
    "def mask2rle(img):\n",
    "    '''\n",
    "    img: numpy array, 1 - mask, 0 - background\n",
    "    Returns run length as string formated\n",
    "    '''\n",
    "    pixels= img.T.flatten()\n",
    "    pixels = np.concatenate([[0], pixels, [0]])\n",
    "    runs = np.where(pixels[1:] != pixels[:-1])[0] + 1\n",
    "    runs[1::2] -= runs[::2]\n",
    "    return ' '.join(str(x) for x in runs)\n",
    "\n",
    "def rle2mask(rle, input_shape):\n",
    "    width, height = input_shape[:2]\n",
    "    \n",
    "    mask= np.zeros( width*height ).astype(np.uint8)\n",
    "    \n",
    "    array = np.asarray([int(x) for x in rle.split()])\n",
    "    starts = array[0::2]\n",
    "    lengths = array[1::2]\n",
    "\n",
    "    current_position = 0\n",
    "    for index, start in enumerate(starts):\n",
    "        mask[int(start):int(start+lengths[index])] = 1\n",
    "        current_position += lengths[index]\n",
    "        \n",
    "    return mask.reshape(height, width).T\n",
    "\n",
    "def build_masks(rles, input_shape, reshape=None):\n",
    "    depth = len(rles)\n",
    "    if reshape is None:\n",
    "        masks = np.zeros((*input_shape, depth))\n",
    "    else:\n",
    "        masks = np.zeros((*reshape, depth))\n",
    "    \n",
    "    for i, rle in enumerate(rles):\n",
    "        if type(rle) is str:\n",
    "            if reshape is None:\n",
    "                masks[:, :, i] = rle2mask(rle, input_shape)\n",
    "            else:\n",
    "                mask = rle2mask(rle, input_shape)\n",
    "                reshaped_mask = np_resize(mask, reshape)\n",
    "                masks[:, :, i] = reshaped_mask\n",
    "    \n",
    "    return masks\n",
    "\n",
    "def build_rles(masks, reshape=None):\n",
    "    width, height, depth = masks.shape\n",
    "    \n",
    "    rles = []\n",
    "    \n",
    "    for i in range(depth):\n",
    "        mask = masks[:, :, i]\n",
    "        \n",
    "        if reshape:\n",
    "            mask = mask.astype(np.float32)\n",
    "            mask = np_resize(mask, reshape).astype(np.int64)\n",
    "        \n",
    "        rle = mask2rle(mask)\n",
    "        rles.append(rle)\n",
    "        \n",
    "    return rles"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "class RAdam(keras.optimizers.Optimizer):\n",
    "    \"\"\"RAdam optimizer.\n",
    "    # Arguments\n",
    "        lr: float >= 0. Learning rate.\n",
    "        beta_1: float, 0 < beta < 1. Generally close to 1.\n",
    "        beta_2: float, 0 < beta < 1. Generally close to 1.\n",
    "        epsilon: float >= 0. Fuzz factor. If `None`, defaults to `K.epsilon()`.\n",
    "        decay: float >= 0. Learning rate decay over each update.\n",
    "        weight_decay: float >= 0. Weight decay for each param.\n",
    "        amsgrad: boolean. Whether to apply the AMSGrad variant of this\n",
    "            algorithm from the paper \"On the Convergence of Adam and\n",
    "            Beyond\".\n",
    "        total_steps: int >= 0. Total number of training steps. Enable warmup by setting a positive value.\n",
    "        warmup_proportion: 0 < warmup_proportion < 1. The proportion of increasing steps.\n",
    "        min_lr: float >= 0. Minimum learning rate after warmup.\n",
    "    # References\n",
    "        - [Adam - A Method for Stochastic Optimization](https://arxiv.org/abs/1412.6980v8)\n",
    "        - [On the Convergence of Adam and Beyond](https://openreview.net/forum?id=ryQu7f-RZ)\n",
    "        - [On The Variance Of The Adaptive Learning Rate And Beyond](https://arxiv.org/pdf/1908.03265v1.pdf)\n",
    "    \"\"\"\n",
    "\n",
    "    def __init__(self, lr=0.001, beta_1=0.9, beta_2=0.999,\n",
    "                 epsilon=None, decay=0., weight_decay=0., amsgrad=False,\n",
    "                 total_steps=0, warmup_proportion=0.1, min_lr=0., **kwargs):\n",
    "        super(RAdam, self).__init__(**kwargs)\n",
    "        with K.name_scope(self.__class__.__name__):\n",
    "            self.iterations = K.variable(0, dtype='int64', name='iterations')\n",
    "            self.lr = K.variable(lr, name='lr')\n",
    "            self.beta_1 = K.variable(beta_1, name='beta_1')\n",
    "            self.beta_2 = K.variable(beta_2, name='beta_2')\n",
    "            self.decay = K.variable(decay, name='decay')\n",
    "            self.weight_decay = K.variable(weight_decay, name='weight_decay')\n",
    "            self.total_steps = K.variable(total_steps, name='total_steps')\n",
    "            self.warmup_proportion = K.variable(warmup_proportion, name='warmup_proportion')\n",
    "            self.min_lr = K.variable(min_lr, name='min_lr')\n",
    "        if epsilon is None:\n",
    "            epsilon = K.epsilon()\n",
    "        self.epsilon = epsilon\n",
    "        self.initial_decay = decay\n",
    "        self.initial_weight_decay = weight_decay\n",
    "        self.initial_total_steps = total_steps\n",
    "        self.amsgrad = amsgrad\n",
    "\n",
    "    def get_updates(self, loss, params):\n",
    "        grads = self.get_gradients(loss, params)\n",
    "        self.updates = [K.update_add(self.iterations, 1)]\n",
    "\n",
    "        lr = self.lr\n",
    "\n",
    "        if self.initial_decay > 0:\n",
    "            lr = lr * (1. / (1. + self.decay * K.cast(self.iterations, K.dtype(self.decay))))\n",
    "\n",
    "        t = K.cast(self.iterations, K.floatx()) + 1\n",
    "\n",
    "        if self.initial_total_steps > 0:\n",
    "            warmup_steps = self.total_steps * self.warmup_proportion\n",
    "            lr = K.switch(\n",
    "                t <= warmup_steps,\n",
    "                lr * (t / warmup_steps),\n",
    "                self.min_lr + (lr - self.min_lr) * (1.0 - K.minimum(t, self.total_steps) / self.total_steps),\n",
    "            )\n",
    "\n",
    "        ms = [K.zeros(K.int_shape(p), dtype=K.dtype(p), name='m_' + str(i)) for (i, p) in enumerate(params)]\n",
    "        vs = [K.zeros(K.int_shape(p), dtype=K.dtype(p), name='v_' + str(i)) for (i, p) in enumerate(params)]\n",
    "\n",
    "        if self.amsgrad:\n",
    "            vhats = [K.zeros(K.int_shape(p), dtype=K.dtype(p), name='vhat_' + str(i)) for (i, p) in enumerate(params)]\n",
    "        else:\n",
    "            vhats = [K.zeros(1, name='vhat_' + str(i)) for i in range(len(params))]\n",
    "\n",
    "        self.weights = [self.iterations] + ms + vs + vhats\n",
    "\n",
    "        beta_1_t = K.pow(self.beta_1, t)\n",
    "        beta_2_t = K.pow(self.beta_2, t)\n",
    "\n",
    "        sma_inf = 2.0 / (1.0 - self.beta_2) - 1.0\n",
    "        sma_t = sma_inf - 2.0 * t * beta_2_t / (1.0 - beta_2_t)\n",
    "\n",
    "        for p, g, m, v, vhat in zip(params, grads, ms, vs, vhats):\n",
    "            m_t = (self.beta_1 * m) + (1. - self.beta_1) * g\n",
    "            v_t = (self.beta_2 * v) + (1. - self.beta_2) * K.square(g)\n",
    "\n",
    "            m_corr_t = m_t / (1.0 - beta_1_t)\n",
    "            if self.amsgrad:\n",
    "                vhat_t = K.maximum(vhat, v_t)\n",
    "                v_corr_t = K.sqrt(vhat_t / (1.0 - beta_2_t) + self.epsilon)\n",
    "                self.updates.append(K.update(vhat, vhat_t))\n",
    "            else:\n",
    "                v_corr_t = K.sqrt(v_t / (1.0 - beta_2_t) + self.epsilon)\n",
    "\n",
    "            r_t = K.sqrt((sma_t - 4.0) / (sma_inf - 4.0) *\n",
    "                         (sma_t - 2.0) / (sma_inf - 2.0) *\n",
    "                         sma_inf / sma_t)\n",
    "\n",
    "            p_t = K.switch(sma_t >= 5, r_t * m_corr_t / v_corr_t, m_corr_t)\n",
    "\n",
    "            if self.initial_weight_decay > 0:\n",
    "                p_t += self.weight_decay * p\n",
    "\n",
    "            p_t = p - lr * p_t\n",
    "\n",
    "            self.updates.append(K.update(m, m_t))\n",
    "            self.updates.append(K.update(v, v_t))\n",
    "            new_p = p_t\n",
    "\n",
    "            # Apply constraints.\n",
    "            if getattr(p, 'constraint', None) is not None:\n",
    "                new_p = p.constraint(new_p)\n",
    "\n",
    "            self.updates.append(K.update(p, new_p))\n",
    "        return self.updates\n",
    "\n",
    "    def get_config(self):\n",
    "        config = {\n",
    "            'lr': float(K.get_value(self.lr)),\n",
    "            'beta_1': float(K.get_value(self.beta_1)),\n",
    "            'beta_2': float(K.get_value(self.beta_2)),\n",
    "            'decay': float(K.get_value(self.decay)),\n",
    "            'weight_decay': float(K.get_value(self.weight_decay)),\n",
    "            'epsilon': self.epsilon,\n",
    "            'amsgrad': self.amsgrad,\n",
    "            'total_steps': float(K.get_value(self.total_steps)),\n",
    "            'warmup_proportion': float(K.get_value(self.warmup_proportion)),\n",
    "            'min_lr': float(K.get_value(self.min_lr)),\n",
    "        }\n",
    "        base_config = super(RAdam, self).get_config()\n",
    "        return dict(list(base_config.items()) + list(config.items()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "class DataGenerator(keras.utils.Sequence):\n",
    "    'Generates data for Keras'\n",
    "    def __init__(self, list_IDs, df, target_df=None, mode='fit',\n",
    "                 base_path=f'{data_dir}train_images',\n",
    "                 batch_size=32, dim=(1400, 2100), n_channels=3, reshape=None,\n",
    "                 augment=False, n_classes=4, random_state=42, shuffle=True, gamma=None):\n",
    "        self.dim = dim\n",
    "        self.batch_size = batch_size\n",
    "        self.df = df\n",
    "        self.mode = mode\n",
    "        self.base_path = base_path\n",
    "        self.target_df = target_df\n",
    "        self.list_IDs = list_IDs\n",
    "        self.reshape = reshape\n",
    "        self.n_channels = n_channels\n",
    "        self.augment = augment\n",
    "        self.n_classes = n_classes\n",
    "        self.shuffle = shuffle\n",
    "        self.random_state = random_state\n",
    "        self.gamma = gamma\n",
    "        self.on_epoch_end()\n",
    "        np.random.seed(self.random_state)\n",
    "\n",
    "    def __len__(self):\n",
    "        'Denotes the number of batches per epoch'\n",
    "        return int(np.floor(len(self.list_IDs) / self.batch_size))\n",
    "\n",
    "    def __getitem__(self, index):\n",
    "        'Generate one batch of data'\n",
    "        # Generate indexes of the batch\n",
    "        indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]\n",
    "\n",
    "        # Find list of IDs\n",
    "        list_IDs_batch = [self.list_IDs[k] for k in indexes]\n",
    "        \n",
    "        X = self.__generate_X(list_IDs_batch)\n",
    "        \n",
    "        if self.mode == 'fit':\n",
    "            y = self.__generate_y(list_IDs_batch)\n",
    "            \n",
    "            if self.augment:\n",
    "                X, y = self.__augment_batch(X, y)\n",
    "            \n",
    "            return X, y\n",
    "        \n",
    "        elif self.mode == 'predict':\n",
    "            return X\n",
    "\n",
    "        else:\n",
    "            raise AttributeError('The mode parameter should be set to \"fit\" or \"predict\".')\n",
    "        \n",
    "    def on_epoch_end(self):\n",
    "        'Updates indexes after each epoch'\n",
    "        self.indexes = np.arange(len(self.list_IDs))\n",
    "        if self.shuffle == True:\n",
    "            np.random.seed(self.random_state)\n",
    "            np.random.shuffle(self.indexes)\n",
    "    \n",
    "    def __generate_X(self, list_IDs_batch):\n",
    "        'Generates data containing batch_size samples'\n",
    "        # Initialization\n",
    "        if self.reshape is None:\n",
    "            X = np.empty((self.batch_size, *self.dim, self.n_channels))\n",
    "        else:\n",
    "            X = np.empty((self.batch_size, *self.reshape, self.n_channels))\n",
    "        \n",
    "        # Generate data\n",
    "        for i, ID in enumerate(list_IDs_batch):\n",
    "            im_name = self.df['ImageId'].iloc[ID]\n",
    "            img_path = f\"{self.base_path}/{im_name}\"\n",
    "            img = self.__load_rgb(img_path)\n",
    "            \n",
    "            if self.reshape is not None:\n",
    "                img = np_resize(img, self.reshape)\n",
    "            \n",
    "            if self.gamma is not None:\n",
    "                img = adjust_gamma(img, gamma=self.gamma)\n",
    "            # Store samples\n",
    "            X[i,] = img\n",
    "\n",
    "        return X\n",
    "    \n",
    "    def __generate_y(self, list_IDs_batch):\n",
    "        if self.reshape is None:\n",
    "            y = np.empty((self.batch_size, *self.dim, self.n_classes), dtype=int)\n",
    "        else:\n",
    "            y = np.empty((self.batch_size, *self.reshape, self.n_classes), dtype=int)\n",
    "        \n",
    "        for i, ID in enumerate(list_IDs_batch):\n",
    "            im_name = self.df['ImageId'].iloc[ID]\n",
    "            image_df = self.target_df[self.target_df['ImageId'] == im_name]\n",
    "            \n",
    "            rles = image_df['EncodedPixels'].values\n",
    "            \n",
    "            if self.reshape is not None:\n",
    "                masks = build_masks(rles, input_shape=self.dim, reshape=self.reshape)\n",
    "            else:\n",
    "                masks = build_masks(rles, input_shape=self.dim)\n",
    "            \n",
    "            y[i, ] = masks\n",
    "\n",
    "        return y\n",
    "    \n",
    "    def __load_grayscale(self, img_path):\n",
    "        img = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)\n",
    "        img = img.astype(np.float32) / 255.\n",
    "        img = np.expand_dims(img, axis=-1)\n",
    "\n",
    "        return img\n",
    "    \n",
    "    def __load_rgb(self, img_path):\n",
    "        img = cv2.imread(img_path)\n",
    "        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)\n",
    "        img = img.astype(np.float32) / 255.\n",
    "\n",
    "        return img\n",
    "    \n",
    "    \n",
    "    def __random_transform(self, img, masks):\n",
    "        composition = albu.Compose([\n",
    "            albu.HorizontalFlip(),\n",
    "            albu.VerticalFlip(),\n",
    "            albu.ShiftScaleRotate(rotate_limit=30, shift_limit=0.1)\n",
    "        ])\n",
    "        \n",
    "        composed = composition(image=img, mask=masks)\n",
    "        aug_img = composed['image']\n",
    "        aug_masks = composed['mask']\n",
    "        \n",
    "        return aug_img, aug_masks\n",
    "    \n",
    "    def __augment_batch(self, img_batch, masks_batch):\n",
    "        for i in range(img_batch.shape[0]):\n",
    "            img_batch[i, ], masks_batch[i, ] = self.__random_transform(\n",
    "                img_batch[i, ], masks_batch[i, ])\n",
    "        \n",
    "        return img_batch, masks_batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def dice_coef(y_true, y_pred, smooth=1):\n",
    "    y_true_f = K.flatten(y_true)\n",
    "    y_pred_f = K.flatten(y_pred)\n",
    "    intersection = K.sum(y_true_f * y_pred_f)\n",
    "    return (2. * intersection + smooth) / (K.sum(y_true_f) + K.sum(y_pred_f) + smooth)\n",
    "\n",
    "def dice_loss(y_true, y_pred):\n",
    "    smooth = 1.\n",
    "    y_true_f = K.flatten(y_true)\n",
    "    y_pred_f = K.flatten(y_pred)\n",
    "    intersection = y_true_f * y_pred_f\n",
    "    score = (2. * K.sum(intersection) + smooth) / (K.sum(y_true_f) + K.sum(y_pred_f) + smooth)\n",
    "    return 1. - score\n",
    "\n",
    "def bce_dice_loss(y_true, y_pred):\n",
    "    return binary_crossentropy(y_true, y_pred) + dice_loss(y_true, y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "le = LabelEncoder()\n",
    "kf = StratifiedKFold(n_splits=5)\n",
    "split_idx = kf.split(mask_count_df.index, le.fit_transform(mask_count_df.values[:,1]))\n",
    "\n",
    "t = [(mask_count_df.index[train], mask_count_df.index[test]) for i, (train,test) in enumerate(split_idx) if i == 0]\n",
    "train_idx, val_idx = t[0][0], t[0][1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "train_idx, val_idx = train_test_split(\n",
    "    mask_count_df.index, random_state=42, test_size=0.20\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Int64Index([ 821, 1885, 3260, 1872, 4464,  716, 5330, 4473, 4116,  468,\n",
       "            ...\n",
       "            5175,  406,  440, 5171,  379, 5150,  342,  368,  488,  489],\n",
       "           dtype='int64', length=1111)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "val_idx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Image_Label</th>\n",
       "      <th>EncodedPixels</th>\n",
       "      <th>ImageId</th>\n",
       "      <th>ClassId</th>\n",
       "      <th>hasMask</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0011165.jpg_Fish</td>\n",
       "      <td>264918 937 266318 937 267718 937 269118 937 27...</td>\n",
       "      <td>0011165.jpg</td>\n",
       "      <td>Fish</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0011165.jpg_Flower</td>\n",
       "      <td>1355565 1002 1356965 1002 1358365 1002 1359765...</td>\n",
       "      <td>0011165.jpg</td>\n",
       "      <td>Flower</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0011165.jpg_Gravel</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0011165.jpg</td>\n",
       "      <td>Gravel</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0011165.jpg_Sugar</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0011165.jpg</td>\n",
       "      <td>Sugar</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>002be4f.jpg_Fish</td>\n",
       "      <td>233813 878 235213 878 236613 878 238010 881 23...</td>\n",
       "      <td>002be4f.jpg</td>\n",
       "      <td>Fish</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          Image_Label                                      EncodedPixels  \\\n",
       "0    0011165.jpg_Fish  264918 937 266318 937 267718 937 269118 937 27...   \n",
       "1  0011165.jpg_Flower  1355565 1002 1356965 1002 1358365 1002 1359765...   \n",
       "2  0011165.jpg_Gravel                                                NaN   \n",
       "3   0011165.jpg_Sugar                                                NaN   \n",
       "4    002be4f.jpg_Fish  233813 878 235213 878 236613 878 238010 881 23...   \n",
       "\n",
       "       ImageId ClassId  hasMask  \n",
       "0  0011165.jpg    Fish     True  \n",
       "1  0011165.jpg  Flower     True  \n",
       "2  0011165.jpg  Gravel    False  \n",
       "3  0011165.jpg   Sugar    False  \n",
       "4  002be4f.jpg    Fish     True  "
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "BATCH_SIZE = 16\n",
    "\n",
    "train_generator = DataGenerator(\n",
    "    train_idx, \n",
    "    df=mask_count_df,\n",
    "    target_df=train_df,\n",
    "    batch_size=BATCH_SIZE,\n",
    "    reshape=(320, 480),\n",
    "    augment=True,\n",
    "    n_channels=3,\n",
    "    n_classes=4,\n",
    "    gamma = 0.8\n",
    ")\n",
    "\n",
    "val_generator = DataGenerator(\n",
    "    val_idx, \n",
    "    df=mask_count_df,\n",
    "    target_df=train_df,\n",
    "    batch_size=BATCH_SIZE, \n",
    "    shuffle=False,\n",
    "    reshape=(320, 480),\n",
    "    augment=False,\n",
    "    n_channels=3,\n",
    "    n_classes=4,\n",
    "    gamma = 0.8\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://github.com/qubvel/classification_models/releases/download/0.0.1/resnet34_imagenet_1000_no_top.h5\n",
      "85524480/85521592 [==============================] - 4s 0us/step\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "data (InputLayer)               (None, 320, 480, 3)  0                                            \n",
      "__________________________________________________________________________________________________\n",
      "bn_data (BatchNormalization)    (None, 320, 480, 3)  9           data[0][0]                       \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_1 (ZeroPadding2D (None, 326, 486, 3)  0           bn_data[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "conv0 (Conv2D)                  (None, 160, 240, 64) 9408        zero_padding2d_1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "bn0 (BatchNormalization)        (None, 160, 240, 64) 256         conv0[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "relu0 (Activation)              (None, 160, 240, 64) 0           bn0[0][0]                        \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_2 (ZeroPadding2D (None, 162, 242, 64) 0           relu0[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "pooling0 (MaxPooling2D)         (None, 80, 120, 64)  0           zero_padding2d_2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit1_bn1 (BatchNormaliz (None, 80, 120, 64)  256         pooling0[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit1_relu1 (Activation) (None, 80, 120, 64)  0           stage1_unit1_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_3 (ZeroPadding2D (None, 82, 122, 64)  0           stage1_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit1_conv1 (Conv2D)     (None, 80, 120, 64)  36864       zero_padding2d_3[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit1_bn2 (BatchNormaliz (None, 80, 120, 64)  256         stage1_unit1_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit1_relu2 (Activation) (None, 80, 120, 64)  0           stage1_unit1_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_4 (ZeroPadding2D (None, 82, 122, 64)  0           stage1_unit1_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit1_conv2 (Conv2D)     (None, 80, 120, 64)  36864       zero_padding2d_4[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit1_sc (Conv2D)        (None, 80, 120, 64)  4096        stage1_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "add_1 (Add)                     (None, 80, 120, 64)  0           stage1_unit1_conv2[0][0]         \n",
      "                                                                 stage1_unit1_sc[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit2_bn1 (BatchNormaliz (None, 80, 120, 64)  256         add_1[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit2_relu1 (Activation) (None, 80, 120, 64)  0           stage1_unit2_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_5 (ZeroPadding2D (None, 82, 122, 64)  0           stage1_unit2_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit2_conv1 (Conv2D)     (None, 80, 120, 64)  36864       zero_padding2d_5[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit2_bn2 (BatchNormaliz (None, 80, 120, 64)  256         stage1_unit2_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit2_relu2 (Activation) (None, 80, 120, 64)  0           stage1_unit2_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_6 (ZeroPadding2D (None, 82, 122, 64)  0           stage1_unit2_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit2_conv2 (Conv2D)     (None, 80, 120, 64)  36864       zero_padding2d_6[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "add_2 (Add)                     (None, 80, 120, 64)  0           stage1_unit2_conv2[0][0]         \n",
      "                                                                 add_1[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit3_bn1 (BatchNormaliz (None, 80, 120, 64)  256         add_2[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit3_relu1 (Activation) (None, 80, 120, 64)  0           stage1_unit3_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_7 (ZeroPadding2D (None, 82, 122, 64)  0           stage1_unit3_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit3_conv1 (Conv2D)     (None, 80, 120, 64)  36864       zero_padding2d_7[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit3_bn2 (BatchNormaliz (None, 80, 120, 64)  256         stage1_unit3_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit3_relu2 (Activation) (None, 80, 120, 64)  0           stage1_unit3_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_8 (ZeroPadding2D (None, 82, 122, 64)  0           stage1_unit3_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage1_unit3_conv2 (Conv2D)     (None, 80, 120, 64)  36864       zero_padding2d_8[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "add_3 (Add)                     (None, 80, 120, 64)  0           stage1_unit3_conv2[0][0]         \n",
      "                                                                 add_2[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit1_bn1 (BatchNormaliz (None, 80, 120, 64)  256         add_3[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit1_relu1 (Activation) (None, 80, 120, 64)  0           stage2_unit1_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_9 (ZeroPadding2D (None, 82, 122, 64)  0           stage2_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit1_conv1 (Conv2D)     (None, 40, 60, 128)  73728       zero_padding2d_9[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit1_bn2 (BatchNormaliz (None, 40, 60, 128)  512         stage2_unit1_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit1_relu2 (Activation) (None, 40, 60, 128)  0           stage2_unit1_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_10 (ZeroPadding2 (None, 42, 62, 128)  0           stage2_unit1_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit1_conv2 (Conv2D)     (None, 40, 60, 128)  147456      zero_padding2d_10[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit1_sc (Conv2D)        (None, 40, 60, 128)  8192        stage2_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "add_4 (Add)                     (None, 40, 60, 128)  0           stage2_unit1_conv2[0][0]         \n",
      "                                                                 stage2_unit1_sc[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit2_bn1 (BatchNormaliz (None, 40, 60, 128)  512         add_4[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit2_relu1 (Activation) (None, 40, 60, 128)  0           stage2_unit2_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_11 (ZeroPadding2 (None, 42, 62, 128)  0           stage2_unit2_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit2_conv1 (Conv2D)     (None, 40, 60, 128)  147456      zero_padding2d_11[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit2_bn2 (BatchNormaliz (None, 40, 60, 128)  512         stage2_unit2_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit2_relu2 (Activation) (None, 40, 60, 128)  0           stage2_unit2_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_12 (ZeroPadding2 (None, 42, 62, 128)  0           stage2_unit2_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit2_conv2 (Conv2D)     (None, 40, 60, 128)  147456      zero_padding2d_12[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_5 (Add)                     (None, 40, 60, 128)  0           stage2_unit2_conv2[0][0]         \n",
      "                                                                 add_4[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit3_bn1 (BatchNormaliz (None, 40, 60, 128)  512         add_5[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit3_relu1 (Activation) (None, 40, 60, 128)  0           stage2_unit3_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_13 (ZeroPadding2 (None, 42, 62, 128)  0           stage2_unit3_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit3_conv1 (Conv2D)     (None, 40, 60, 128)  147456      zero_padding2d_13[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit3_bn2 (BatchNormaliz (None, 40, 60, 128)  512         stage2_unit3_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit3_relu2 (Activation) (None, 40, 60, 128)  0           stage2_unit3_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_14 (ZeroPadding2 (None, 42, 62, 128)  0           stage2_unit3_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit3_conv2 (Conv2D)     (None, 40, 60, 128)  147456      zero_padding2d_14[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_6 (Add)                     (None, 40, 60, 128)  0           stage2_unit3_conv2[0][0]         \n",
      "                                                                 add_5[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit4_bn1 (BatchNormaliz (None, 40, 60, 128)  512         add_6[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit4_relu1 (Activation) (None, 40, 60, 128)  0           stage2_unit4_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_15 (ZeroPadding2 (None, 42, 62, 128)  0           stage2_unit4_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit4_conv1 (Conv2D)     (None, 40, 60, 128)  147456      zero_padding2d_15[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit4_bn2 (BatchNormaliz (None, 40, 60, 128)  512         stage2_unit4_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit4_relu2 (Activation) (None, 40, 60, 128)  0           stage2_unit4_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_16 (ZeroPadding2 (None, 42, 62, 128)  0           stage2_unit4_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage2_unit4_conv2 (Conv2D)     (None, 40, 60, 128)  147456      zero_padding2d_16[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_7 (Add)                     (None, 40, 60, 128)  0           stage2_unit4_conv2[0][0]         \n",
      "                                                                 add_6[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit1_bn1 (BatchNormaliz (None, 40, 60, 128)  512         add_7[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit1_relu1 (Activation) (None, 40, 60, 128)  0           stage3_unit1_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_17 (ZeroPadding2 (None, 42, 62, 128)  0           stage3_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit1_conv1 (Conv2D)     (None, 20, 30, 256)  294912      zero_padding2d_17[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit1_bn2 (BatchNormaliz (None, 20, 30, 256)  1024        stage3_unit1_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit1_relu2 (Activation) (None, 20, 30, 256)  0           stage3_unit1_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_18 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit1_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit1_conv2 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_18[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit1_sc (Conv2D)        (None, 20, 30, 256)  32768       stage3_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "add_8 (Add)                     (None, 20, 30, 256)  0           stage3_unit1_conv2[0][0]         \n",
      "                                                                 stage3_unit1_sc[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit2_bn1 (BatchNormaliz (None, 20, 30, 256)  1024        add_8[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit2_relu1 (Activation) (None, 20, 30, 256)  0           stage3_unit2_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_19 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit2_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit2_conv1 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_19[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit2_bn2 (BatchNormaliz (None, 20, 30, 256)  1024        stage3_unit2_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit2_relu2 (Activation) (None, 20, 30, 256)  0           stage3_unit2_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_20 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit2_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit2_conv2 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_20[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_9 (Add)                     (None, 20, 30, 256)  0           stage3_unit2_conv2[0][0]         \n",
      "                                                                 add_8[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit3_bn1 (BatchNormaliz (None, 20, 30, 256)  1024        add_9[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit3_relu1 (Activation) (None, 20, 30, 256)  0           stage3_unit3_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_21 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit3_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit3_conv1 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_21[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit3_bn2 (BatchNormaliz (None, 20, 30, 256)  1024        stage3_unit3_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit3_relu2 (Activation) (None, 20, 30, 256)  0           stage3_unit3_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_22 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit3_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit3_conv2 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_22[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_10 (Add)                    (None, 20, 30, 256)  0           stage3_unit3_conv2[0][0]         \n",
      "                                                                 add_9[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit4_bn1 (BatchNormaliz (None, 20, 30, 256)  1024        add_10[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit4_relu1 (Activation) (None, 20, 30, 256)  0           stage3_unit4_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_23 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit4_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit4_conv1 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_23[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit4_bn2 (BatchNormaliz (None, 20, 30, 256)  1024        stage3_unit4_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit4_relu2 (Activation) (None, 20, 30, 256)  0           stage3_unit4_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_24 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit4_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit4_conv2 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_24[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_11 (Add)                    (None, 20, 30, 256)  0           stage3_unit4_conv2[0][0]         \n",
      "                                                                 add_10[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit5_bn1 (BatchNormaliz (None, 20, 30, 256)  1024        add_11[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit5_relu1 (Activation) (None, 20, 30, 256)  0           stage3_unit5_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_25 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit5_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit5_conv1 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_25[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit5_bn2 (BatchNormaliz (None, 20, 30, 256)  1024        stage3_unit5_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit5_relu2 (Activation) (None, 20, 30, 256)  0           stage3_unit5_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_26 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit5_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit5_conv2 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_26[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_12 (Add)                    (None, 20, 30, 256)  0           stage3_unit5_conv2[0][0]         \n",
      "                                                                 add_11[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit6_bn1 (BatchNormaliz (None, 20, 30, 256)  1024        add_12[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit6_relu1 (Activation) (None, 20, 30, 256)  0           stage3_unit6_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_27 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit6_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit6_conv1 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_27[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit6_bn2 (BatchNormaliz (None, 20, 30, 256)  1024        stage3_unit6_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit6_relu2 (Activation) (None, 20, 30, 256)  0           stage3_unit6_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_28 (ZeroPadding2 (None, 22, 32, 256)  0           stage3_unit6_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage3_unit6_conv2 (Conv2D)     (None, 20, 30, 256)  589824      zero_padding2d_28[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_13 (Add)                    (None, 20, 30, 256)  0           stage3_unit6_conv2[0][0]         \n",
      "                                                                 add_12[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit1_bn1 (BatchNormaliz (None, 20, 30, 256)  1024        add_13[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit1_relu1 (Activation) (None, 20, 30, 256)  0           stage4_unit1_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_29 (ZeroPadding2 (None, 22, 32, 256)  0           stage4_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit1_conv1 (Conv2D)     (None, 10, 15, 512)  1179648     zero_padding2d_29[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit1_bn2 (BatchNormaliz (None, 10, 15, 512)  2048        stage4_unit1_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit1_relu2 (Activation) (None, 10, 15, 512)  0           stage4_unit1_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_30 (ZeroPadding2 (None, 12, 17, 512)  0           stage4_unit1_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit1_conv2 (Conv2D)     (None, 10, 15, 512)  2359296     zero_padding2d_30[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit1_sc (Conv2D)        (None, 10, 15, 512)  131072      stage4_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "add_14 (Add)                    (None, 10, 15, 512)  0           stage4_unit1_conv2[0][0]         \n",
      "                                                                 stage4_unit1_sc[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit2_bn1 (BatchNormaliz (None, 10, 15, 512)  2048        add_14[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit2_relu1 (Activation) (None, 10, 15, 512)  0           stage4_unit2_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_31 (ZeroPadding2 (None, 12, 17, 512)  0           stage4_unit2_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit2_conv1 (Conv2D)     (None, 10, 15, 512)  2359296     zero_padding2d_31[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit2_bn2 (BatchNormaliz (None, 10, 15, 512)  2048        stage4_unit2_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit2_relu2 (Activation) (None, 10, 15, 512)  0           stage4_unit2_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_32 (ZeroPadding2 (None, 12, 17, 512)  0           stage4_unit2_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit2_conv2 (Conv2D)     (None, 10, 15, 512)  2359296     zero_padding2d_32[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_15 (Add)                    (None, 10, 15, 512)  0           stage4_unit2_conv2[0][0]         \n",
      "                                                                 add_14[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit3_bn1 (BatchNormaliz (None, 10, 15, 512)  2048        add_15[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit3_relu1 (Activation) (None, 10, 15, 512)  0           stage4_unit3_bn1[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_33 (ZeroPadding2 (None, 12, 17, 512)  0           stage4_unit3_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit3_conv1 (Conv2D)     (None, 10, 15, 512)  2359296     zero_padding2d_33[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit3_bn2 (BatchNormaliz (None, 10, 15, 512)  2048        stage4_unit3_conv1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit3_relu2 (Activation) (None, 10, 15, 512)  0           stage4_unit3_bn2[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_34 (ZeroPadding2 (None, 12, 17, 512)  0           stage4_unit3_relu2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "stage4_unit3_conv2 (Conv2D)     (None, 10, 15, 512)  2359296     zero_padding2d_34[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "add_16 (Add)                    (None, 10, 15, 512)  0           stage4_unit3_conv2[0][0]         \n",
      "                                                                 add_15[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "bn1 (BatchNormalization)        (None, 10, 15, 512)  2048        add_16[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "relu1 (Activation)              (None, 10, 15, 512)  0           bn1[0][0]                        \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage0_upsample (UpSamp (None, 20, 30, 512)  0           relu1[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, 20, 30, 768)  0           decoder_stage0_upsample[0][0]    \n",
      "                                                                 stage4_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage0_conv1 (Conv2D)   (None, 20, 30, 256)  1769472     concatenate_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage0_bn1 (BatchNormal (None, 20, 30, 256)  1024        decoder_stage0_conv1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage0_relu1 (Activatio (None, 20, 30, 256)  0           decoder_stage0_bn1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage0_conv2 (Conv2D)   (None, 20, 30, 256)  589824      decoder_stage0_relu1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage0_bn2 (BatchNormal (None, 20, 30, 256)  1024        decoder_stage0_conv2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage0_relu2 (Activatio (None, 20, 30, 256)  0           decoder_stage0_bn2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage1_upsample (UpSamp (None, 40, 60, 256)  0           decoder_stage0_relu2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_2 (Concatenate)     (None, 40, 60, 384)  0           decoder_stage1_upsample[0][0]    \n",
      "                                                                 stage3_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage1_conv1 (Conv2D)   (None, 40, 60, 128)  442368      concatenate_2[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage1_bn1 (BatchNormal (None, 40, 60, 128)  512         decoder_stage1_conv1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage1_relu1 (Activatio (None, 40, 60, 128)  0           decoder_stage1_bn1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage1_conv2 (Conv2D)   (None, 40, 60, 128)  147456      decoder_stage1_relu1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage1_bn2 (BatchNormal (None, 40, 60, 128)  512         decoder_stage1_conv2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage1_relu2 (Activatio (None, 40, 60, 128)  0           decoder_stage1_bn2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage2_upsample (UpSamp (None, 80, 120, 128) 0           decoder_stage1_relu2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_3 (Concatenate)     (None, 80, 120, 192) 0           decoder_stage2_upsample[0][0]    \n",
      "                                                                 stage2_unit1_relu1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage2_conv1 (Conv2D)   (None, 80, 120, 64)  110592      concatenate_3[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage2_bn1 (BatchNormal (None, 80, 120, 64)  256         decoder_stage2_conv1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage2_relu1 (Activatio (None, 80, 120, 64)  0           decoder_stage2_bn1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage2_conv2 (Conv2D)   (None, 80, 120, 64)  36864       decoder_stage2_relu1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage2_bn2 (BatchNormal (None, 80, 120, 64)  256         decoder_stage2_conv2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage2_relu2 (Activatio (None, 80, 120, 64)  0           decoder_stage2_bn2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage3_upsample (UpSamp (None, 160, 240, 64) 0           decoder_stage2_relu2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_4 (Concatenate)     (None, 160, 240, 128 0           decoder_stage3_upsample[0][0]    \n",
      "                                                                 relu0[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage3_conv1 (Conv2D)   (None, 160, 240, 32) 36864       concatenate_4[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage3_bn1 (BatchNormal (None, 160, 240, 32) 128         decoder_stage3_conv1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage3_relu1 (Activatio (None, 160, 240, 32) 0           decoder_stage3_bn1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage3_conv2 (Conv2D)   (None, 160, 240, 32) 9216        decoder_stage3_relu1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage3_bn2 (BatchNormal (None, 160, 240, 32) 128         decoder_stage3_conv2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage3_relu2 (Activatio (None, 160, 240, 32) 0           decoder_stage3_bn2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage4_upsample (UpSamp (None, 320, 480, 32) 0           decoder_stage3_relu2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage4_conv1 (Conv2D)   (None, 320, 480, 16) 4608        decoder_stage4_upsample[0][0]    \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage4_bn1 (BatchNormal (None, 320, 480, 16) 64          decoder_stage4_conv1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage4_relu1 (Activatio (None, 320, 480, 16) 0           decoder_stage4_bn1[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage4_conv2 (Conv2D)   (None, 320, 480, 16) 2304        decoder_stage4_relu1[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage4_bn2 (BatchNormal (None, 320, 480, 16) 64          decoder_stage4_conv2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "decoder_stage4_relu2 (Activatio (None, 320, 480, 16) 0           decoder_stage4_bn2[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "final_conv (Conv2D)             (None, 320, 480, 4)  580         decoder_stage4_relu2[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "sigmoid (Activation)            (None, 320, 480, 4)  0           final_conv[0][0]                 \n",
      "==================================================================================================\n",
      "Total params: 24,456,589\n",
      "Trainable params: 3,167,495\n",
      "Non-trainable params: 21,289,094\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = sm.Unet(backbone_name = 'resnet34', encoder_weights='imagenet',\n",
    "    classes=4,\n",
    "    input_shape=(320, 480, 3),\n",
    "    activation='sigmoid',encoder_freeze=True)\n",
    "\n",
    "\n",
    "model.compile(optimizer=RAdam(), loss=bce_dice_loss, metrics=[dice_coef])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/12\n",
      "277/277 [==============================] - 778s 3s/step - loss: 1.1311 - dice_coef: 0.3403 - val_loss: 0.9801 - val_dice_coef: 0.4265\n",
      "Epoch 2/12\n",
      "277/277 [==============================] - 711s 3s/step - loss: 0.9028 - dice_coef: 0.4458 - val_loss: 0.8945 - val_dice_coef: 0.4998\n",
      "Epoch 3/12\n",
      "277/277 [==============================] - 711s 3s/step - loss: 0.8596 - dice_coef: 0.4783 - val_loss: 0.8902 - val_dice_coef: 0.4954\n",
      "Epoch 4/12\n",
      "277/277 [==============================] - 741s 3s/step - loss: 0.8355 - dice_coef: 0.4948 - val_loss: 0.8182 - val_dice_coef: 0.5033\n",
      "Epoch 5/12\n",
      "277/277 [==============================] - 708s 3s/step - loss: 0.8235 - dice_coef: 0.5034 - val_loss: 0.7993 - val_dice_coef: 0.5294\n",
      "Epoch 6/12\n",
      "277/277 [==============================] - 720s 3s/step - loss: 0.8166 - dice_coef: 0.5081 - val_loss: 0.8306 - val_dice_coef: 0.5085\n",
      "Epoch 7/12\n",
      "277/277 [==============================] - 713s 3s/step - loss: 0.8067 - dice_coef: 0.5140 - val_loss: 0.8168 - val_dice_coef: 0.5260\n",
      "Epoch 8/12\n",
      "277/277 [==============================] - 723s 3s/step - loss: 0.8041 - dice_coef: 0.5159 - val_loss: 0.7857 - val_dice_coef: 0.5243\n",
      "Epoch 9/12\n",
      "277/277 [==============================] - 703s 3s/step - loss: 0.7986 - dice_coef: 0.5192 - val_loss: 0.7841 - val_dice_coef: 0.5399\n",
      "Epoch 10/12\n",
      "277/277 [==============================] - 759s 3s/step - loss: 0.7947 - dice_coef: 0.5215 - val_loss: 0.7930 - val_dice_coef: 0.5118\n",
      "Epoch 11/12\n",
      "277/277 [==============================] - 747s 3s/step - loss: 0.7900 - dice_coef: 0.5245 - val_loss: 0.7896 - val_dice_coef: 0.5346\n",
      "Epoch 12/12\n",
      "277/277 [==============================] - 787s 3s/step - loss: 0.7887 - dice_coef: 0.5255 - val_loss: 0.7668 - val_dice_coef: 0.5389\n"
     ]
    }
   ],
   "source": [
    "checkpoint = ModelCheckpoint('res34_unet_freeze.h5', save_best_only=True)\n",
    "es = EarlyStopping(monitor='val_dice_coef', patience=4, verbose=1, mode='max', restore_best_weights=True)\n",
    "\n",
    "history = model.fit_generator(\n",
    "    train_generator,\n",
    "    validation_data=val_generator,\n",
    "    callbacks=[checkpoint, es],\n",
    "    epochs=12\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f653e1fd908>"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "with open('history.json', 'w') as f:\n",
    "    json.dump(history.history, f)\n",
    "\n",
    "history_df = pd.DataFrame(history.history)\n",
    "history_df[['loss', 'val_loss']].plot()\n",
    "history_df[['dice_coef', 'val_dice_coef']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "del history_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "277/277 [==============================] - 701s 3s/step - loss: 0.7977 - dice_coef: 0.5193 - val_loss: 0.8495 - val_dice_coef: 0.5006\n",
      "Epoch 2/15\n",
      "277/277 [==============================] - 688s 2s/step - loss: 0.7958 - dice_coef: 0.5183 - val_loss: 0.8297 - val_dice_coef: 0.5110\n",
      "Epoch 3/15\n",
      "277/277 [==============================] - 1034s 4s/step - loss: 0.7928 - dice_coef: 0.5216 - val_loss: 0.8909 - val_dice_coef: 0.4831\n",
      "Epoch 4/15\n",
      "277/277 [==============================] - 716s 3s/step - loss: 0.7919 - dice_coef: 0.5212 - val_loss: 0.8030 - val_dice_coef: 0.5228\n",
      "Epoch 5/15\n",
      "277/277 [==============================] - 783s 3s/step - loss: 0.7883 - dice_coef: 0.5242 - val_loss: 0.7797 - val_dice_coef: 0.5291\n",
      "Epoch 6/15\n",
      "277/277 [==============================] - 792s 3s/step - loss: 0.7815 - dice_coef: 0.5292 - val_loss: 0.8559 - val_dice_coef: 0.5062\n",
      "Epoch 7/15\n",
      "277/277 [==============================] - 740s 3s/step - loss: 0.7766 - dice_coef: 0.5321 - val_loss: 0.8184 - val_dice_coef: 0.5130\n",
      "Epoch 8/15\n",
      "277/277 [==============================] - 762s 3s/step - loss: 0.7732 - dice_coef: 0.5345 - val_loss: 0.7604 - val_dice_coef: 0.5374\n",
      "Epoch 9/15\n",
      "277/277 [==============================] - 727s 3s/step - loss: 0.7657 - dice_coef: 0.5398 - val_loss: 0.9660 - val_dice_coef: 0.4593\n",
      "Epoch 10/15\n",
      "277/277 [==============================] - 799s 3s/step - loss: 0.7660 - dice_coef: 0.5397 - val_loss: 0.8059 - val_dice_coef: 0.5169\n",
      "Epoch 11/15\n",
      "277/277 [==============================] - 801s 3s/step - loss: 0.7588 - dice_coef: 0.5452 - val_loss: 0.8174 - val_dice_coef: 0.5113\n",
      "Epoch 12/15\n",
      "277/277 [==============================] - 832s 3s/step - loss: 0.7543 - dice_coef: 0.5478 - val_loss: 0.8656 - val_dice_coef: 0.5179\n",
      "Epoch 13/15\n",
      "277/277 [==============================] - 760s 3s/step - loss: 0.7508 - dice_coef: 0.5495 - val_loss: 0.8486 - val_dice_coef: 0.5231\n",
      "Epoch 14/15\n",
      "277/277 [==============================] - 760s 3s/step - loss: 0.7421 - dice_coef: 0.5556 - val_loss: 0.7699 - val_dice_coef: 0.5422\n",
      "Epoch 15/15\n",
      "277/277 [==============================] - 730s 3s/step - loss: 0.7441 - dice_coef: 0.5548 - val_loss: 0.8438 - val_dice_coef: 0.5091\n"
     ]
    }
   ],
   "source": [
    "model.load_weights('res34_unet_freeze.h5')\n",
    "set_trainable(model)\n",
    "model.compile(optimizer=RAdam(lr=0.0007), loss=bce_dice_loss, metrics=[dice_coef])\n",
    "\n",
    "checkpoint = ModelCheckpoint('res34_unet_unfreeze.h5', save_best_only=True)\n",
    "es = EarlyStopping(monitor='val_dice_coef', patience=8, verbose=1, mode='max', restore_best_weights=True)\n",
    "\n",
    "train_generator = DataGenerator(\n",
    "    train_idx, \n",
    "    df=mask_count_df,\n",
    "    target_df=train_df,\n",
    "    batch_size=BATCH_SIZE,\n",
    "    reshape=(320, 480),\n",
    "    augment=True,\n",
    "    n_channels=3,\n",
    "    n_classes=4,\n",
    "    gamma = 0.8\n",
    ")\n",
    "\n",
    "val_generator = DataGenerator(\n",
    "    val_idx, \n",
    "    df=mask_count_df,\n",
    "    target_df=train_df,\n",
    "    batch_size=BATCH_SIZE, \n",
    "    reshape=(320, 480),\n",
    "    augment=False,\n",
    "    n_channels=3,\n",
    "    n_classes=4,\n",
    "    gamma = 0.8\n",
    ")\n",
    "\n",
    "\n",
    "history = model.fit_generator(\n",
    "    train_generator,\n",
    "    validation_data=val_generator,\n",
    "    callbacks=[checkpoint, es],\n",
    "    epochs=15\n",
    ")\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('history_2.json', 'w') as f:\n",
    "    json.dump(history.history, f)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f651cbd7518>"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAD8CAYAAAB+UHOxAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzt3Xl4VNX5wPHvyb6TZBIgECCBgAgEWQKCKJtsWoVaQfFn61ZQ6661dWm1ra3WWtvaqtWiomitqFgVrQKyowKyhFUgLGEJCZAFsu85vz9OEgIkZCaZmTuTvJ/n4cksd+59E5L73nuW9yitNUIIIdofH6sDEEIIYQ1JAEII0U5JAhBCiHZKEoAQQrRTkgCEEKKdkgQghBDtlF0JQCk1RSm1Rym1Tyn1aCPv36KUylZKban9N6vBe92VUkuUUruUUt8rpRKcF74QQoiWUs3NA1BK+QJpwEQgA9gA3KC1/r7BNrcAKVrrexr5/Ergaa31V0qpMKBGa13itO9ACCFEi9hzBzAc2Ke1PqC1rgDmA9Ps2blSqh/gp7X+CkBrXSQnfyGE8Ax+dmzTFTjS4HkGcHEj212rlBqNuVt4UGt9BOgDnFJK/RdIBJYCj2qtq5s6WExMjE5ISLAzfCGEEACbNm3K0VrHOvIZexKAauS1s9uNPgPe01qXK6XuBOYB42v3fxkwGDgMvA/cArxxxgGUuh24HaB79+5s3LjRgW9BCCGEUuqQo5+xpwkoA+jW4Hk8kNlwA611rta6vPbpa8DQBp9NrW0+qgI+AYacfQCt9RytdYrWOiU21qEEJoQQooXsSQAbgN5KqUSlVAAwE1jYcAOlVFyDp1OBXQ0+G6WUqjurjwe+RwghhOWabQLSWlcppe4BFgO+wFyt9U6l1FPARq31QuA+pdRUoArIwzTzoLWuVko9DCxTSilgE+YOQQghhMWaHQbqbikpKfrsPoDKykoyMjIoKyuzKCrRmKCgIOLj4/H397c6FCHaPaXUJq11iiOfsacT2HIZGRmEh4eTkJCAuZEQVtNak5ubS0ZGBomJiVaHI4RoAa8oBVFWVobNZpOTvwdRSmGz2eSuTAgv5hUJAJCTvweS/xMhvJtXNAEJIYQ325VVwLJdxwkJ8CMi2J+IIPO1Q7B//fOwQD+3X1RJAhBCCBc5kF3E35bu5fNtmTQ33sZHUZsM6hKDX4PHZyWNIPP+6cctG4ghCaCFfvvb3xIWFkZBQQGjR49mwoQJVodULzs7m6uuuoqKigr+8Y9/cNlll1kdkhDtSsbJEv6xbC8fbT5KoJ8Pd43txU8v7YmvUhSUVZJfWklBaWWDx1VnvV5FfmklJwqKzGtllZRV1jg9TkkArfTUU09ZHcI5li1bRt++fZk3b57VoQjRrpwoLOPl5ft477sjoODmkQncNa4XMWGB9dt0CPE/o7SCvcqrqusTRUFpbbIoq6p/fM+fHN+n1yWA3322k+8zC5y6z35dIvjN1f2b3e7pp5/m7bffplu3bsTGxjJ06FBuueUWrrrqKqZPn86GDRu4//77KS4uJjAwkGXLlhESEsKjjz7KypUrKS8v5+677+aOO+5o8hjPPfcc77zzDj4+PlxxxRU8++yzbNmyhTvvvJOSkhJ69erF3LlziYqKYv/+/dx9991kZ2cTEhLCa6+9RllZGb/85S8pLS1l0KBBrF27luDgYGf+uIQQZzlZXMGrq/cz79uDVFZrrkvpxr3jk+gS6by/vUA/X2LDfYkND2z0/XNq8dvB6xKAVTZt2sT8+fNJTU2lqqqKIUOGMHTo0Pr3KyoquP7663n//fcZNmwYBQUFBAcH88Ybb9ChQwc2bNhAeXk5o0aNYtKkSY2Onf/yyy/55JNPWL9+PSEhIeTl5QFw00038eKLLzJmzBiefPJJfve73/HCCy9w++238+qrr9K7d2/Wr1/PXXfdxfLly3nqqafYuHEjL730ktt+PkK0R4VllbzxdTpvrEmnqKKKaRd14YEJfUiICbU6NLt4XQKw50rdFdasWcM111xDSEgIAFOnTj3j/T179hAXF8ewYcMAiIiIAGDJkiVs27aNBQsWAJCfn8/evXsbTQBLly7l1ltvrT9GdHQ0+fn5nDp1ijFjxgBw8803M2PGDIqKivj222+ZMWNG/efLy8vP2acQwvnKKqt5e+1BXlm5n5MllUzu34mHJl7ABZ3DrQ7NIV6XAKx0viFaWutG39da8+KLLzJ58uRm99/UPhpTU1NDZGQkW7ZssWt7IUTrVVTV8P6Gw7y4fB8nCssZ3SeWhyf1YWB8pNWhtYjXTASz2ujRo/n4448pLS2lsLCQzz777Iz3+/btS2ZmJhs2bACgsLCQqqoqJk+ezCuvvEJlZSUAaWlpFBcXN3qMSZMmMXfuXEpKzKJpeXl5dOjQgaioKNasWQPAO++8w5gxY4iIiCAxMZEPP/wQMMlj69atLvnehWjvqqpr+HDjEcb/ZSVPfLqTBFso798+grdvG+61J3+QOwC7DRkyhOuvv55BgwbRo0ePc4ZWBgQE8P7773PvvfdSWlpKcHAwS5cuZdasWRw8eJAhQ4agtSY2NpZPPvmk0WNMmTKFLVu2kJKSQkBAAFdeeSXPPPMM8+bNq+8E7tmzJ2+++SYA7777Lj/72c/4wx/+QGVlJTNnzuSiiy5y+c9CiPaipkbzxY4s/vpVGgeyi0nu2oGnr0lmdO+YNjET3iuqge7atYsLL7zQoojE+cj/jWiLtNYs332C55eksSurgD6dwnho4gVM7t/JY0/8bbYaqBBCuMu3+3J4fskeNh8+RffoEF64fhBXX9QFXx/PPPG3hiQAC2zfvp2f/OQnZ7wWGBjI+vXrLYpIiPZNa83mw6f4y5I9fLs/l7gOQTxzTTIzUuLx9227XaWSACyQnJwso3eEcDOtNbnFFRzMKSY9p5iDucUczCmpf1xSUY0tNIAnrurHjRd3J8jf1+qQXU4SgBCiTTlZXEF6bjEHc8y/9NyS+seF5VX12/n6KLpFBZMQE8rwxGj6dApn2qAuhAa2n9Ni+/lOhRBtRkFZ5ekr+ZwSDuaax+k5xeSXVtZv56OgS2QwiTGhXDOkKwm2UBJjQkmICSU+KrhNN+/YQxKAEMJjVddo0o4XsunQSbYeOcWB2iv53OKKM7br0iGIhJhQfjAwjkSbOcEnxoTQLTqEQL+235TTUpIAhBAeo7CsktTDp9h06CSbD58k9fApimqbbWyhASR1DGNiv04kxITWX833sIW0i/Z6V5AE4CJhYWEUFRXZtW3DiqKzZs3ioYceol+/fi6O0H67d+9m5syZKKVYsGABvXr1sjok0QZorTmcV8KmQyfr/+05XojWpunmgs4R/HBwF4b2iGJo92i6RQd77Bh8byUJwMO8/vrrVodwjk8++YRp06bxu9/9zupQhBcrq6xmx9H8+pP95sMnySkyTTnhgX4M6h7JFQPiGNojiou6dSA8qGWrXAn7eV8C+PJROLbdufvsnAxXPHveTR555BF69OjBXXfdBZgVwZRSrF69mpMnT1JZWckf/vAHpk2b1uzhtNbce++9LF++nMTERBrOxh47dizPP/88KSkpLFq0iMcff5zq6mpiYmJYtmwZxcXF3HvvvWzfvp2qqip++9vfNnnM6upqHnnkERYvXoxSitmzZ3PvvfeybNkyHn74Yaqqqhg2bBivvPIKgYGBbNq0iYceeoiioiJiYmJ46623SE1N5YUXXsDX15fVq1ezYsUKB36woj07UVDG5sMn2XjwJJsOn2TH0Xwqq83veoIthNF9Ys3VfY8oencMb5MTrTyd9yUAi8ycOZMHHnigPgF88MEHLFq0iAcffJCIiAhycnIYMWIEU6dObfY29eOPP2bPnj1s376d48eP069fP2677bYztsnOzmb27NmsXr2axMTE+rUBnn76acaPH8/cuXM5deoUw4cPZ8KECYSGnlt/fM6cOaSnp5Oamoqfnx95eXmUlZVxyy23sGzZMvr06cNNN93EK6+8wt133829997Lp59+SmxsLO+//z6/+tWvmDt3LnfeeSdhYWE8/PDDTvppirZoz7FCvkvPNVf4h09yJK8UgAA/Hy6K78BtlyYytHsUQ3pEnbFClrCO9yWAZq7UXWXw4MGcOHGCzMxMsrOziYqKIi4ujgcffJDVq1fj4+PD0aNHOX78OJ07dz7vvlavXs0NN9yAr68vXbp0Yfz48edss27dOkaPHl2/bkB0dDRg1hdYuHAhzz//PABlZWUcPny40Xo8S5cu5c4778TPz69+H1u3biUxMZE+ffoAZn2Bl19+mQkTJrBjxw4mTpwImLuHuLi4Fv60RHuhtWb13hz+uWIf69PNRUpseCApPaK4eWQCQ3pE0b9LhIzE8VDelwAsNH36dBYsWMCxY8eYOXMm7777LtnZ2WzatAl/f38SEhIoKyuza1/N3SWcb32Bjz76iAsuuKDZYzS2j6aK/2mt6d+/P2vXrm12v0JU12gW7zzGP1fuY8fRAjpHBPHrH1zI5P6diY+Szlpv0b5nQTho5syZzJ8/nwULFjB9+nTy8/Pp2LEj/v7+rFixgkOHDtm1n9GjRzN//nyqq6vJyspqtF195MiRrFq1ivT0dID6JqDJkyfz4osv1p/IU1NTmzzOpEmTePXVV6mqqqrfR9++fTl48CD79u0DTq8vcMEFF5CdnV2fACorK9m5c6edPxnRXlRU1fDBhiNM/Osq7np3M8Xl1Tx37UBW/3Icsy7rSbfoEDn5exG5A3BA//79KSwspGvXrsTFxXHjjTdy9dVXk5KSwqBBg+jbt69d+7nmmmtYvnw5ycnJ9OnTp365x4ZiY2OZM2cOP/rRj6ipqaFjx4589dVXPPHEEzzwwAMMHDgQrTUJCQl8/vnnjR5n1qxZpKWlMXDgQPz9/Zk9ezb33HMPb775JjNmzKjvBL7zzjsJCAhgwYIF3HfffeTn51NVVcUDDzxA//7WLMEpPEtJRRXzvzvCa2sOkJVfRr+4CF7+vyFMGdBZOm+9mKwHIFpF/m/atvySSuatPcib36RzsqSS4YnR3DW2F2P6xMqVvoeR9QCEEE5xoqCMN75O59/rDlFcUc3lfTty17heDO0RbXVo3ktr8LCkKQnAhdxV93/x4sU88sgjZ7yWmJjIxx9/7NTjiLbvcG4J/1q9nw83ZVBVXcNVA7vws7G9uDAuwurQvNvmt2Hp7+Du7yDUZnU09bwmATQ1KsaTuavu/+TJk5k8ebLLj3M2T2s+FC23+1gBr6zcz+fbsvBVimuHxnPH6J4kxJw7v0Q4qPA4LP41lOfD95/AsJ9aHVE9uxKAUmoK8HfAF3hda/3sWe/fAvwZOFr70kta69cbvB8B7AI+1lrf42iQQUFB5ObmYrPZvC4JtFVaa3JzcwkKCrI6FNEKmw6d5JWV+1i66wQhAb7cNiqBWZf1pFOE/L86zVdPQFUpRMTD9gXelQCUUr7Ay8BEIAPYoJRaqLX+/qxN3z/Pyf33wKqWBhkfH09GRgbZ2dkt3YVwgaCgIOLj460OQzhIa82avTn8c+U+1h3IIzLEnwcn9OHmS3oQGRJgdXhtS/oa2PY+jP4F+AXC8j/AqcMQ2d3qyAD77gCGA/u01gcAlFLzgWnA2QmgUUqpoUAnYBHgUA91HX9///oZsUK0VXnFFeSXVuLno/DxUfgqhY8P+CqFb+1rfj4Kn9rn5n3774hr6idv7Wf70fz6yVs3DO/erlbBcpuqCvjfzyGyB1z2cyg8ZhLAjo/g0getjg6wLwF0BY40eJ4BXNzIdtcqpUYDacCDWusjSikf4C/AT4DLmzqAUup24HaA7t09IzMK4S47jubzr9UH+GJ7FtU1jver+PqcTgi+Pgof1eC1BomirLKGnKJyEmwh/OnaZH44uKuUaHCldS9Dzh74vw/BPxiiEyF+mGkG8qIE0Nglxtm/pZ8B72mty5VSdwLzgPHAXcAXtcmgyQNorecAc8DMA7AncCG8mdaalWnZzFl1gLUHcgkL9OO2UQn06xJBdY25Wq/Wmuqa0/9q6p5rbd6vgeqamtrtOP1+zdnbmcdoGH9hR64YECeTt1zt1GFY9Rz0vQr6TDr9evJ18OUv4Pj30Mn6NT/sSQAZQLcGz+OBzIYbaK1zGzx9DfhT7eORwGVKqbuAMCBAKVWktX605SEL4b0qqmr4dMtRXl+Tzp7jhXSOCOLxK/syc3h3IqT+fdux6DHzdcpZxSv7/xAWPQo7FkCnJ90f11nsSQAbgN5KqUTMKJ+ZwP813EApFae1zqp9OhUz4get9Y0NtrkFSJGTv2iP8ksree+7w7z5TTrHC8rp2zmcv153EVcN7EKAn5TkalP2LILdn8OE30FktzPfC+sIPcfC9g9h/BOWTwxrNgForauUUvcAizHDQOdqrXcqpZ4CNmqtFwL3KaWmAlVAHnCLC2MWwhr5GRDUAQLD7f5I5qlS5n6dzvwNRygqr+LSpBiem34Ro3vHyJDmtqiixDTxxPaFEXc1vk3yDPjkTsjYAN2Guze+s9jV9a+1/gL44qzXnmzw+DHgsWb28RbwlsMRCuEJyvLhlVHmj/cHzze7+c7MfF5bfYDPt2WhgasGxjH7sp4M6NrB9bEK63z9V9P+f8v/wK+JIbV9fwB+QbDtA+9IAEK0exvegLJTcHxHk5vUja+fs/oAX+/LITTAl5svSeC2SxPpGhnsxmCFJXL2wjd/h4EzIeHSprcLioA+U2DnxzDlj+BrXd+PJAAhmlNRAmtfNo9z953zdmV1DZ9tzWTO6gPsPlZIx/BAHpnSl/+7uDsdgqVjt13QGr54GPyCYdLvm99+4HWmLMSBVdB7guvja4IkACGak/pvKMmB3pNh72IoPQXBkRSW1XXsHiQrv4zeHcN4bvpApg3qIuPr25ud/4UDK+HK501Hb3OSJpj+pO0fSgIQwmNVV8K3/4BuI2DITbB3MTmHv+e1/VH8Z/1hCsurGNEzmmeuSWZMn1iHZuaKNqKsABY9DnGDIOU2+z7jFwj9psGO/5o7zIAQ18bYVBiWHFUIb7H9Q8g/QuUVz7O3PJp+wDPvfM4n1aO4MjmO20f3ZGB8pNVRCiuteAaKjsMN/wEfB+78kmeYMtFpi2DAj1wX33lIAhDiLFXVNezLLmL74TzGLvsjhb6JTHmnBqoz2R2omBpfyoPXjaNbtDVXbcKDZG2D7/5lrvy7DnXssz1GQXicKQ0hCUAI96up0aTnFrMt4xTbMvLZnpHPzswCSiurmeLzHTMCDvOfqMe5NSWRgV0j0cu6M9Z2CuTkL2pq4H8PQXA0XP6E45/38YUB18L6f0FJHoS4f7U1SQCi3dBak3GylG0Z+fUn/B1H8yksrwIgyN+HAV06cMPw7gzsGsEVa/+Iru7F/fc8fPrWfmtSoyOBRDuU+o6ZzPXDVyE4qmX7SJ4Oa1+CXQth6C1ODc8ekgBEm3Usv6z+RL/taD7bM05xsqQSgABfHy6MC2fa4C4MjI9kYHwHkmLD8POtLcuwbylkb4epL53ZrhvTGw6v88j1XYUbFefC0t9A90vgopkt30/cILD1Ns1AkgCEaJni8io2HTrJliOn6k/6JwrLAVMauU+ncCb168zAbh0Y2DWSCzqHn78Gz5q/QkRXGHj9ma/bkqCy2NR2j4hz4XckPNrS30B5IfzgL627EFDKdAav/CPkH4UOXZ0Xox0kAQivVFhWycaDJ1mXnsv6A3nsOJpPVY1GKegVG8alSTEMjO9Acnwk/eIiCA5wYHTG4XVw6BuY8qdzp/PbepmvufskAbRXh9eb5p9L7nNOSefk6bDyGTOX4JJ7W78/B0gCEF4hv7SSDel5rE/PZX26OeHXaPD3VVwUH8kdY3pycaKNwd0jCW9tWeU1f4UQmxn3fzZbkvmauw8SL2vdcYT3qa4yHb8RXWHMI87Zp60XdBlihhxLAhACThZXsL7uhH8gj13HCtAaAvx8GNQtknvG92ZEYjSDu0c5dnXfnKxtZrbv+F83PjknIh58A6UjuL36bo6pB3XdOxAY5rz9Js+AxY9BdhrE9nHefpshCUB4hJyicr5Lz2P9gVzWHchjz/FCwIzMGdI9igcu78PFPaMZ1C2SIH8Xlln4+m8QEA7DZjf+vo+PuWKTBND+FGTCiqchaSJceLVz9z3gR7DkV7XrBPzKufs+D0kAwhInCspYV3vCX5+ex74TRQAE+/uSkhDF1RfFMaKnjYHxke5bMCVnn6nQeOkDEHye2b22JDjxvXtiEp5j8a+gpgqu/LPzR4CFd4bE0SYBjHvcbSPMJAGI06oqzC+4C+qS1NRoVuw5wdJdx1l/II8DOcUAhAX6kZIQxbVD4rm4ZzTJXTvg72vRClnfvGBqtDS1kEcdWxLs+cLUCbKwlK9wo/3LTSftuF+Zxd1dIXkGfHo3HN0M8Q7OKm4hSQACaqph63umpolfENy1runFLBxUWFbJgk0ZvPXtQQ7llhAe5MfwhGhmDu/GiJ42+sVFnB57b6X8DNg6H1Jubb6aoy3JJMpTh0+PChJtV1U5/O9hiO5lRv64yoVXw+cPmbsASQDC5bSGvUtg6W9Nk4atN+TuhS3/tr+qYRMO5Rbz1rcH+XBjBkXlVQztEcUvJl/A5P6drbvCP59vXwK0faMwGo4EkgTQ9n3zd8jbDz/+L/gHue44QR2gzyTY8RFMftqxwnItJAmgvcrYBF89CYe+huieMOMtuHAazJ0Mq5+HQTea5hAHaK1ZeyCXuV8fZNnu4/gqxVUD47h1VCIXdfPgipnFObDpLTPpK7J789s3TABMdmVkwmp5B8zfQ/9rIOly1x8veQbs+gzSV0OvcS4/nCSA9ibvACx7ynR2hsTAFX82U9DrmnzGPQ7v/NCUqR3exEiYs5RVVrNwSyZzv0ln97FCokMDuGdcEj8e0YNOES68YnKW9a9CVRmMesC+7UOiISjSLAEo2i6t4ctHTD/P5Gfcc8zekyAwwpSGkAQgnKYoG1Y/Bxvngm8AjP6lae4Iijhzu55jTX2TNX+BwT8G/6bXsj1eUMa/1x3i3fWHySuuoG/ncJ67diBTB3Vx7VBNZyorgPVzTPurveOvlTI1gWQoaNu2+3PTRDr5GYjo4p5j+geb38VdC02ZCVc2OSEJoO2rKIa1/zTtmJUlZnbr2EfNsLPGKGXuAuZdZZpFRvzsnE22HjnFm9+k8/m2LKq1ZsKFnbh1VAIje9pQ3lYgbeMbUJ4Plz3k2OdsSWY9V9E2lRfBl49CpwEw/A73Hjt5Bmx510xI7DfNpYeSBNBWVVeZeiUrn4WiY9D3Krj8N/Zd5SZeBgmXmZIIQ26GgBCqqmtYtPMYc79OZ/PhU4QF+nHTyARuvqQHPWyhrv9+XKGy1Cz23uty6DLYsc/aepmRU+VFzp0RKjzD6uegIAOmvwG+bj5NJo6G0I5mNJAkAOEQrc0Y9aW/hZw06HYxXDcPuo9wbD/jHoc3r6D02zm8xdW8vdYsfN7DFsJvru7H9KHxra+5Y7XUf0NxNlz2c8c/W9cRnHcA4gY6Ny5hrRO7zIXB4B87/nfjDHULxWycC6Wnzj8psZUkAbQlR76DJU/AkXVmSOf170LfH7RoVuHeoGQIG0b0ir/wYnk3Bid15ffTBjCub0d828LC59WVplms2wjocYnjn284EkgSQNuhNfzv5xAYDhOesi6O5Bmw/hXTDzH4xy47jCSAtiBnr7ni3/05hHWCq16AwT9x+Na1pkazKi2bud+ks2ZvDsP9pvKB3xOsHLOXjldc65rYrVK72Ds/+GvLpt1H9zRfpSO4bdn2vikFfvXfIdRmXRxdh0BUovk9lQQgGlV4HFY9C5vmmdED434NI++CAPva5LXW7M8uZu2BXNbtz2XdgVxyiyvoFBHILyZfwA3DJ8Ina+i47VUYd9e5I4a8VU2NKfrWKRl6T2zZPgJCTWVQSQBtR+lJWPJr6JoCgxspBe5OdQvFrHneLD7U1KCNVpIE4I3KC+HbF83s1epyGPZTM6wzLPa8H9Nacyi3hLUHclm7P5e1B3LJrl01K65DEGP6xDKub0emDGgwW3fsY/DaOPjuXzD6F67+ztxj9+emf2T63NYV3ZKqoG3L8j9ASS78+CNT9dVqyTNMZ/SO/5oLOxeQBOBNqivN0MxVfzKdl/1+CJc/ed5yBEfyTp/w1x3IJSu/DIDY8EBG9rQxspeNkT1t9LCFND6Es+sQuOBKk3CG326mq3szrc0ch+ie5ufXGrYk2LFA1gf2diV5sOEN8+/iOyDuIqsjMmL7mFi2fygJQGDWDV3zF+hxKdzwfqMFo7LyS83Vfe0VfsbJUgBsoQGM6GljRO0Jv1dsqP1j9sc+Cv8aDeteMY+92f7lkLUFpr7Y+lortiQoyzdXjaExzolPuE92Gqz7pykCWFVq6vyPe9zqqM6UPMM0S+Xud0ndKUkA3mTnx9BrvClKVXvyPlFQZtrwa6/yD+aWABAZ4s/FidHMujSRkb1i6NMprOWTtOIuMrMT175srpCCo5z1Hblf/WLvM1u/r4YjgSQBeAet4cBKc+Lfu8Ss7jbwOlMC3Bnr+zpb/x+ZkX3bF8BYJy1B2YAkACfIyi9l5Z5sVuw+wa5jBQT6+RLs70uQvw9B/nWPzdfgAPM4yN/n9HM/X4ICTn+mfvuA058LKUgnKO8ARYNns2r7MdYeyGHt/lz2Z5u6+uFBflycGM2PR/RgZC8bF3aOwMeZwzXHPmaKVK192SyX6I0OrzPF76Y865xy1w0XiLdivLiwX1W5aUpZ+084sRNCY83vdMpPm+07s1SHrpBwqYl9zC+d3tRoVwJQSk0B/g74Aq9rrZ896/1bgD8DR2tfeklr/bpSahDwChABVANPa63fd1LslqmqriH1yClW7D7Bij3Z7MoqAKBrZDBDekRRXVNDWWUNpRXVFJZVkV1YTlllNaWV1ZRWVFNWVUNFVY1Dx/yp7xc84Q9TvgghQ28mNMCXYYnRXJfSjZG9bPTv0sG14/M79TcVEde9Yq6WQqJddyxXOd9i7y0R2QN8/KUonCcrzqlt33/N9Jt17AdTXzJNKy6us+M0ydPhs/shayt0GeTUXTebAJTJA5NiAAAgAElEQVRSvsDLwEQgA9iglFqotT57Tbz3tdb3nPVaCXCT1nqvUqoLsEkptVhrfcoZwbtTblE5q9KyWbEnm9Vp2eSXVuLno0hJiOKxK/oyrm9Hene0v5mlukZTVlldnxjKKqspraihrMokibrXymqTxoSNL5BblsiPx13G8ESLVs4a8yjs/AS+/QdM+K17j91aZyz27qTSFb5+ZnUoGQnkeU7sqm3ff9+MlEuaCCPvNsUOva3Dvt80syDN9g/dnwCA4cA+rfUBAKXUfGAa0OyiqFrrtAaPM5VSJ4BYwOMTQE2NZkdmPit2Z7Nizwm2ZpxCa4gJC2RSv06M69uRS3vHENHCcgi+PorQQD9CA+34LygrgGWpMPIu7hxj4QIkHfuaq5H1c2DE3Z5963y25hZ7bylbkumgE9bTGvYvM82U+5eb1e0G3WDuWGMvsDq6lguOMmWid3wEE59y6kIx9iSArsCRBs8zgIsb2e5apdRoIA14UGvd8DMopYYDAcA5fy1KqduB2wG6de+B1tqSqpL5pZV8vTeHFXtOsHJPNjlF5SgFF8VH8uCEPoy7oCP9uzi5bd0eB1ZATSX09oDFR8Y8Yn4Rv3nBrFrkDXL3w/efmOX8nF1XxdYL9i0zy2q6YQUn0YjKUtj2gbniz95tZsOP/zUMvc3a2bzOlDwd9vzPzFJOHO203dqTABo72+mznn8GvKe1LldK3QnMA8bX70CpOOAd4Gat9TmN31rrOcAcgMC43rrPr7/EFhpITHgAMWGBxIQFYgsLILb2cUyYec8WGkh0aECL27611qQdL2LFnhOs2H2CjYdOUl2j6RDsz+g+sYzvG8vo3rHYwhxbGcvp0paY8ffdGsu7bhbT26ycteF1s56Ai2YoOtU3L5g1EEbe7fx925JME0N+BkT1cP7+RdMKj5vfw41vmKG4nZLhh6/CgB85vJqdx+szBQLCTDOQmxNABtCtwfN4ILPhBlrr3AZPXwP+VPdEKRUB/A/4tdZ6XXMHi+sQxKzLepJTWE5OUTk5RRXsOVZIblEFFdXndpz6KIgOPTNR1CeJsABiwgOJrX3dFhpIVU0N3+7Lrb/KP3rKjJO/MC6CO0b3ZHzfjgzqFukZC5WDKVuwd4kpWezusrRNGf0Lc8X19QtwxbPNb2+l/KOw5T2z6llzi723RMOhoJIA3OPYDnO1v/1DqK6APleYiVIJl3lf+769AkJMSffvP4Urn3dagrPnjLIB6K2USsSM8pkJ/F/DDZRScVrrrNqnU4Fdta8HAB8Db2utP7QnoJiwQB6Z0vec17XWFJRVmaRQaBJDTlE5uUXlZNc+zikq59DhYnIKKyitrG50/74+iuoaTUiAL5cmxXDP+CTGXhBLXIemV76yVNYWKD4BfTyg+aeOrZdpW904F0bd577Vklpibe1i76Puc83+GyYAd6wZ215pDXu/Mv+f6avAP8SM5rr4ZxCTZHV07pE8A7bNh31LTZVfJ2g2AWitq5RS9wCLMcNA52qtdyqlngI2aq0XAvcppaYCVUAecEvtx68DRgO22qGiALdorbc4GqhSig7B/nQI9qdXbPMLcBSXV5FbVEF2UXl9csgprKCiupqRPWMYlhhFoJ8XtNnuXQIoSJpgdSRnGv0LM4NyzV/hB89bHU3j6hZ7T77OvsXeWyKsk+lclpFArnN0Eyz+FRxeC+FxZmGjobd451Dk1ug51qzjvf1D9yUAAK31F8AXZ732ZIPHjwGPNfK5fwP/bmWMLVI3wqa7LcSKwztP2iKIH+Z5M02jEkyZ2s3zYNT9ENmt2Y+43fpXTQfhpXYu9t4SSnlPUbiCLDMPwhmT4Nwh/ygs+50p0Rwaa8qcD7rRe+J3Nl8/07+x+W1TEDIwvNW79JCGbtGowuOQmQp9JlkdSeMue/h0cTVPc8Zi7y4eAmhL8vwEUJIH/xgE/xgM6/9lEqOnKi+C5U/Di0PNvJNLH4J7N0PKre335F8neQZUlcGuz52yO0kAnmzfV+arJwz/bExkNxh6s1l7+OQhq6M508a5LVvsvSVsSXDqCFSWuf5YLXV0szlx+AfDl7+EF5JNJ355odWRnVZTY5bpfHGoKYPc90q4ZwNM+E3bWYuiteKHmebM7XZ1qTZLEoAnS1sM4V2gc7LVkTTt0odA+cLqP1sdyWn1i72Pd3yx95awJQEaTqa7/lgtlZlqvs5eDrd8YX6nlv4G/jYAVvzR3CFYKX0NzBkDn94NHeLhp1+Z9RpkZNWZ6haKObASik60eneSADxVVQXsX2FWrPLkoW0duppb8y3/MQuke4LUf5uRUy1Z7L0lGhaF81SZqWad6KAISBgFP/nYJIMeo8yqci8km6qThcfdG1fufph/I8y7yqzIde0bMGspdBvu3ji8SfIM0NWmeayVJAF4qsNroaLQs4Z/NuXSB8HXH1Z5wF1AdSV88w8zaa7HKPccs24oqCcXhcvacu7dUNehcMN/4GffmolGa1+Cvw80dWdOHWl8P85SehIWPQ4vX2yuZi9/0jT3JE/37AseT9DxQug0wCnNQJIAPFXaYlOrPHGM1ZE0L7wzDJtlxijnWHwVvH0B5B82V//uOpEERZjhoJ5aE6jwOBQcbbo5rFN/mP4G3LPRXF1uest0GH9yt/P/P6srTSf0PwabyVyDbjAdvJf93PRPCPskz4CM7yCvdc2OkgA81d7Fpg54YPNzHjzCqPtN8a1Vf2p+W1epqYGv/2qujnq7eeSUJ48EyqqddtNcJUlbL5j2Ety/xdTJ37EAXh4GH95qZt+2htawZxH8c6TphO48EO5cY1ZmC+/Uun23RwOuNV93fNSq3UgC8ES5+83JxBuaf+qEdYThs81tafYea2KoW+z9sofc34zgyXMBMlMBZU669ugQD1c+Bw9sNwX09n4Fr46C/8yEjI2OH//YDnjnh/De9eb5De/DTZ969uAGTxfZDbpfYv7e9Nml2ewnCcAT7V1ivrr7Kra1Lrnf1Npf6eb6QFrD9wvNlaUzFntvCVsSlOSYtm1Pk7nFzIVw9G4yrCNM/B08uB3GPg5H1sHrl8Pb08yoneZOPIXHYeF98K/LzGImVzwHd62FC6ZIO78zJE831U+Pt/zuTBKAJ0pbDDF9zGIj3iTUZtYM3vkxHN/pnmMe/x7engof/MTMcp0+15qyzPU1gTxkJFRDmakQ14qFRIKjzHq0D2yHib83i63MuwrmTjaVas9OBJWlZnLgi0Ngy7umXs99qeZ3w7dl62eIRvT7Ifj4taozWBKApykvNDW/ve3qv87Ie8wUdVffBZSehC9+Ca9eCse2ww/+Arevcs+4/8bYepuvuR42EqggC4qOOefnEhhuiurdv81UpCzIhP/MgH+NNkMSa6pNJ/xLw2HZU6Z2zd3fwZRnTBIRzhVqM1WCt39k+r9awEPqC4t6B1bWlridYnUkLRMSbVZgWvWsWYYxzs52Z3vVVJv6Q8t+D2WnIOU2GPcr6wuDRSWA8vG8foD6DmAnJkb/INPfM/QWU6fn67/BhzebNSvK8k3b/g8/c2rdetGE5BlmwMiRZivtN6r93AFobX5RXxrmlBl0LpO2GAI7QPcRVkfSciN+Zk4Gzr4LOLTWzBb9/EGzuPcda8yVv9UnfzA1aiJ7eF4CyEw1ickVHa6+/qYg4N3fwfQ3TT3+aS+bOzE5+btH3ytNaewWNgO1jzuAmmrTQbjhdfN88zxTztjT1NU87zXOu9tKgyNh5L2w4g/mBNTaq8/8o/DVk2ZYYkS8Odn0v8bzOhI9cShoZirE9jULiriKj6+pUjngR647hmhcQKgpDb3z4xZ9vO3fAVSWwgc3mZP/qPtNu+TGN6G6yurIzpW11bTXetPwz6ZcfIdp913xx5bvo7LM1Bh6KcUM8RzziJktOuBHnnfyh9MLxLdiWJ5Tae2cBCw8W/KMFo8+a9sJoCTPDFnb/T8zBG3iUzD8djMrMu1Lq6M7V/3iLxOtjqT1giJqx5AvdnzsuNam3O3Lw2H5H8xiOHd/B+Med+2VbGvZekFlCRRmNb+tOxRkQnG2JIC2rtd4CG5ZM2jbTQAnD8Ebk8wY6OvmmStSMJ2rHbrBd69ZG19j0haZ+ixhsVZH4hzDbzdDM1c8Y/9nTuw2k4bev9G0bd60EK5/xzuqQsbUjQTykGagugqgrRkCKjyfrz/0b9ncl7aZALK2whsTzdXPTZ9Cv2mn3/PxNdUr01dBdpp1MZ6tKNvUbG8LzT91AsNg1AOwfxkcbmaUQukpWPQYvHKJOXFd8We482vo6QW1kOp4WlG4zFRTqrvzAKsjEa42bFaLPtb2EsD+5fDmleAbAD9dAj1GnrvN4JvM+3Wdwp5g31eA9t7x/00ZNgtCOzZ9F1BTbYqPvTgU1r1iFvq+NxUuvt0sgedNwruAX7DnFIXL2mJGS0mRtbavU/8WfaxtJYCt8+HdGWZM9k+/anopwLBYM4tu63tm+TlPkLYYwjpD3EVWR+JcASGmXHT6Kjj49ZnvHV4Hr42Dz+43zSd3rIKrXzATXLyRj4/n1ASq7wCW5h/RtLaRAOrWpf34DlMD/tYvICLu/J8ZPhvKC8xEFqtVV5o7F09f/KWlUm41yW3FM+b/qiATPpptSgkUZZtFQG79sm0kP09JAPlHoCRXEoA4L+9PADXV8MXDZup58nVw4wIzCak58cNMdcQNr1s/bO/wOpOM2lL7f0P+wabe+6Fv4LP74MUU+P5Ts6j8vRvb1iIgtiQ4edAkdSvVdQDLCCBxHt6dAM4e43/Nv8yMTHsoZe4CTnwPh751bZzNSVtk+iR6jrU2DlcachNEdIXNb5uJbnevh8ufMBNZ2hJbb7Nc38lD1saRuQV8/M3aCEI0wct62RooyYP/XA8ZG8wY/7phno4YMN2sg7rhdbNOqlX2LjFNV4Hh1sXgav5B5u6sLL/xjvm2or4q6F6ISbIujsxUs3SgX6B1MQiP5513ACcPmjH+WVvPHOPvqIAQU8tk10IoPObUEO2Wl24WMWmrzT8NderXtk/+4BkLxMsMYGEn70sAmVvg9SbG+LdEym1QUwWb5jknPkd56+IvonEh0WZWppUJ4ORBUylVEoBohnclgH3L4K0fmNvapsb4O8rWy5Qa2PSmNR13aYtNs0HdlaPwfnU1gaziihLQok3yngSw5T34z3XNj/FviWGzTP2W3f9z3j7tUV4EB9dA73bQ/NOeWF0VNDPVDCro2M+6GIRX8PwEoDWsfh4+udP+Mf6O6j0JOnR3/8zg9FW1i79IAmhTYpLMBYVVkwwzU83MUHtHxIl2y7MTQE01/O/nsPz3jo3xd5SPLwy7zVyNn9jl/P03JW0xBIRD9zbeMdre1I0EyrOgGUhryNwqzT/CLp6bAOrG+G98w/Ex/i0x+CbwDXTfXUDDxV/kSq1tsbIoXN4BKM+XCqDCLp6ZAEryYN7UM+v4+7g41FCbWWhk63woK3DtscAsZF6YKc0/bVF0T/PVio5gmQEsHGDXWVUpNUUptUcptU8p9Wgj79+ilMpWSm2p/TerwXs3K6X21v67udmDVVc4Z4x/SwybDRVF7qkPtHex+doWFn8RZ/IPNmtOWNERnJlq7mQ7Xuj+Ywuv02wCUEr5Ai8DVwD9gBuUUo0NL3hfaz2o9t/rtZ+NBn4DXAwMB36jlIo67wFz0pw3xt9R8UPNlZM76gOlLTbHCu/k2uMIa1hVFC5rq6n/781rSgu3secOYDiwT2t9QGtdAcwH7D0zTwa+0lrnaa1PAl8BU84fkZ/zxvi3xLDZkL373NLFzlScY5ZJ7HP+H4XwYrbe7l8fuKbGTJSU5h9hJ3sSQFfgSIPnGbWvne1apdQ2pdQCpVQ3Rz6rlLpdKbVRKbUxG5tzx/g7asCPzGLmG1y4ZOS+pbTJxV/EabYk0xlbnOO+Y+bth4pCSQDCbvYkgMbq9J59WfMZkKC1HggsBerqKtjzWbTWc7TWKVrrlNhYi9fD9Q+urQ/0ualb7wppi80qWTJSo+1qWBTOXTJlBrBwjD0JIAPo1uB5PHDGmVFrnau1Lq99+how1N7PeqSUn4KuMUsVOlt1lVkjt/ck149sEtaxoihcZqpZkjLGwjto4VXsOQNtAHorpRKVUgHATGBhww2UUg2n5k4F6mZTLQYmKaWiajt/J9W+5tmiE83qXJvegqoK5+77yHpTErmPNP+0aZHdTT1+dyeAzsnet5aysEyzCUBrXQXcgzlx7wI+0FrvVEo9pZSaWrvZfUqpnUqprcB9wC21n80Dfo9JIhuAp2pf83zDZkPRcdj9mXP3m7bInBh6jnPufoVn8fE18wHcNRegptqMAJLmH+EAuy4VtNZfAF+c9dqTDR4/BjzWxGfnAnNbEaM1kiaYwnPfvQ4DrnXefvcugR6XQFCE8/YpPFNMb/fdAeTug8piWQNYOEQaoZvi42P6Ag5/C8d3OmefJw+ZIaYy+7d9sPUypRlqql1/LJkBLFpAEsD5DP4x+AXBd04aElq/+IskgHbBlmRmtucfaX7b1spMBf8QiOnj+mOJNkMSwPmERJt1g7d9YDpuWyttsWkXtnKtWOE+9UXh3NAMlLkF4i4yfQ9C2EkSQHOG/dS0rW6d37r9VBRD+mq5+m9P6ucCuDgBVFfBsW0yr0Q4TBJAc7oOga5DW18fKH01VJdL+397EhoLgRGuTwA5aVBZIu3/wmGSAOwxbLb5I0tf1fJ9pC2GgDCzqploH5Ryz/KQ0gEsWkgSgD36XwMhtpZ3BmttOoB7jpXFX9obdywQn7XFXFzYpG9JOEYSgD38g2DwT2DPF5Cf4fjnj++EgqPS/NMe2ZLMKKDKUtcdIzO1tgNY/pyFY+Q3xl4pt5kr+Y1vOv7ZusVfpPpn+2PrBWgzH8AVqivN6nLS/CNaQBKAvaJ6mPr9m+dBVXnz2zeUtthcoYV3dk1swnO5eiRQ9m6oKpMEIFpEEoAjhs8yq5V9v7D5beuU5EHGBln8pb1ydVXQuhLQMgRUtIAkAEf0HG8mcjmyWMy+paa0tIz/b58CwyGss+s6gjNTzVDTuoXohXCAJABH1NUHOrIesrbZ95m0xWY8uNyit1+uLAonHcCiFeS3xlGDbzSLbmx4vfltq6vMHUDSRPkDbc9ctUB8VYUZYSYVQEULyVnJUcFRkDwdtn8IpafOv23GBig7JYu/tHe2JCjJNf1BzpS9y8wul7tL0UKSAFpi+Gwz9X7Lf86/Xdoi8PGDXuPdE5fwTPUjgZzcDyAzgEUrSQJoibiLIH64aQaqqWl6u71LoPtICOrgvtiE53HVUNDMVPO7FZXo3P2KdkMSQEsNnw15++HAisbfP3UETnwvs38FRPYA5euCBLDFDP9Uyrn7Fe2GJICW6jcNQmKa7gyun/0rCaDd8wswy4s6MwFUldd2AEvzj2g5SQAt5RcIQ2827fynDp/7ftoS80cf09vtoQkP5OyicMd3Qk2lJADRKpIAWmPorebrxrPWvK8oMaWje0+W23Nh2JJMk+H5+owckVU7A1iGgIpWkATQGpHdoM8VsPltqCw7/frBNaY+iwz/FHVsvczIscJM5+wvM9UMSY7s4Zz9iXZJEkBrDZ9lxnh//8np19IWg38o9LjUuriEZ3H2SKDMVNP8I3eYohUkAbRW4ljzx123WEzDxV/8gywMTHgUZyaAyjI4sUva/0WrSQJoLR8fGDYLjm40V2UndpkFQKT5RzQU0QX8Q5zTEXx8J9RUSQVQ0WqSAJzhohvMH/eG12XxF9E4pZxXEyhzs/kqdwCilfysDqBNCI6EgdfB1vnmVr9zsrniE6IhWxJkbW39fjK3mDkoHeJbvy/RrskdgLMMm21G/hzfIYu/iMbZkuDkIVPFszWytpjhn9IBLFpJEoCzdB5g6v6AzP4VjbMlga6Gkwdbvo+KEukAFk4jCcCZxj8BA2dC1yFWRyI8kTNGAh3fYZKIJADhBNIH4EwJo8w/IRrjjPWBpQS0cCK5AxDCXYKjTOdtqxLAFgjtCOFxzotLtFt2JQCl1BSl1B6l1D6l1KPn2W66UkorpVJqn/srpeYppbYrpXYppR5zVuBCeKXWFoWTGcDCiZpNAEopX+Bl4AqgH3CDUqpfI9uFA/cB6xu8PAMI1FonA0OBO5RSCa0PWwgvZUtq+R1AeRHk7JHmH+E09twBDAf2aa0PaK0rgPnAtEa2+z3wHNCgKhoaCFVK+QHBQAVQ0LqQhfBitl5QdAzKWvBncGw76BqpACqcxp4E0BU40uB5Ru1r9ZRSg4FuWuvPz/rsAqAYyAIOA89rrc9ZGVspdbtSaqNSamN2drYj8QvhXepGAuW1oBmorgS0lIAQTmJPAmissVHXv6mUD/A34OeNbDccqAa6AInAz5VSPc/ZmdZztNYpWuuU2NhYuwIXwivVLRDUkn6AzFTT+RshHcDCOewZBpoBdGvwPB5oWNQ8HBgArFSmY6ozsFApNRX4P2CR1roSOKGU+gZIAQ44IXYhvE9UIqBa1g9Q1wEshJPYcwewAeitlEpUSgUAM4GFdW9qrfO11jFa6wStdQKwDpiqtd6IafYZr4xQYASw2+nfhRDewj/ILCTkaAIoL4ScvdL8I5yq2QSgta4C7gEWA7uAD7TWO5VST9Ve5Z/Py0AYsAOTSN7UWm9rZcxCeLeWjATK2gZouQMQTmXXTGCt9RfAF2e99mQT245t8LgIMxRUCFHHlmQqx2pt/3j++hnAcgcgnEdmAgvhbrYkKC+AohP2fyYzFSK6QlhH18Ul2h1JAEK4W0uKwmVtkeYf4XSSAIRwN0cTQFm+2Vaaf4STSQIQwt06xINvoP0JoG4VMbkDEE4mCUAId/Pxheie9k8Gy6ybASwJQDiXJAAhrODIAvGZqdChO4TaXBuTaHckAQhhBVsS5B2A6qrmt81MlfZ/4RKSAISwQkxvqKmE/MPn3670JJxMlwQgXEISgBBWqB8J1Ew/gHQACxeSBCCEFewdClo3A1hqAAkXkAQghBVCbBDUwb4EEJUAIdFuCUu0L5IAhLCCUvYVhcvcIlf/wmUkAQhhFVsS5JwnAZTkwalD0v4vXEYSgBBWsfWGggyoKGn8/foKoJIAhGtIAhDCKrZe5mteEwvk1XcAX+SeeES7IwlACKs0NxIoa4spGREc6b6YRLsiCUAIq0T3NF+bSgCZUgJauJYkACGsEhgG4V0anwxWlA35RyQBCJeSBCCElZoqCpdVVwFUhoAK15EEIISVYnpD7t5zX68vAS0dwMJ1JAEIYSVbkin4VpJ35uuZqWaYaFCENXGJdkESgBBWamokkJSAFm4gCUAIKzWWAAqPQ2GmdAALl5MEIISVIruDj9+ZCaCuA1gSgHAxSQBCWMnX31T7bJgAMlMBBZ0HWhWVaCckAQhhtbOLwmWmQkwfM09ACBeSBCCE1WxJkLcfamrMc5kBLNzEz+oAhGj3bElQVQYFR01/QNExSQDCLSQBCGG1hiOBKkvNYxkCKtxAEoAQVmuYAIqzQflA52RrYxLtgiQAIawW3hn8Q01RuLz9ENsXAkKtjkq0A9IJLITVlKotCre3dgawtP8L97ArASilpiil9iil9imlHj3PdtOVUlopldLgtYFKqbVKqZ1Kqe1KqSBnBC5EmxLTG458Z5qApAKocJNmE4BSyhd4GbgC6AfcoJTq18h24cB9wPoGr/kB/wbu1Fr3B8YClU6JXIi2xJYE5QXmsdwBCDex5w5gOLBPa31Aa10BzAemNbLd74HngLIGr00CtmmttwJorXO11tWtjFmItqeuI1j5QucB1sYi2g17EkBX4EiD5xm1r9VTSg0GummtPz/rs30ArZRarJTarJT6ZWMHUErdrpTaqJTamJ2d7UD4QrQRdQvEd7wQ/IOtjUW0G/YkANXIa7r+TaV8gL8BP29kOz/gUuDG2q/XKKUuP2dnWs/RWqdorVNiY2PtClyINiW6NgHI+H/hRvYMA80AujV4Hg9kNngeDgwAViqlADoDC5VSU2s/u0prnQOglPoCGAIsa33oQrQhwZEw4bfQ65zrIyFcxp47gA1Ab6VUolIqAJgJLKx7U2udr7WO0VonaK0TgHXAVK31RmAxMFApFVLbITwG+N7p34UQbcGlD0KcVAAV7tNsAtBaVwH3YE7mu4APtNY7lVJP1V7ln++zJ4G/YpLIFmCz1vp/rQ9bCCFEaymtdfNbuVFKSoreuHGj1WEIIYRXUUpt0lqnNL/laTITWAgh2ilJAEII0U5JAhBCiHZKEoAQQrRTkgCEEKKdkgQghBDtlMcNA1VKFQJ7rI6jFWKAHKuDaAWJ31oSv3W8OXaAC7TW4Y58wBNXBNvj6FhWT6KU2ijxW0fit5Y3x+/NsYOJ39HPSBOQEEK0U5IAhBCinfLEBDDH6gBaSeK3lsRvLW+O35tjhxbE73GdwEIIIdzDE+8AhBBCuIFHJQCl1BSl1B6l1D6l1KNWx+MIpVQ3pdQKpdQupdROpdT9VsfkKKWUr1IqVSl19tKeHk8pFamUWqCU2l37fzDS6pgcoZR6sPb3ZodS6j2lVJDVMZ2PUmquUuqEUmpHg9eilVJfKaX21n6NsjLG82ki/j/X/v5sU0p9rJSKtDLG82ks/gbvPayU0kqpmOb24zEJQCnlC7wMXAH0A25QSvWzNiqHVAE/11pfCIwA7vay+AHux6z54I3+DizSWvcFLsKLvg+lVFfgPiBFaz0A8MUsvOTJ3gKmnPXao8AyrXVvzKp/nnwR9xbnxv8VMEBrPRBIAx5zd1AOeItz40cp1Q2YCBy2ZycekwCA4cA+rfUBrXUFMB+YZnFMdtNaZ2mtN9c+LsScgLpaG5X9lFLxwA+A162OxVFKqQhgNPAGgNa6Qmt9ytqoHOYHBNeunBfCmcuuehyt9Wog76yXpwHzah/PA37o1qAc0Fj8WusltQtggVnZMN7tgdmpiZ8/mPXZf0mDdafnHskAAAJkSURBVNvPx5MSQFfgSIPnGXjRCbQhpVQCMBhYb20kDnkB84tTY3UgLdATyAberG3Cel0pFWp1UPbSWh8FnsdctWUB+VrrJdZG1SKdtNZZYC6IgI4Wx9MatwFfWh2EI2pXaDyqtd5q72c8KQGoRl7zuiFKSqkw4CPgAa11gdXx2EMpdRVwQmu9yepYWsgPGAK8orUeDBTj2c0PZ6htK58GJAJdgFCl1I+tjar9Ukr9CtOk+67VsdhLKRUC/Ap40pHPeVICyAC6NXgej4ffBp9NKeWPOfm/q7X+r9XxOGAUMFUpdRDT9DZeKfVva0NySAaQobWuu+NagEkI3mICkK61ztZaVwL/BS6xOKaWOK6UigOo/XrC4ngcppS6GbgKuFF71xj5XpgLiK21f8fxwGalVOfzfciTEsAGoLdSKlEpFYDpBFtocUx2U0opTBv0Lq31X62OxxFa68e01vFa6wTMz3251tprrkC11seAI0qpC2pfuhz43sKQHHUYGKGUCqn9PbocL+rEbmAhcHPt45uBTy2MxWFKqSnAI8BUrXWJ1fE4Qmu9XWvdUWudUPt3nAEMqf3baJLHJIDazpd7gMWYX/4PtNY7rY3KIaOAn2CunrfU/rvS6qDakXuBd5VS24BBwDMWx2O32juXBcBmYDvm79KjZ6Uqpd4D1gIXKKUylFI/BZ4FJiql9mJGojxrZYzn00T8LwHhwFe1f7+vWhrkeTQRv+P78a67HCGEEM7iMXcAQggh3EsSgBBCtFOSAIQQop2SBCCEEO2UJAAhhGinJAEIIUQ7JQlACCHaKUkAQgjRTv0/fznZO7LjkOcAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "history_df = pd.DataFrame(history.history)\n",
    "history_df[['loss', 'val_loss']].plot()\n",
    "history_df[['dice_coef', 'val_dice_coef']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.load_weights('res34_unet_unfreeze.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "500/500 [==============================] - 52s 105ms/step\n",
      "500/500 [==============================] - 52s 104ms/step\n",
      "500/500 [==============================] - 52s 104ms/step\n",
      "500/500 [==============================] - 52s 103ms/step\n",
      "500/500 [==============================] - 52s 104ms/step\n",
      "500/500 [==============================] - 52s 104ms/step\n",
      "500/500 [==============================] - 51s 102ms/step\n",
      "198/198 [==============================] - 20s 103ms/step\n"
     ]
    }
   ],
   "source": [
    "\n",
    "test_df = []\n",
    "\n",
    "for i in range(0, test_imgs.shape[0], 500):\n",
    "    batch_idx = list(\n",
    "        range(i, min(test_imgs.shape[0], i + 500))\n",
    "    )\n",
    "\n",
    "    test_generator = DataGenerator(\n",
    "        batch_idx,\n",
    "        df=test_imgs,\n",
    "        shuffle=False,\n",
    "        mode='predict',\n",
    "        dim=(350, 525),\n",
    "        reshape=(320, 480),\n",
    "        n_channels=3,\n",
    "        base_path=f'{data_dir}test_images',\n",
    "        target_df=sub_df,\n",
    "        batch_size=1,\n",
    "        n_classes=4,\n",
    "        gamma = 0.8\n",
    "    )\n",
    "\n",
    "    batch_pred_masks = model.predict_generator(\n",
    "        test_generator, \n",
    "        workers=1,\n",
    "        verbose=1\n",
    "    )\n",
    "\n",
    "    for j, b in enumerate(batch_idx):\n",
    "        filename = test_imgs['ImageId'].iloc[b]\n",
    "        image_df = sub_df[sub_df['ImageId'] == filename].copy()\n",
    "\n",
    "        pred_masks = batch_pred_masks[j, ].round().astype(int)\n",
    "        pred_rles = build_rles(pred_masks, reshape=(350, 525))\n",
    "\n",
    "        image_df['EncodedPixels'] = pred_rles\n",
    "        test_df.append(image_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_df = pd.concat(test_df)\n",
    "test_df.drop(columns='ImageId', inplace=True)\n",
    "test_df.to_csv('res32-unet.csv', index=False)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}