DeepCell_tn_nuclear_K1a.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "330e05e3",
   "metadata": {},
   "source": [
    "**Further train DeepCell Models with K's Data and Make Predictions**<br>\n",
    "Starting from tn nuclear, further train with K's data cut into 7x4=28 512x512 patches.<br>\n",
    "A total of 12,574 training instances. 200 epochs, semanticdatagenerator, not resized<br>\n",
    "Best mAP 0.37 at mpp=1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "a72e9c51",
   "metadata": {},
   "outputs": [],
   "source": [
    "starting_model_path = 'tn1.0_nuclear_20221102.h5' # the model trained with nucleus data from the Tissuenet V1.0 dataset\n",
    "model_name = 'tn1.0_nuclear_K_512x512'\n",
    "model_path = '{}.h5'.format(model_name)\n",
    "epochs=200 # 20 sec per step \n",
    "INPUT_PATH=\"images/training_512x512/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "90f2183c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.7.1\n",
      "0.11.1\n"
     ]
    }
   ],
   "source": [
    "import syotil\n",
    "\n",
    "import numpy as np\n",
    "from skimage import io\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline\n",
    "from timeit import default_timer\n",
    "import os\n",
    "from skimage import img_as_ubyte\n",
    "\n",
    "import deepcell\n",
    "from deepcell import image_generators\n",
    "from deepcell.utils.train_utils import rate_scheduler, get_callbacks, count_gpus\n",
    "from deepcell_toolbox.utils import resize, tile_image, untile_image\n",
    "from deepcell_toolbox.deep_watershed import deep_watershed\n",
    "from deepcell.losses import weighted_categorical_crossentropy\n",
    "from deepcell.model_zoo.panopticnet import PanopticNet\n",
    "\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras.optimizers import SGD, Adam\n",
    "from tensorflow.keras.losses import MSE\n",
    "\n",
    "print(tf.__version__)\n",
    "print(deepcell.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "938b06af",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "28\n",
      "['images/training_512x512/M872956_JML_Position10_CD3_train_512x512patch3_img.png', 'images/training_512x512/M872956_JML_Position8_CD8_train_512x512patch1_img.png', 'images/training_512x512/M872956_JML_Position9_CD3_train_512x512patch2_img.png', 'images/training_512x512/M872956_JML_Position8_CD4_train_512x512patch4_img.png', 'images/training_512x512/M872956_JML_Position8_CD8_train_512x512patch4_img.png', 'images/training_512x512/M872956_JML_Position9_CD3_train_512x512patch4_img.png', 'images/training_512x512/M926910_CFL_Position13_CD3_train_512x512patch4_img.png', 'images/training_512x512/M872956_JML_Position8_CD4_train_512x512patch3_img.png', 'images/training_512x512/M872956_JML_Position8_CD4_train_512x512patch1_img.png', 'images/training_512x512/M926910_CFL_Position7_CD3_train_512x512patch1_img.png', 'images/training_512x512/M872956_JML_Position10_CD3_train_512x512patch2_img.png', 'images/training_512x512/M872956_JML_Position8_CD4_train_512x512patch2_img.png', 'images/training_512x512/M926910_CFL_Position7_CD3_train_512x512patch2_img.png', 'images/training_512x512/M872956_JML_Position10_CD3_train_512x512patch4_img.png', 'images/training_512x512/M872956_JML_Position10_CD3_train_512x512patch1_img.png', 'images/training_512x512/M926910_CFL_Position13_CD3_train_512x512patch3_img.png', 'images/training_512x512/M926910_CFL_Position7_CD3_train_512x512patch3_img.png', 'images/training_512x512/M872956_JML_Position9_CD3_train_512x512patch1_img.png', 'images/training_512x512/M872956_JML_Position8_CD3_train_512x512patch3_img.png', 'images/training_512x512/M872956_JML_Position8_CD3_train_512x512patch1_img.png', 'images/training_512x512/M872956_JML_Position8_CD8_train_512x512patch3_img.png', 'images/training_512x512/M872956_JML_Position8_CD3_train_512x512patch4_img.png', 'images/training_512x512/M872956_JML_Position8_CD8_train_512x512patch2_img.png', 'images/training_512x512/M926910_CFL_Position13_CD3_train_512x512patch2_img.png', 'images/training_512x512/M926910_CFL_Position7_CD3_train_512x512patch4_img.png', 'images/training_512x512/M926910_CFL_Position13_CD3_train_512x512patch1_img.png', 'images/training_512x512/M872956_JML_Position9_CD3_train_512x512patch3_img.png', 'images/training_512x512/M872956_JML_Position8_CD3_train_512x512patch2_img.png']\n"
     ]
    }
   ],
   "source": [
    "import glob\n",
    "FILENAMES = glob.glob(INPUT_PATH+\"*_img.png\")\n",
    "print(len(FILENAMES))\n",
    "print(FILENAMES)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "1bac06f7",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-03 20:57:43.087132: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 14657 MB memory:  -> device: 0, name: Tesla V100-PCIE-16GB, pci bus id: 0000:04:00.0, compute capability: 7.0\n",
      "2022-11-03 20:57:43.088589: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:1 with 14657 MB memory:  -> device: 1, name: Tesla V100-PCIE-16GB, pci bus id: 0000:06:00.0, compute capability: 7.0\n",
      "2022-11-03 20:57:43.089800: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:2 with 14657 MB memory:  -> device: 2, name: Tesla V100-PCIE-16GB, pci bus id: 0000:08:00.0, compute capability: 7.0\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(28, 512, 512, 1)\n",
      "(28, 512, 512, 1)\n"
     ]
    }
   ],
   "source": [
    "imgs = [io.imread(CURR_IM_NAME)[:,:,0] for CURR_IM_NAME in FILENAMES]\n",
    "X_train = tf.stack(imgs)\n",
    "X_train = np.expand_dims(X_train, axis=-1)\n",
    "print(X_train.shape)\n",
    "\n",
    "masks = [io.imread(CURR_IM_NAME.replace(\"img\",\"masks\")) for CURR_IM_NAME in FILENAMES]\n",
    "y_train = tf.stack(masks)\n",
    "y_train = np.expand_dims(y_train, axis=-1)\n",
    "print(y_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "73d0c5d8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "50193\n"
     ]
    }
   ],
   "source": [
    "# create dataloader\n",
    "\n",
    "seed=0\n",
    "min_objects = 2\n",
    "\n",
    "transforms = ['inner-distance', 'outer-distance', 'fgbg']\n",
    "transforms_kwargs = {'outer-distance': {'erosion_width': 0}}\n",
    "\n",
    "# use augmentation for training but not validation\n",
    "# no need to define width_shift_range or height_shift_range \n",
    "# because crops will be taken randomly over the image and not just the center (cropping.py)\n",
    "# datagen = image_generators.CroppingDataGenerator(\n",
    "#     crop_size=(512, 512),\n",
    "datagen = image_generators.SemanticDataGenerator(\n",
    "    rotation_range=180,\n",
    "    fill_mode='reflect',\n",
    "    zoom_range=(0.75, 1.25),\n",
    "    horizontal_flip=True,\n",
    "    vertical_flip=True)\n",
    "\n",
    "datagen_val = image_generators.CroppingDataGenerator()\n",
    "\n",
    "batch_size = 4 # 8 causes memory outage\n",
    "\n",
    "train_data = datagen.flow(\n",
    "    {'X': X_train, 'y': y_train},\n",
    "    seed=seed,\n",
    "    transforms=transforms,\n",
    "    transforms_kwargs=transforms_kwargs,\n",
    "    min_objects=min_objects,\n",
    "    batch_size=batch_size)\n",
    "\n",
    "val_data = None\n",
    "# datagen_val.flow(\n",
    "#     {'X': X_val, 'y': y_val},\n",
    "#     seed=seed,\n",
    "#     transforms=transforms,\n",
    "#     transforms_kwargs=transforms_kwargs,\n",
    "#     min_objects=min_objects,\n",
    "#     batch_size=batch_size)\n",
    "\n",
    "# get number of training and validation instances\n",
    "\n",
    "cnts_train = [np.max(y_train[i,...]) for i in range(y_train.shape[0])]\n",
    "print(np.sum(cnts_train)) # total number of training instances"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "0b79383a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/app/software/scikit-image/0.19.1-foss-2021b/lib/python3.9/site-packages/skimage/io/_plugins/matplotlib_plugin.py:150: UserWarning: Low image data range; displaying image with stretched contrast.\n",
      "  lo, hi, cmap = _get_display_range(image)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "i=5\n",
    "plt.subplot(1, 2, 1) # row 1, col 2 index 1\n",
    "io.imshow(X_train[i,:,:,0])\n",
    "plt.subplot(1, 2, 2) # row 1, col 2 index 1\n",
    "# tmp = syotil.masks_to_outlines(y_train[i,:,:,0]); io.imshow(tmp)\n",
    "io.imshow(y_train[i,:,:,0])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9591736f",
   "metadata": {},
   "source": [
    "**The two cells below define and train the model.** They can be skipped if a trained model will be loaded."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "7ebc07f0",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-03 10:20:21.637819: W tensorflow/python/util/util.cc:368] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.\n",
      "/app/software/TensorFlow/2.7.1-foss-2021b-CUDA-11.4.1/lib/python3.9/site-packages/keras/optimizer_v2/adam.py:105: UserWarning: The `lr` argument is deprecated, use `learning_rate` instead.\n",
      "  super(Adam, self).__init__(name, **kwargs)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[('semantic_0', (None, 512, 512, 1)),\n",
       " ('semantic_1', (None, 512, 512, 1)),\n",
       " ('semantic_2', (None, 512, 512, 2))]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "semantic_classes = [1, 1, 2] # inner distance, outer distance, fgbg\n",
    "\n",
    "model = PanopticNet(\n",
    "    backbone='resnet50',\n",
    "    input_shape=(512,512,1),\n",
    "    norm_method='whole_image',\n",
    "    num_semantic_classes=semantic_classes)\n",
    "\n",
    "lr = 1e-4\n",
    "optimizer = Adam(lr=lr, clipnorm=0.001)\n",
    "lr_sched = rate_scheduler(lr=lr, decay=0.99)\n",
    "\n",
    "# Create a dictionary of losses for each semantic head\n",
    "\n",
    "def semantic_loss(n_classes):\n",
    "    def _semantic_loss(y_pred, y_true):\n",
    "        if n_classes > 1:\n",
    "            return 0.01 * weighted_categorical_crossentropy(\n",
    "                y_pred, y_true, n_classes=n_classes)\n",
    "        return MSE(y_pred, y_true)\n",
    "    return _semantic_loss\n",
    "\n",
    "loss = {}\n",
    "\n",
    "# Give losses for all of the semantic heads\n",
    "for layer in model.layers:\n",
    "    if layer.name.startswith('semantic_'):\n",
    "        n_classes = layer.output_shape[-1]\n",
    "        loss[layer.name] = semantic_loss(n_classes)\n",
    "        \n",
    "model.compile(loss=loss, optimizer=optimizer)\n",
    "\n",
    "model.load_weights(starting_model_path, by_name=True)\n",
    "\n",
    "[(layer.name, layer.output_shape) for layer in filter(lambda x: x.name.startswith('semantic_'), model.layers)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "b1fab4f5",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-03 10:20:25.664503: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /device:GPU:0 with 10429 MB memory:  -> device: 0, name: NVIDIA GeForce GTX 1080 Ti, pci bus id: 0000:5e:00.0, compute capability: 6.1\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training on 1 GPUs.\n",
      "Epoch 1/200\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-03 10:20:39.503155: I tensorflow/stream_executor/cuda/cuda_dnn.cc:366] Loaded cuDNN version 8202\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7/7 [==============================] - ETA: 0s - loss: 0.0239 - semantic_0_loss: 0.0093 - semantic_1_loss: 0.0122 - semantic_2_loss: 0.0024WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 25s 1s/step - loss: 0.0239 - semantic_0_loss: 0.0093 - semantic_1_loss: 0.0122 - semantic_2_loss: 0.0024 - lr: 1.0000e-04\n",
      "Epoch 2/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0213 - semantic_0_loss: 0.0085 - semantic_1_loss: 0.0110 - semantic_2_loss: 0.0018WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0213 - semantic_0_loss: 0.0085 - semantic_1_loss: 0.0110 - semantic_2_loss: 0.0018 - lr: 9.9000e-05\n",
      "Epoch 3/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0205 - semantic_0_loss: 0.0081 - semantic_1_loss: 0.0106 - semantic_2_loss: 0.0017WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0205 - semantic_0_loss: 0.0081 - semantic_1_loss: 0.0106 - semantic_2_loss: 0.0017 - lr: 9.8010e-05\n",
      "Epoch 4/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0202 - semantic_0_loss: 0.0081 - semantic_1_loss: 0.0104 - semantic_2_loss: 0.0017WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0202 - semantic_0_loss: 0.0081 - semantic_1_loss: 0.0104 - semantic_2_loss: 0.0017 - lr: 9.7030e-05\n",
      "Epoch 5/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0200 - semantic_0_loss: 0.0080 - semantic_1_loss: 0.0103 - semantic_2_loss: 0.0017WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0200 - semantic_0_loss: 0.0080 - semantic_1_loss: 0.0103 - semantic_2_loss: 0.0017 - lr: 9.6060e-05\n",
      "Epoch 6/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0199 - semantic_0_loss: 0.0079 - semantic_1_loss: 0.0103 - semantic_2_loss: 0.0017WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0199 - semantic_0_loss: 0.0079 - semantic_1_loss: 0.0103 - semantic_2_loss: 0.0017 - lr: 9.5099e-05\n",
      "Epoch 7/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0194 - semantic_0_loss: 0.0077 - semantic_1_loss: 0.0100 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0194 - semantic_0_loss: 0.0077 - semantic_1_loss: 0.0100 - semantic_2_loss: 0.0016 - lr: 9.4148e-05\n",
      "Epoch 8/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0193 - semantic_0_loss: 0.0077 - semantic_1_loss: 0.0099 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0193 - semantic_0_loss: 0.0077 - semantic_1_loss: 0.0099 - semantic_2_loss: 0.0016 - lr: 9.3207e-05\n",
      "Epoch 9/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0190 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0190 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016 - lr: 9.2274e-05\n",
      "Epoch 10/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0185 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0096 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0185 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0096 - semantic_2_loss: 0.0016 - lr: 9.1352e-05\n",
      "Epoch 11/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0189 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0099 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0189 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0099 - semantic_2_loss: 0.0016 - lr: 9.0438e-05\n",
      "Epoch 12/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0191 - semantic_0_loss: 0.0077 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0191 - semantic_0_loss: 0.0077 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016 - lr: 8.9534e-05\n",
      "Epoch 13/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0187 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0096 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0187 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0096 - semantic_2_loss: 0.0016 - lr: 8.8638e-05\n",
      "Epoch 14/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0191 - semantic_0_loss: 0.0076 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0191 - semantic_0_loss: 0.0076 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016 - lr: 8.7752e-05\n",
      "Epoch 15/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0185 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0096 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0185 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0096 - semantic_2_loss: 0.0016 - lr: 8.6875e-05\n",
      "Epoch 16/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0191 - semantic_0_loss: 0.0076 - semantic_1_loss: 0.0099 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0191 - semantic_0_loss: 0.0076 - semantic_1_loss: 0.0099 - semantic_2_loss: 0.0016 - lr: 8.6006e-05\n",
      "Epoch 17/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0188 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0188 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016 - lr: 8.5146e-05\n",
      "Epoch 18/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0188 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0097 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0188 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0097 - semantic_2_loss: 0.0016 - lr: 8.4294e-05\n",
      "Epoch 19/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0189 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0189 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0098 - semantic_2_loss: 0.0016 - lr: 8.3451e-05\n",
      "Epoch 20/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0183 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0183 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0016 - lr: 8.2617e-05\n",
      "Epoch 21/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0182 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0182 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0015 - lr: 8.1791e-05\n",
      "Epoch 22/200\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7/7 [==============================] - ETA: 0s - loss: 0.0185 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0185 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0016 - lr: 8.0973e-05\n",
      "Epoch 23/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0187 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0097 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0187 - semantic_0_loss: 0.0075 - semantic_1_loss: 0.0097 - semantic_2_loss: 0.0016 - lr: 8.0163e-05\n",
      "Epoch 24/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0182 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0182 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015 - lr: 7.9361e-05\n",
      "Epoch 25/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0183 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0183 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0015 - lr: 7.8568e-05\n",
      "Epoch 26/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0185 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0096 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0185 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0096 - semantic_2_loss: 0.0015 - lr: 7.7782e-05\n",
      "Epoch 27/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0181 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0181 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015 - lr: 7.7004e-05\n",
      "Epoch 28/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 7.6234e-05\n",
      "Epoch 29/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0185 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0016WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0185 - semantic_0_loss: 0.0074 - semantic_1_loss: 0.0095 - semantic_2_loss: 0.0016 - lr: 7.5472e-05\n",
      "Epoch 30/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0179 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0179 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 7.4717e-05\n",
      "Epoch 31/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0183 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0183 - semantic_0_loss: 0.0073 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015 - lr: 7.3970e-05\n",
      "Epoch 32/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0181 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0181 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015 - lr: 7.3230e-05\n",
      "Epoch 33/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0179 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0179 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 7.2498e-05\n",
      "Epoch 34/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015 - lr: 7.1773e-05\n",
      "Epoch 35/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0179 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0092 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0179 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0092 - semantic_2_loss: 0.0015 - lr: 7.1055e-05\n",
      "Epoch 36/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 7.0345e-05\n",
      "Epoch 37/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0172 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0172 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 6.9641e-05\n",
      "Epoch 38/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 6.8945e-05\n",
      "Epoch 39/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0178 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0092 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0178 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0092 - semantic_2_loss: 0.0015 - lr: 6.8255e-05\n",
      "Epoch 40/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0180 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0180 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 6.7573e-05\n",
      "Epoch 41/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 6.6897e-05\n",
      "Epoch 42/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0181 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0181 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0094 - semantic_2_loss: 0.0015 - lr: 6.6228e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 43/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 6.5566e-05\n",
      "Epoch 44/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 6.4910e-05\n",
      "Epoch 45/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0174 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0174 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015 - lr: 6.4261e-05\n",
      "Epoch 46/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0178 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0092 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0178 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0092 - semantic_2_loss: 0.0015 - lr: 6.3619e-05\n",
      "Epoch 47/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0175 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0175 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015 - lr: 6.2982e-05\n",
      "Epoch 48/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 6.2353e-05\n",
      "Epoch 49/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 6.1729e-05\n",
      "Epoch 50/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 6.1112e-05\n",
      "Epoch 51/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 6.0501e-05\n",
      "Epoch 52/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 5.9896e-05\n",
      "Epoch 53/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0172 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0172 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 5.9297e-05\n",
      "Epoch 54/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0177 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0177 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 5.8704e-05\n",
      "Epoch 55/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0177 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0177 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 5.8117e-05\n",
      "Epoch 56/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0178 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0092 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0178 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0092 - semantic_2_loss: 0.0015 - lr: 5.7535e-05\n",
      "Epoch 57/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0174 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0174 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 5.6960e-05\n",
      "Epoch 58/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0181 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0181 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 5.6391e-05\n",
      "Epoch 59/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 5.5827e-05\n",
      "Epoch 60/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 5.5268e-05\n",
      "Epoch 61/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 5.4716e-05\n",
      "Epoch 62/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015 - lr: 5.4169e-05\n",
      "Epoch 63/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 5.3627e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 64/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 5.3091e-05\n",
      "Epoch 65/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 5.2560e-05\n",
      "Epoch 66/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 5.2034e-05\n",
      "Epoch 67/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0180 - semantic_0_loss: 0.0072 - semantic_1_loss: 0.0093 - semantic_2_loss: 0.0015 - lr: 5.1514e-05\n",
      "Epoch 68/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 5.0999e-05\n",
      "Epoch 69/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 5.0489e-05\n",
      "Epoch 70/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0174 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0174 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0091 - semantic_2_loss: 0.0015 - lr: 4.9984e-05\n",
      "Epoch 71/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0176 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0176 - semantic_0_loss: 0.0071 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015 - lr: 4.9484e-05\n",
      "Epoch 72/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0174 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0174 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 4.8989e-05\n",
      "Epoch 73/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 4.8499e-05\n",
      "Epoch 74/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015 - lr: 4.8014e-05\n",
      "Epoch 75/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0172 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0172 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 4.7534e-05\n",
      "Epoch 76/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 4.7059e-05\n",
      "Epoch 77/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 4.6588e-05\n",
      "Epoch 78/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 4.6122e-05\n",
      "Epoch 79/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0172 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0172 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 4.5661e-05\n",
      "Epoch 80/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 4.5204e-05\n",
      "Epoch 81/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0174 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0174 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015 - lr: 4.4752e-05\n",
      "Epoch 82/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015 - lr: 4.4305e-05\n",
      "Epoch 83/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 4.3862e-05\n",
      "Epoch 84/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015 - lr: 4.3423e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 85/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 4.2989e-05\n",
      "Epoch 86/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0175 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0090 - semantic_2_loss: 0.0015 - lr: 4.2559e-05\n",
      "Epoch 87/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 4.2133e-05\n",
      "Epoch 88/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 4.1712e-05\n",
      "Epoch 89/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 4.1295e-05\n",
      "Epoch 90/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 4.0882e-05\n",
      "Epoch 91/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 4.0473e-05\n",
      "Epoch 92/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0014 - lr: 4.0068e-05\n",
      "Epoch 93/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.9668e-05\n",
      "Epoch 94/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 3.9271e-05\n",
      "Epoch 95/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.8878e-05\n",
      "Epoch 96/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 3.8490e-05\n",
      "Epoch 97/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0174 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0174 - semantic_0_loss: 0.0070 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 3.8105e-05\n",
      "Epoch 98/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 3.7724e-05\n",
      "Epoch 99/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 3.7346e-05\n",
      "Epoch 100/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0014 - lr: 3.6973e-05\n",
      "Epoch 101/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.6603e-05\n",
      "Epoch 102/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0172 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0172 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 3.6237e-05\n",
      "Epoch 103/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 3.5875e-05\n",
      "Epoch 104/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0014 - lr: 3.5516e-05\n",
      "Epoch 105/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 3.5161e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 106/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0014 - lr: 3.4809e-05\n",
      "Epoch 107/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.4461e-05\n",
      "Epoch 108/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 3.4117e-05\n",
      "Epoch 109/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 3.3775e-05\n",
      "Epoch 110/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.3438e-05\n",
      "Epoch 111/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 3.3103e-05\n",
      "Epoch 112/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 3.2772e-05\n",
      "Epoch 113/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 3.2445e-05\n",
      "Epoch 114/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 3.2120e-05\n",
      "Epoch 115/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.1799e-05\n",
      "Epoch 116/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0173 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0089 - semantic_2_loss: 0.0015 - lr: 3.1481e-05\n",
      "Epoch 117/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0170 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.1166e-05\n",
      "Epoch 118/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.0854e-05\n",
      "Epoch 119/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 3.0546e-05\n",
      "Epoch 120/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 3.0240e-05\n",
      "Epoch 121/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015 - lr: 2.9938e-05\n",
      "Epoch 122/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0015 - lr: 2.9639e-05\n",
      "Epoch 123/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 2.9342e-05\n",
      "Epoch 124/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.9049e-05\n",
      "Epoch 125/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.8758e-05\n",
      "Epoch 126/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.8471e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 127/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.8186e-05\n",
      "Epoch 128/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.7904e-05\n",
      "Epoch 129/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.7625e-05\n",
      "Epoch 130/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0069 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 2.7349e-05\n",
      "Epoch 131/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 9s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.7075e-05\n",
      "Epoch 132/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.6805e-05\n",
      "Epoch 133/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.6537e-05\n",
      "Epoch 134/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 2.6271e-05\n",
      "Epoch 135/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 2.6009e-05\n",
      "Epoch 136/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.5748e-05\n",
      "Epoch 137/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.5491e-05\n",
      "Epoch 138/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 2.5236e-05\n",
      "Epoch 139/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.4984e-05\n",
      "Epoch 140/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.4734e-05\n",
      "Epoch 141/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.4487e-05\n",
      "Epoch 142/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.4242e-05\n",
      "Epoch 143/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.3999e-05\n",
      "Epoch 144/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.3759e-05\n",
      "Epoch 145/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.3522e-05\n",
      "Epoch 146/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 2.3286e-05\n",
      "Epoch 147/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.3054e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 148/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 2.2823e-05\n",
      "Epoch 149/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 2.2595e-05\n",
      "Epoch 150/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.2369e-05\n",
      "Epoch 151/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 2.2145e-05\n",
      "Epoch 152/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.1924e-05\n",
      "Epoch 153/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.1704e-05\n",
      "Epoch 154/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.1487e-05\n",
      "Epoch 155/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.1273e-05\n",
      "Epoch 156/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.1060e-05\n",
      "Epoch 157/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.0849e-05\n",
      "Epoch 158/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0160 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0160 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014 - lr: 2.0641e-05\n",
      "Epoch 159/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 2.0434e-05\n",
      "Epoch 160/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 2.0230e-05\n",
      "Epoch 161/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 2.0028e-05\n",
      "Epoch 162/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0163 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 1.9827e-05\n",
      "Epoch 163/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 1.9629e-05\n",
      "Epoch 164/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.9433e-05\n",
      "Epoch 165/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0161 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0082 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0161 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0082 - semantic_2_loss: 0.0014 - lr: 1.9239e-05\n",
      "Epoch 166/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.9046e-05\n",
      "Epoch 167/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.8856e-05\n",
      "Epoch 168/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.8667e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 169/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.8480e-05\n",
      "Epoch 170/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 1.8296e-05\n",
      "Epoch 171/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0163 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0163 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 1.8113e-05\n",
      "Epoch 172/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.7932e-05\n",
      "Epoch 173/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.7752e-05\n",
      "Epoch 174/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.7575e-05\n",
      "Epoch 175/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0167 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0167 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 1.7399e-05\n",
      "Epoch 176/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.7225e-05\n",
      "Epoch 177/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0162 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0162 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014 - lr: 1.7053e-05\n",
      "Epoch 178/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 1.6882e-05\n",
      "Epoch 179/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.6713e-05\n",
      "Epoch 180/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 1.6546e-05\n",
      "Epoch 181/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0171 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0088 - semantic_2_loss: 0.0015 - lr: 1.6381e-05\n",
      "Epoch 182/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0067 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.6217e-05\n",
      "Epoch 183/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.6055e-05\n",
      "Epoch 184/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.5894e-05\n",
      "Epoch 185/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0166 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.5735e-05\n",
      "Epoch 186/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.5578e-05\n",
      "Epoch 187/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0161 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0161 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014 - lr: 1.5422e-05\n",
      "Epoch 188/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.5268e-05\n",
      "Epoch 189/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 1.5115e-05\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 190/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 1.4964e-05\n",
      "Epoch 191/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 1.4814e-05\n",
      "Epoch 192/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0165 - semantic_0_loss: 0.0066 - semantic_1_loss: 0.0085 - semantic_2_loss: 0.0014 - lr: 1.4666e-05\n",
      "Epoch 193/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0160 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0082 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0160 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0082 - semantic_2_loss: 0.0014 - lr: 1.4520e-05\n",
      "Epoch 194/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0161 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0082 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0161 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0082 - semantic_2_loss: 0.0014 - lr: 1.4374e-05\n",
      "Epoch 195/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0164 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0084 - semantic_2_loss: 0.0014 - lr: 1.4231e-05\n",
      "Epoch 196/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0160 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0082 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0160 - semantic_0_loss: 0.0064 - semantic_1_loss: 0.0082 - semantic_2_loss: 0.0014 - lr: 1.4088e-05\n",
      "Epoch 197/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0169 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0087 - semantic_2_loss: 0.0014 - lr: 1.3948e-05\n",
      "Epoch 198/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0168 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0168 - semantic_0_loss: 0.0068 - semantic_1_loss: 0.0086 - semantic_2_loss: 0.0014 - lr: 1.3808e-05\n",
      "Epoch 199/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0161 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0161 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014 - lr: 1.3670e-05\n",
      "Epoch 200/200\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.0162 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014WARNING:tensorflow:Can save best model only with val_loss available, skipping.\n",
      "7/7 [==============================] - 8s 1s/step - loss: 0.0162 - semantic_0_loss: 0.0065 - semantic_1_loss: 0.0083 - semantic_2_loss: 0.0014 - lr: 1.3533e-05\n"
     ]
    }
   ],
   "source": [
    "# fit the model\n",
    "print('Training on', count_gpus(), 'GPUs.')\n",
    "\n",
    "train_callbacks = get_callbacks(\n",
    "    model_path,\n",
    "    lr_sched=lr_sched,\n",
    "#     monitor=\"val_loss\",\n",
    "    # monitor='loss', # training loss\n",
    "    verbose=1)          \n",
    "            \n",
    "loss_history = model.fit(\n",
    "    train_data,\n",
    "    steps_per_epoch=train_data.y.shape[0] // batch_size, # \n",
    "    epochs=epochs, \n",
    "#     validation_data=val_data,\n",
    "#     validation_steps=val_data.y.shape[0] // batch_size,\n",
    "    callbacks=train_callbacks)\n",
    "\n",
    "model.save_weights(model_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "42277d43",
   "metadata": {},
   "source": [
    "**Make prediction on K's data.**<br>\n",
    "Using NuclearSegmentation allows setting image_mpp, which has a substantial influence on performance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "dbf36df1",
   "metadata": {},
   "outputs": [],
   "source": [
    "prediction_model = PanopticNet(\n",
    "    backbone='resnet50',\n",
    "    norm_method='whole_image',\n",
    "    num_semantic_classes=[1, 1], # inner distance, outer distance\n",
    "    input_shape= (512,512,1)\n",
    ")\n",
    "\n",
    "prediction_model.load_weights(model_path, by_name=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "37a766dc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('semantic_0', (None, 512, 512, 1)), ('semantic_1', (None, 512, 512, 1))]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from deepcell.applications import NuclearSegmentation\n",
    "app = NuclearSegmentation(prediction_model)\n",
    "[(layer.name, layer.output_shape) for layer in filter(lambda x: x.name.startswith('semantic_'), app.model.layers)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "20301e2d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/yfong/deeplearning/dense_cell_segmentation\n",
      "['M926910_CFL_Position7_CD3_test_img.png', 'M872956_JML_Position10_CD3_test_img.png', 'M872956_JML_Position8_CD8_test_img.png', 'M926910_CFL_Position13_CD3_test_img.png', 'M872956_JML_Position8_CD3_test_img.png', 'M872956_JML_Position8_CD4_test_img.png', 'M872956_JML_Position9_CD3_test_img.png']\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "print(os.getcwd())\n",
    "INPUT_PATH=\"images/test/\"\n",
    "FILENAMES = [f for f in os.listdir(\"images/training/testimages\")]\n",
    "print(FILENAMES)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "a2eaeaa5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.65 0.75 0.85 0.95 1.05 1.15 1.25]\n",
      "M926910_CFL_Position7_CD3_test_img.png\n",
      "M872956_JML_Position10_CD3_test_img.png\n",
      "M872956_JML_Position8_CD8_test_img.png\n",
      "M926910_CFL_Position13_CD3_test_img.png\n",
      "M872956_JML_Position8_CD3_test_img.png\n",
      "M872956_JML_Position8_CD4_test_img.png\n",
      "M872956_JML_Position9_CD3_test_img.png\n",
      "[0.65 0.75 0.85 0.95 1.05 1.15 1.25]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[5, 4, 2, 4, 2, 1, 4]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# detailed mapping of image_mpp parameter\n",
    "\n",
    "mpps = np.linspace(start=0.65, stop=1.25, num=7)\n",
    "print(mpps)\n",
    "\n",
    "AP_mat=[]\n",
    "for CURR_IM_NAME in FILENAMES:\n",
    "    print(CURR_IM_NAME)\n",
    "    im0 = io.imread(os.path.join(INPUT_PATH, CURR_IM_NAME))\n",
    "    mask_true=io.imread(os.path.join(INPUT_PATH, CURR_IM_NAME.replace(\"img\",\"masks\")))\n",
    "\n",
    "    x = np.expand_dims(im0, axis=-1)\n",
    "    x = np.expand_dims(x, axis=0)\n",
    "    y, tile_info = app._tile_input(x)\n",
    "    AP_arr=[]\n",
    "    for mpp in mpps:\n",
    "        prd = app._untile_output(app.predict(y, image_mpp=mpp), tile_info)[0,:,:,0]\n",
    "        a=syotil.csi(mask_true, prd)\n",
    "        AP_arr.append(a)\n",
    "    AP_mat.append(AP_arr)\n",
    "\n",
    "#AP_mat\n",
    "print(mpps)\n",
    "[x.index(np.max(x)) for x in AP_mat]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "f31fa740",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1, 1040, 233, 1)\n",
      "(1, 1568, 512, 1)\n",
      "{'padding': True, 'x_pad': (264, 264), 'y_pad': (139, 140)}\n",
      "(1, 1040, 233, 1)\n",
      "(1, 1568, 512, 1)\n",
      "{'padding': True, 'x_pad': (264, 264), 'y_pad': (139, 140)}\n",
      "(1, 1040, 233, 1)\n",
      "(1, 1568, 512, 1)\n",
      "{'padding': True, 'x_pad': (264, 264), 'y_pad': (139, 140)}\n",
      "(1, 1040, 233, 1)\n",
      "(1, 1568, 512, 1)\n",
      "{'padding': True, 'x_pad': (264, 264), 'y_pad': (139, 140)}\n",
      "(1, 1040, 233, 1)\n",
      "(1, 1568, 512, 1)\n",
      "{'padding': True, 'x_pad': (264, 264), 'y_pad': (139, 140)}\n",
      "(1, 1040, 233, 1)\n",
      "(1, 1568, 512, 1)\n",
      "{'padding': True, 'x_pad': (264, 264), 'y_pad': (139, 140)}\n",
      "(1, 1040, 233, 1)\n",
      "(1, 1568, 512, 1)\n",
      "{'padding': True, 'x_pad': (264, 264), 'y_pad': (139, 140)}\n",
      "{'M926910_CFL_Position7_CD3_test_img.png': 0.3236641221374046, 'M872956_JML_Position10_CD3_test_img.png': 0.3854166666666667, 'M872956_JML_Position8_CD8_test_img.png': 0.4330357142857143, 'M926910_CFL_Position13_CD3_test_img.png': 0.2606741573033708, 'M872956_JML_Position8_CD3_test_img.png': 0.4029126213592233, 'M872956_JML_Position8_CD4_test_img.png': 0.4395833333333333, 'M872956_JML_Position9_CD3_test_img.png': 0.35660377358490564, 'mAP': 0.37169862695294553}\n"
     ]
    }
   ],
   "source": [
    "mpp_select=1\n",
    "\n",
    "APs={}\n",
    "for CURR_IM_NAME in FILENAMES:\n",
    "    im0 = io.imread(os.path.join(INPUT_PATH, CURR_IM_NAME))\n",
    "    mask_true=io.imread(os.path.join(INPUT_PATH, CURR_IM_NAME.replace(\"img\",\"masks\")))\n",
    "\n",
    "    x = np.expand_dims(im0, axis=-1)\n",
    "    x = np.expand_dims(x, axis=0)\n",
    "    y, tile_info = app._tile_input(x)\n",
    "    print(x.shape)\n",
    "    print(y.shape)\n",
    "    print(tile_info)\n",
    "    pred = app.predict(y, image_mpp=mpp_select)\n",
    "    prd = app._untile_output(pred, tile_info)\n",
    "    #io.imshow(prd[0,:,:,0])\n",
    "    plt.show()\n",
    "    \n",
    "    APs[CURR_IM_NAME] = syotil.csi(mask_true, prd[0,:,:,0])# masks may lose one pixel if dimension is odd pixels\n",
    "\n",
    "APs[\"mAP\"]=np.mean(list(APs.values()))\n",
    "print(APs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "7bc06419",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                         0         1\n",
      "0   M926910_CFL_Position7_CD3_test_img.png  0.323664\n",
      "1  M872956_JML_Position10_CD3_test_img.png  0.385417\n",
      "2   M872956_JML_Position8_CD8_test_img.png  0.433036\n",
      "3  M926910_CFL_Position13_CD3_test_img.png  0.260674\n",
      "4   M872956_JML_Position8_CD3_test_img.png  0.402913\n",
      "5   M872956_JML_Position8_CD4_test_img.png  0.439583\n",
      "6   M872956_JML_Position9_CD3_test_img.png  0.356604\n",
      "7                                      mAP  0.371699\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "df = pd.DataFrame([FILENAMES+[\"mAP\"], list(APs.values())])\n",
    "print(df.transpose())\n",
    "df.to_csv('images/training/csi_tn_'+model_name+'.txt', index=False, header=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "520fe391",
   "metadata": {},
   "source": [
    "**mAP**<br>\n",
    "image_mpp=1:   0.37<br>\n",
    "image_mpp=1.3: 0.30<br>"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}