notebook/2.1.Image_segmentation_train.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "# Image segmentation training - Mask R-CNN\n",
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this notebook, we will show how to train a deep learning model to do instance segmentation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "## Set path and Load library"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "button": false,
    "collapsed": false,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": false
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from mrcnn.main import *\n",
    "# Set path to Mask RCNN folder\n",
    "ROOT_DIR = os.path.abspath(\"/home/ubuntu/github/MaskRCNN\")\n",
    "# Directory to save logs and trained model\n",
    "MODEL_DIR = os.path.join(ROOT_DIR, \"logs\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")\n",
    "%matplotlib inline \n",
    "from mrcnn.config import Config\n",
    "COCO_MODEL_PATH = os.path.join(ROOT_DIR, \"mask_rcnn_coco.h5\")\n",
    "if not os.path.exists(COCO_MODEL_PATH):\n",
    "    utils.download_trained_weights(COCO_MODEL_PATH)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "## Load data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "Input training and validation image path here. Validation dataset may be left blank. Note that all training images must be of equal size and square. The variable 'size' corresponds to the image shape: (size, size, 3). "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [],
   "source": [
    "#Training images should be square\n",
    "size=400\n",
    "\n",
    "images = sorted((glob.glob(\"./MaskRCNN/train/*crop.png\")))\n",
    "labels = sorted((glob.glob(\"./MaskRCNN/train/*label.png\")))\n",
    "\n",
    "images_val = sorted((glob.glob(\"./MaskRCNN/val/*crop.png\")))\n",
    "labels_val = sorted((glob.glob(\"./MaskRCNN/val/*label.png\")))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "If validation dataset does not exist, fill it in with the same variables as the training dataset. Fields cannot be left blank."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "## Set parameters for training"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Key parameter\n",
    "* GPU count\n",
    "* IMAGES_PER_GPU\n",
    "* DETECTION_MAX_INSTANCES: max instance in one image\n",
    "* IMAGE_MIN_DIM: random cropping will be performed on the input image to this size. I recommend 128 for 200X image.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "button": false,
    "collapsed": false,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": false
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Configurations:\n",
      "BACKBONE                       resnet50\n",
      "BACKBONE_STRIDES               [4, 8, 16, 32, 64]\n",
      "BATCH_SIZE                     2\n",
      "BBOX_STD_DEV                   [0.1 0.1 0.2 0.2]\n",
      "COMPUTE_BACKBONE_SHAPE         None\n",
      "DETECTION_MAX_INSTANCES        256\n",
      "DETECTION_MIN_CONFIDENCE       0.7\n",
      "DETECTION_NMS_THRESHOLD        0.3\n",
      "FPN_CLASSIF_FC_LAYERS_SIZE     1024\n",
      "GPU_COUNT                      1\n",
      "GRADIENT_CLIP_NORM             5.0\n",
      "IMAGES_PER_GPU                 2\n",
      "IMAGE_MAX_DIM                  128\n",
      "IMAGE_META_SIZE                14\n",
      "IMAGE_MIN_DIM                  128\n",
      "IMAGE_MIN_SCALE                0\n",
      "IMAGE_RESIZE_MODE              crop\n",
      "IMAGE_SHAPE                    [128 128   3]\n",
      "LEARNING_MOMENTUM              0.9\n",
      "LEARNING_RATE                  0.001\n",
      "LOSS_WEIGHTS                   {'rpn_class_loss': 1.0, 'rpn_bbox_loss': 1.0, 'mrcnn_class_loss': 1.0, 'mrcnn_bbox_loss': 1.0, 'mrcnn_mask_loss': 1.0}\n",
      "MASK_POOL_SIZE                 14\n",
      "MASK_SHAPE                     [28, 28]\n",
      "MAX_GT_INSTANCES               256\n",
      "MEAN_PIXEL                     [0 0 0]\n",
      "MINI_MASK_SHAPE                (56, 56)\n",
      "NAME                           cell\n",
      "NUM_CLASSES                    2\n",
      "POOL_SIZE                      7\n",
      "POST_NMS_ROIS_INFERENCE        1000\n",
      "POST_NMS_ROIS_TRAINING         2000\n",
      "ROI_POSITIVE_RATIO             0.33\n",
      "RPN_ANCHOR_RATIOS              [0.5, 1, 2]\n",
      "RPN_ANCHOR_SCALES              (8, 16, 32, 64, 128)\n",
      "RPN_ANCHOR_STRIDE              1\n",
      "RPN_BBOX_STD_DEV               [0.1 0.1 0.2 0.2]\n",
      "RPN_NMS_THRESHOLD              0.7\n",
      "RPN_TRAIN_ANCHORS_PER_IMAGE    256\n",
      "STEPS_PER_EPOCH                300\n",
      "TOP_DOWN_PYRAMID_SIZE          256\n",
      "TRAIN_BN                       False\n",
      "TRAIN_ROIS_PER_IMAGE           320\n",
      "USE_MINI_MASK                  True\n",
      "USE_RPN_ROIS                   True\n",
      "VALIDATION_STEPS               50\n",
      "WEIGHT_DECAY                   0.0001\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "class CellConfig(Config):\n",
    "    \"\"\"Configuration for training on the toy shapes dataset.\n",
    "    Derives from the base Config class and overrides values specific\n",
    "    to the toy shapes dataset.\n",
    "    \"\"\"\n",
    "    # Give the configuration a recognizable name\n",
    "    NAME = \"cell\"\n",
    "    BACKBONE = 'resnet50'\n",
    "    # Train on 1 GPU and 8 images per GPU. We can put multiple images on each\n",
    "    # GPU because the images are small. Batch size is 8 (GPUs * images/GPU).\n",
    "    GPU_COUNT = 1\n",
    "    IMAGES_PER_GPU = 2\n",
    "    #RPN_NMS_THRESHOLD = 0.8\n",
    "    DETECTION_MAX_INSTANCES = 256 #500\n",
    "    MAX_GT_INSTANCES = 256\n",
    "    # Number of classes (including background)\n",
    "    NUM_CLASSES = 1 + 1  # background + 1 cell\n",
    "    # Use small images for faster training. Set the limits of the small side\n",
    "    # the large side, and that determines the image shape.\n",
    "    IMAGE_MIN_DIM = 128\n",
    "    IMAGE_MAX_DIM = 128 #128\n",
    "    RPN_TRAIN_ANCHORS_PER_IMAGE = 256 #1024 #256 #1024 #256\n",
    "    IMAGE_RESIZE_MODE = 'crop'\n",
    "    ROI_POSITIVE_RATIO = 0.33\n",
    "    # Use smaller anchors because our image and objects are small\n",
    "    RPN_ANCHOR_SCALES = (8, 16, 32, 64, 128)  # anchor side in pixels\n",
    "    # Reduce training ROIs per image because the images are small and have\n",
    "    # few objects. Aim to allow ROI sampling to pick 33% positive ROIs.\n",
    "    TRAIN_ROIS_PER_IMAGE = 320 #1000 #320 #1000 #320 # 80\n",
    "    #LOSS_WEIGHTS  ={}\n",
    "    # Use a small epoch since the data is simple\n",
    "    STEPS_PER_EPOCH = 1000 #1000 (for a comprehensive training, 1000 is recommended)\n",
    "    # use small validation steps since the epoch is small\n",
    "    VALIDATION_STEPS =50# 10 #5\n",
    "    MEAN_PIXEL = np.array([0,0,0])  #([45,25,40]) # based on your own data\n",
    "\n",
    "    # ALSO CHANGED CONFIG.PY to change IMAGE SHAPE\n",
    "    \n",
    "    \n",
    "config = CellConfig()\n",
    "config.display()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "## Load data and configuration into model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "button": false,
    "collapsed": false,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": false
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image Count: 1\n",
      "Class Count: 2\n",
      "  0. BG                                                \n",
      "  1. cell                                              \n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbb099adf60>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "config = CellConfig()\n",
    "#config.display()\n",
    "\n",
    "# Training dataset\n",
    "dataset_train = CellDataset()\n",
    "dataset_train.setImages(images)\n",
    "dataset_train.setLabels(labels)\n",
    "dataset_train.setSize(size)\n",
    "dataset_train.load_cells('dummy_path')\n",
    "dataset_train.prepare()\n",
    "\n",
    "#All entries must be nonzero!\n",
    "# Validation dataset\n",
    "dataset_val = CellDataset()\n",
    "dataset_val.setImages(images)\n",
    "dataset_val.setLabels(labels)\n",
    "dataset_val.setSize(size)\n",
    "dataset_val.load_cells('dummy_path')\n",
    "dataset_val.prepare()\n",
    "\n",
    "print(\"Image Count: {}\".format(len(dataset_train.image_ids)))\n",
    "print(\"Class Count: {}\".format(dataset_train.num_classes))\n",
    "for i, info in enumerate(dataset_train.class_info):\n",
    "    print(\"{:3}. {:50}\".format(i, info['name']))\n",
    "    \n",
    "for i in range(0,min(len(images),5)):\n",
    "    image = dataset_train.load_image(i)\n",
    "    mask, class_ids = dataset_train.load_mask(i)\n",
    "    visualize.display_top_masks(image, mask, class_ids, dataset_train.class_names)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "## Training configuration"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "### Model backend and Pre-trained weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "button": false,
    "collapsed": true,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": true
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [],
   "source": [
    "# Create model in training mode\n",
    "model = modellib.MaskRCNN(mode=\"training\", config=config,\n",
    "                          model_dir=MODEL_DIR)\n",
    "#Which weights to start with?\n",
    "init_with = \"imagnet\"  # imagenet, coco, or last\n",
    "\n",
    "if init_with == \"imagenet\":\n",
    "    model.load_weights(model.get_imagenet_weights(), by_name=True)\n",
    "elif init_with == \"coco\":\n",
    "    # Load weights trained on MS COCO, but skip layers that\n",
    "    # are different due to the different number of classes\n",
    "    # See README for instructions to download the COCO weights\n",
    "    model.load_weights(COCO_MODEL_PATH, by_name=True,\n",
    "                       exclude=[\"mrcnn_class_logits\", \"mrcnn_bbox_fc\", \n",
    "                                \"mrcnn_bbox\", \"mrcnn_mask\"])\n",
    "elif init_with == \"last\":\n",
    "    # Load the last model you trained and continue training\n",
    "    model.load_weights(model.find_last()[1], by_name=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "### Data Augmentation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "button": false,
    "collapsed": true,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": true
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [],
   "source": [
    "#Data Augmentation\n",
    "\n",
    "augmentation = iaa.SomeOf((0, 5), [\n",
    "        iaa.Fliplr(0.5),\n",
    "        iaa.Flipud(0.5),\n",
    "        iaa.OneOf([iaa.Affine(rotate=90),\n",
    "                   iaa.Affine(rotate=180),\n",
    "                   iaa.Affine(rotate=270)]),\n",
    "        iaa.Affine(rotate=(-90,90)),\n",
    "    iaa.Sometimes(0.2, iaa.GaussianBlur(sigma=(0.0,5.0)))\n",
    "    ,iaa.Sometimes(0.2, iaa.Affine(scale={\"x\": (0.7, 1.5), \"y\": (0.7, 1.5)})),\n",
    "    iaa.Sometimes(0.2,  iaa.Multiply((0.7, 1.5))),\n",
    "    iaa.Sometimes(0.1, iaa.Affine(shear=(-45, 45)))\n",
    "])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "button": false,
    "collapsed": true,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": true
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "## Train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "### Train on the heads layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "button": false,
    "collapsed": false,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": false
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/ubuntu/anaconda3/envs/github2/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/base.py:198: retry (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use the retry module or similar alternatives.\n",
      "\n",
      "Starting at epoch 0. LR=0.001\n",
      "\n",
      "Checkpoint Path: /home/ubuntu/github/MaskRCNN/logs/cell20191219T1724/mask_rcnn_cell_{epoch:04d}.h5\n",
      "Selecting layers to train\n",
      "fpn_c5p5               (Conv2D)\n",
      "fpn_c4p4               (Conv2D)\n",
      "fpn_c3p3               (Conv2D)\n",
      "fpn_c2p2               (Conv2D)\n",
      "fpn_p5                 (Conv2D)\n",
      "fpn_p2                 (Conv2D)\n",
      "fpn_p3                 (Conv2D)\n",
      "fpn_p4                 (Conv2D)\n",
      "In model:  rpn_model\n",
      "    rpn_conv_shared        (Conv2D)\n",
      "    rpn_class_raw          (Conv2D)\n",
      "    rpn_bbox_pred          (Conv2D)\n",
      "mrcnn_mask_conv1       (TimeDistributed)\n",
      "mrcnn_mask_bn1         (TimeDistributed)\n",
      "mrcnn_mask_conv2       (TimeDistributed)\n",
      "mrcnn_mask_bn2         (TimeDistributed)\n",
      "mrcnn_class_conv1      (TimeDistributed)\n",
      "mrcnn_class_bn1        (TimeDistributed)\n",
      "mrcnn_mask_conv3       (TimeDistributed)\n",
      "mrcnn_mask_bn3         (TimeDistributed)\n",
      "mrcnn_class_conv2      (TimeDistributed)\n",
      "mrcnn_class_bn2        (TimeDistributed)\n",
      "mrcnn_mask_conv4       (TimeDistributed)\n",
      "mrcnn_mask_bn4         (TimeDistributed)\n",
      "mrcnn_bbox_fc          (TimeDistributed)\n",
      "mrcnn_mask_deconv      (TimeDistributed)\n",
      "mrcnn_class_logits     (TimeDistributed)\n",
      "mrcnn_mask             (TimeDistributed)\n",
      "Epoch 1/10\n",
      "300/300 [==============================] - 793s 3s/step - loss: 4.2064 - rpn_class_loss: 0.7209 - rpn_bbox_loss: 1.5385 - mrcnn_class_loss: 0.5210 - mrcnn_bbox_loss: 0.7903 - mrcnn_mask_loss: 0.6357 - val_loss: 3.1272 - val_rpn_class_loss: 0.3979 - val_rpn_bbox_loss: 0.9635 - val_mrcnn_class_loss: 0.5951 - val_mrcnn_bbox_loss: 0.5932 - val_mrcnn_mask_loss: 0.5774\n",
      "Epoch 2/10\n",
      "300/300 [==============================] - 738s 2s/step - loss: 3.1078 - rpn_class_loss: 0.4337 - rpn_bbox_loss: 0.9870 - mrcnn_class_loss: 0.5000 - mrcnn_bbox_loss: 0.6067 - mrcnn_mask_loss: 0.5805 - val_loss: 2.7847 - val_rpn_class_loss: 0.3201 - val_rpn_bbox_loss: 0.8847 - val_mrcnn_class_loss: 0.4717 - val_mrcnn_bbox_loss: 0.5537 - val_mrcnn_mask_loss: 0.5545\n",
      "Epoch 3/10\n",
      "300/300 [==============================] - 763s 3s/step - loss: 2.8210 - rpn_class_loss: 0.3816 - rpn_bbox_loss: 0.8684 - mrcnn_class_loss: 0.4620 - mrcnn_bbox_loss: 0.5511 - mrcnn_mask_loss: 0.5580 - val_loss: 2.4715 - val_rpn_class_loss: 0.2725 - val_rpn_bbox_loss: 0.7743 - val_mrcnn_class_loss: 0.4360 - val_mrcnn_bbox_loss: 0.4634 - val_mrcnn_mask_loss: 0.5253\n",
      "Epoch 4/10\n",
      "300/300 [==============================] - 748s 2s/step - loss: 2.6939 - rpn_class_loss: 0.3584 - rpn_bbox_loss: 0.8335 - mrcnn_class_loss: 0.4394 - mrcnn_bbox_loss: 0.5201 - mrcnn_mask_loss: 0.5425 - val_loss: 2.4180 - val_rpn_class_loss: 0.2617 - val_rpn_bbox_loss: 0.7796 - val_mrcnn_class_loss: 0.4119 - val_mrcnn_bbox_loss: 0.4480 - val_mrcnn_mask_loss: 0.5168\n",
      "Epoch 5/10\n",
      "300/300 [==============================] - 752s 3s/step - loss: 2.6036 - rpn_class_loss: 0.3473 - rpn_bbox_loss: 0.8001 - mrcnn_class_loss: 0.4289 - mrcnn_bbox_loss: 0.4946 - mrcnn_mask_loss: 0.5327 - val_loss: 2.2559 - val_rpn_class_loss: 0.2309 - val_rpn_bbox_loss: 0.7277 - val_mrcnn_class_loss: 0.3827 - val_mrcnn_bbox_loss: 0.4140 - val_mrcnn_mask_loss: 0.5006\n",
      "Epoch 6/10\n",
      "300/300 [==============================] - 756s 3s/step - loss: 2.5212 - rpn_class_loss: 0.3293 - rpn_bbox_loss: 0.7770 - mrcnn_class_loss: 0.4177 - mrcnn_bbox_loss: 0.4768 - mrcnn_mask_loss: 0.5203 - val_loss: 2.1621 - val_rpn_class_loss: 0.2269 - val_rpn_bbox_loss: 0.7048 - val_mrcnn_class_loss: 0.3765 - val_mrcnn_bbox_loss: 0.3776 - val_mrcnn_mask_loss: 0.4764\n",
      "Epoch 7/10\n",
      "300/300 [==============================] - 744s 2s/step - loss: 2.4439 - rpn_class_loss: 0.3149 - rpn_bbox_loss: 0.7516 - mrcnn_class_loss: 0.4079 - mrcnn_bbox_loss: 0.4551 - mrcnn_mask_loss: 0.5143 - val_loss: 2.1146 - val_rpn_class_loss: 0.2222 - val_rpn_bbox_loss: 0.6425 - val_mrcnn_class_loss: 0.3839 - val_mrcnn_bbox_loss: 0.3790 - val_mrcnn_mask_loss: 0.4871\n",
      "Epoch 8/10\n",
      "300/300 [==============================] - 751s 3s/step - loss: 2.4154 - rpn_class_loss: 0.3104 - rpn_bbox_loss: 0.7388 - mrcnn_class_loss: 0.4081 - mrcnn_bbox_loss: 0.4531 - mrcnn_mask_loss: 0.5050 - val_loss: 2.0478 - val_rpn_class_loss: 0.2064 - val_rpn_bbox_loss: 0.6379 - val_mrcnn_class_loss: 0.3695 - val_mrcnn_bbox_loss: 0.3634 - val_mrcnn_mask_loss: 0.4705\n",
      "Epoch 9/10\n",
      "300/300 [==============================] - 741s 2s/step - loss: 2.3496 - rpn_class_loss: 0.2970 - rpn_bbox_loss: 0.7188 - mrcnn_class_loss: 0.3977 - mrcnn_bbox_loss: 0.4351 - mrcnn_mask_loss: 0.5010 - val_loss: 2.0916 - val_rpn_class_loss: 0.2213 - val_rpn_bbox_loss: 0.6136 - val_mrcnn_class_loss: 0.3561 - val_mrcnn_bbox_loss: 0.3670 - val_mrcnn_mask_loss: 0.5336\n",
      "Epoch 10/10\n",
      "300/300 [==============================] - 737s 2s/step - loss: 2.2727 - rpn_class_loss: 0.2880 - rpn_bbox_loss: 0.6967 - mrcnn_class_loss: 0.3819 - mrcnn_bbox_loss: 0.4165 - mrcnn_mask_loss: 0.4896 - val_loss: 1.9406 - val_rpn_class_loss: 0.1952 - val_rpn_bbox_loss: 0.5823 - val_mrcnn_class_loss: 0.3681 - val_mrcnn_bbox_loss: 0.3344 - val_mrcnn_mask_loss: 0.4605\n",
      "Last trained model path: /home/ubuntu/github/MaskRCNN/logs/cell20191219T1724/mask_rcnn_cell_0010.h5\n"
     ]
    }
   ],
   "source": [
    "#Run \"tensorboard --logdir logs/\" in terminal to visualize results\n",
    "#Default Port #: 6006\n",
    "\n",
    "model.train(dataset_train, dataset_val,\n",
    "            learning_rate=config.LEARNING_RATE, \n",
    "            epochs=10,augmentation=augmentation,\n",
    "            layers='heads') #all heads\n",
    "\n",
    "model_path = model.find_last()\n",
    "print(\"Last trained model path: \" + str(model_path))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "### Train on all layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "button": false,
    "collapsed": false,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": false
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Starting at epoch 10. LR=0.001\n",
      "\n",
      "Checkpoint Path: /home/ubuntu/github/MaskRCNN/logs/cell20191219T1724/mask_rcnn_cell_{epoch:04d}.h5\n",
      "Selecting layers to train\n",
      "conv1                  (Conv2D)\n",
      "bn_conv1               (BatchNorm)\n",
      "res2a_branch2a         (Conv2D)\n",
      "bn2a_branch2a          (BatchNorm)\n",
      "res2a_branch2b         (Conv2D)\n",
      "bn2a_branch2b          (BatchNorm)\n",
      "res2a_branch2c         (Conv2D)\n",
      "res2a_branch1          (Conv2D)\n",
      "bn2a_branch2c          (BatchNorm)\n",
      "bn2a_branch1           (BatchNorm)\n",
      "res2b_branch2a         (Conv2D)\n",
      "bn2b_branch2a          (BatchNorm)\n",
      "res2b_branch2b         (Conv2D)\n",
      "bn2b_branch2b          (BatchNorm)\n",
      "res2b_branch2c         (Conv2D)\n",
      "bn2b_branch2c          (BatchNorm)\n",
      "res2c_branch2a         (Conv2D)\n",
      "bn2c_branch2a          (BatchNorm)\n",
      "res2c_branch2b         (Conv2D)\n",
      "bn2c_branch2b          (BatchNorm)\n",
      "res2c_branch2c         (Conv2D)\n",
      "bn2c_branch2c          (BatchNorm)\n",
      "res3a_branch2a         (Conv2D)\n",
      "bn3a_branch2a          (BatchNorm)\n",
      "res3a_branch2b         (Conv2D)\n",
      "bn3a_branch2b          (BatchNorm)\n",
      "res3a_branch2c         (Conv2D)\n",
      "res3a_branch1          (Conv2D)\n",
      "bn3a_branch2c          (BatchNorm)\n",
      "bn3a_branch1           (BatchNorm)\n",
      "res3b_branch2a         (Conv2D)\n",
      "bn3b_branch2a          (BatchNorm)\n",
      "res3b_branch2b         (Conv2D)\n",
      "bn3b_branch2b          (BatchNorm)\n",
      "res3b_branch2c         (Conv2D)\n",
      "bn3b_branch2c          (BatchNorm)\n",
      "res3c_branch2a         (Conv2D)\n",
      "bn3c_branch2a          (BatchNorm)\n",
      "res3c_branch2b         (Conv2D)\n",
      "bn3c_branch2b          (BatchNorm)\n",
      "res3c_branch2c         (Conv2D)\n",
      "bn3c_branch2c          (BatchNorm)\n",
      "res3d_branch2a         (Conv2D)\n",
      "bn3d_branch2a          (BatchNorm)\n",
      "res3d_branch2b         (Conv2D)\n",
      "bn3d_branch2b          (BatchNorm)\n",
      "res3d_branch2c         (Conv2D)\n",
      "bn3d_branch2c          (BatchNorm)\n",
      "res4a_branch2a         (Conv2D)\n",
      "bn4a_branch2a          (BatchNorm)\n",
      "res4a_branch2b         (Conv2D)\n",
      "bn4a_branch2b          (BatchNorm)\n",
      "res4a_branch2c         (Conv2D)\n",
      "res4a_branch1          (Conv2D)\n",
      "bn4a_branch2c          (BatchNorm)\n",
      "bn4a_branch1           (BatchNorm)\n",
      "res4b_branch2a         (Conv2D)\n",
      "bn4b_branch2a          (BatchNorm)\n",
      "res4b_branch2b         (Conv2D)\n",
      "bn4b_branch2b          (BatchNorm)\n",
      "res4b_branch2c         (Conv2D)\n",
      "bn4b_branch2c          (BatchNorm)\n",
      "res4c_branch2a         (Conv2D)\n",
      "bn4c_branch2a          (BatchNorm)\n",
      "res4c_branch2b         (Conv2D)\n",
      "bn4c_branch2b          (BatchNorm)\n",
      "res4c_branch2c         (Conv2D)\n",
      "bn4c_branch2c          (BatchNorm)\n",
      "res4d_branch2a         (Conv2D)\n",
      "bn4d_branch2a          (BatchNorm)\n",
      "res4d_branch2b         (Conv2D)\n",
      "bn4d_branch2b          (BatchNorm)\n",
      "res4d_branch2c         (Conv2D)\n",
      "bn4d_branch2c          (BatchNorm)\n",
      "res4e_branch2a         (Conv2D)\n",
      "bn4e_branch2a          (BatchNorm)\n",
      "res4e_branch2b         (Conv2D)\n",
      "bn4e_branch2b          (BatchNorm)\n",
      "res4e_branch2c         (Conv2D)\n",
      "bn4e_branch2c          (BatchNorm)\n",
      "res4f_branch2a         (Conv2D)\n",
      "bn4f_branch2a          (BatchNorm)\n",
      "res4f_branch2b         (Conv2D)\n",
      "bn4f_branch2b          (BatchNorm)\n",
      "res4f_branch2c         (Conv2D)\n",
      "bn4f_branch2c          (BatchNorm)\n",
      "res5a_branch2a         (Conv2D)\n",
      "bn5a_branch2a          (BatchNorm)\n",
      "res5a_branch2b         (Conv2D)\n",
      "bn5a_branch2b          (BatchNorm)\n",
      "res5a_branch2c         (Conv2D)\n",
      "res5a_branch1          (Conv2D)\n",
      "bn5a_branch2c          (BatchNorm)\n",
      "bn5a_branch1           (BatchNorm)\n",
      "res5b_branch2a         (Conv2D)\n",
      "bn5b_branch2a          (BatchNorm)\n",
      "res5b_branch2b         (Conv2D)\n",
      "bn5b_branch2b          (BatchNorm)\n",
      "res5b_branch2c         (Conv2D)\n",
      "bn5b_branch2c          (BatchNorm)\n",
      "res5c_branch2a         (Conv2D)\n",
      "bn5c_branch2a          (BatchNorm)\n",
      "res5c_branch2b         (Conv2D)\n",
      "bn5c_branch2b          (BatchNorm)\n",
      "res5c_branch2c         (Conv2D)\n",
      "bn5c_branch2c          (BatchNorm)\n",
      "fpn_c5p5               (Conv2D)\n",
      "fpn_c4p4               (Conv2D)\n",
      "fpn_c3p3               (Conv2D)\n",
      "fpn_c2p2               (Conv2D)\n",
      "fpn_p5                 (Conv2D)\n",
      "fpn_p2                 (Conv2D)\n",
      "fpn_p3                 (Conv2D)\n",
      "fpn_p4                 (Conv2D)\n",
      "In model:  rpn_model\n",
      "    rpn_conv_shared        (Conv2D)\n",
      "    rpn_class_raw          (Conv2D)\n",
      "    rpn_bbox_pred          (Conv2D)\n",
      "mrcnn_mask_conv1       (TimeDistributed)\n",
      "mrcnn_mask_bn1         (TimeDistributed)\n",
      "mrcnn_mask_conv2       (TimeDistributed)\n",
      "mrcnn_mask_bn2         (TimeDistributed)\n",
      "mrcnn_class_conv1      (TimeDistributed)\n",
      "mrcnn_class_bn1        (TimeDistributed)\n",
      "mrcnn_mask_conv3       (TimeDistributed)\n",
      "mrcnn_mask_bn3         (TimeDistributed)\n",
      "mrcnn_class_conv2      (TimeDistributed)\n",
      "mrcnn_class_bn2        (TimeDistributed)\n",
      "mrcnn_mask_conv4       (TimeDistributed)\n",
      "mrcnn_mask_bn4         (TimeDistributed)\n",
      "mrcnn_bbox_fc          (TimeDistributed)\n",
      "mrcnn_mask_deconv      (TimeDistributed)\n",
      "mrcnn_class_logits     (TimeDistributed)\n",
      "mrcnn_mask             (TimeDistributed)\n",
      "Epoch 11/20\n",
      "300/300 [==============================] - 785s 3s/step - loss: 2.2766 - rpn_class_loss: 0.2851 - rpn_bbox_loss: 0.6939 - mrcnn_class_loss: 0.3856 - mrcnn_bbox_loss: 0.4272 - mrcnn_mask_loss: 0.4847 - val_loss: 1.9824 - val_rpn_class_loss: 0.2303 - val_rpn_bbox_loss: 0.5405 - val_mrcnn_class_loss: 0.3637 - val_mrcnn_bbox_loss: 0.3586 - val_mrcnn_mask_loss: 0.4893\n",
      "Epoch 12/20\n",
      "300/300 [==============================] - 745s 2s/step - loss: 2.1796 - rpn_class_loss: 0.2694 - rpn_bbox_loss: 0.6623 - mrcnn_class_loss: 0.3695 - mrcnn_bbox_loss: 0.4076 - mrcnn_mask_loss: 0.4708 - val_loss: 1.8569 - val_rpn_class_loss: 0.2148 - val_rpn_bbox_loss: 0.5312 - val_mrcnn_class_loss: 0.3121 - val_mrcnn_bbox_loss: 0.3260 - val_mrcnn_mask_loss: 0.4727\n",
      "Epoch 13/20\n",
      "300/300 [==============================] - 745s 2s/step - loss: 2.1209 - rpn_class_loss: 0.2598 - rpn_bbox_loss: 0.6448 - mrcnn_class_loss: 0.3607 - mrcnn_bbox_loss: 0.3930 - mrcnn_mask_loss: 0.4626 - val_loss: 1.8232 - val_rpn_class_loss: 0.2184 - val_rpn_bbox_loss: 0.5283 - val_mrcnn_class_loss: 0.2853 - val_mrcnn_bbox_loss: 0.3321 - val_mrcnn_mask_loss: 0.4590\n",
      "Epoch 14/20\n",
      "300/300 [==============================] - 736s 2s/step - loss: 2.0611 - rpn_class_loss: 0.2469 - rpn_bbox_loss: 0.6291 - mrcnn_class_loss: 0.3490 - mrcnn_bbox_loss: 0.3807 - mrcnn_mask_loss: 0.4554 - val_loss: 1.7262 - val_rpn_class_loss: 0.1966 - val_rpn_bbox_loss: 0.5040 - val_mrcnn_class_loss: 0.2987 - val_mrcnn_bbox_loss: 0.2869 - val_mrcnn_mask_loss: 0.4401\n",
      "Epoch 15/20\n",
      "300/300 [==============================] - 743s 2s/step - loss: 2.0275 - rpn_class_loss: 0.2435 - rpn_bbox_loss: 0.6173 - mrcnn_class_loss: 0.3422 - mrcnn_bbox_loss: 0.3744 - mrcnn_mask_loss: 0.4501 - val_loss: 1.7293 - val_rpn_class_loss: 0.1830 - val_rpn_bbox_loss: 0.5034 - val_mrcnn_class_loss: 0.2925 - val_mrcnn_bbox_loss: 0.3079 - val_mrcnn_mask_loss: 0.4425\n",
      "Epoch 16/20\n",
      "300/300 [==============================] - 748s 2s/step - loss: 2.0047 - rpn_class_loss: 0.2376 - rpn_bbox_loss: 0.6105 - mrcnn_class_loss: 0.3373 - mrcnn_bbox_loss: 0.3727 - mrcnn_mask_loss: 0.4466 - val_loss: 1.6712 - val_rpn_class_loss: 0.1881 - val_rpn_bbox_loss: 0.4789 - val_mrcnn_class_loss: 0.2850 - val_mrcnn_bbox_loss: 0.2913 - val_mrcnn_mask_loss: 0.4278\n",
      "Epoch 17/20\n",
      "300/300 [==============================] - 752s 3s/step - loss: 1.9486 - rpn_class_loss: 0.2301 - rpn_bbox_loss: 0.5917 - mrcnn_class_loss: 0.3303 - mrcnn_bbox_loss: 0.3562 - mrcnn_mask_loss: 0.4402 - val_loss: 1.6574 - val_rpn_class_loss: 0.1831 - val_rpn_bbox_loss: 0.4663 - val_mrcnn_class_loss: 0.3000 - val_mrcnn_bbox_loss: 0.2867 - val_mrcnn_mask_loss: 0.4212\n",
      "Epoch 18/20\n",
      "300/300 [==============================] - 746s 2s/step - loss: 1.9898 - rpn_class_loss: 0.2343 - rpn_bbox_loss: 0.6014 - mrcnn_class_loss: 0.3384 - mrcnn_bbox_loss: 0.3738 - mrcnn_mask_loss: 0.4419 - val_loss: 1.8769 - val_rpn_class_loss: 0.1965 - val_rpn_bbox_loss: 0.4860 - val_mrcnn_class_loss: 0.3675 - val_mrcnn_bbox_loss: 0.3882 - val_mrcnn_mask_loss: 0.4386\n",
      "Epoch 19/20\n",
      "300/300 [==============================] - 736s 2s/step - loss: 1.9285 - rpn_class_loss: 0.2267 - rpn_bbox_loss: 0.5818 - mrcnn_class_loss: 0.3331 - mrcnn_bbox_loss: 0.3550 - mrcnn_mask_loss: 0.4320 - val_loss: 1.5901 - val_rpn_class_loss: 0.1637 - val_rpn_bbox_loss: 0.4690 - val_mrcnn_class_loss: 0.2713 - val_mrcnn_bbox_loss: 0.2582 - val_mrcnn_mask_loss: 0.4278\n",
      "Epoch 20/20\n",
      "300/300 [==============================] - 740s 2s/step - loss: 1.8444 - rpn_class_loss: 0.2128 - rpn_bbox_loss: 0.5598 - mrcnn_class_loss: 0.3145 - mrcnn_bbox_loss: 0.3343 - mrcnn_mask_loss: 0.4229 - val_loss: 1.3930 - val_rpn_class_loss: 0.1431 - val_rpn_bbox_loss: 0.4210 - val_mrcnn_class_loss: 0.2415 - val_mrcnn_bbox_loss: 0.2014 - val_mrcnn_mask_loss: 0.3860\n"
     ]
    }
   ],
   "source": [
    "model.train(dataset_train, dataset_val, \n",
    "            learning_rate=config.LEARNING_RATE, \n",
    "            epochs=20,augmentation=augmentation,\n",
    "            layers='all') #all\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "button": false,
    "deletable": true,
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "source": [
    "### Train on all layers with lower learning rate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "button": false,
    "collapsed": false,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": false
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Starting at epoch 20. LR=0.0001\n",
      "\n",
      "Checkpoint Path: /home/ubuntu/github/MaskRCNN/logs/cell20191219T1724/mask_rcnn_cell_{epoch:04d}.h5\n",
      "Selecting layers to train\n",
      "conv1                  (Conv2D)\n",
      "bn_conv1               (BatchNorm)\n",
      "res2a_branch2a         (Conv2D)\n",
      "bn2a_branch2a          (BatchNorm)\n",
      "res2a_branch2b         (Conv2D)\n",
      "bn2a_branch2b          (BatchNorm)\n",
      "res2a_branch2c         (Conv2D)\n",
      "res2a_branch1          (Conv2D)\n",
      "bn2a_branch2c          (BatchNorm)\n",
      "bn2a_branch1           (BatchNorm)\n",
      "res2b_branch2a         (Conv2D)\n",
      "bn2b_branch2a          (BatchNorm)\n",
      "res2b_branch2b         (Conv2D)\n",
      "bn2b_branch2b          (BatchNorm)\n",
      "res2b_branch2c         (Conv2D)\n",
      "bn2b_branch2c          (BatchNorm)\n",
      "res2c_branch2a         (Conv2D)\n",
      "bn2c_branch2a          (BatchNorm)\n",
      "res2c_branch2b         (Conv2D)\n",
      "bn2c_branch2b          (BatchNorm)\n",
      "res2c_branch2c         (Conv2D)\n",
      "bn2c_branch2c          (BatchNorm)\n",
      "res3a_branch2a         (Conv2D)\n",
      "bn3a_branch2a          (BatchNorm)\n",
      "res3a_branch2b         (Conv2D)\n",
      "bn3a_branch2b          (BatchNorm)\n",
      "res3a_branch2c         (Conv2D)\n",
      "res3a_branch1          (Conv2D)\n",
      "bn3a_branch2c          (BatchNorm)\n",
      "bn3a_branch1           (BatchNorm)\n",
      "res3b_branch2a         (Conv2D)\n",
      "bn3b_branch2a          (BatchNorm)\n",
      "res3b_branch2b         (Conv2D)\n",
      "bn3b_branch2b          (BatchNorm)\n",
      "res3b_branch2c         (Conv2D)\n",
      "bn3b_branch2c          (BatchNorm)\n",
      "res3c_branch2a         (Conv2D)\n",
      "bn3c_branch2a          (BatchNorm)\n",
      "res3c_branch2b         (Conv2D)\n",
      "bn3c_branch2b          (BatchNorm)\n",
      "res3c_branch2c         (Conv2D)\n",
      "bn3c_branch2c          (BatchNorm)\n",
      "res3d_branch2a         (Conv2D)\n",
      "bn3d_branch2a          (BatchNorm)\n",
      "res3d_branch2b         (Conv2D)\n",
      "bn3d_branch2b          (BatchNorm)\n",
      "res3d_branch2c         (Conv2D)\n",
      "bn3d_branch2c          (BatchNorm)\n",
      "res4a_branch2a         (Conv2D)\n",
      "bn4a_branch2a          (BatchNorm)\n",
      "res4a_branch2b         (Conv2D)\n",
      "bn4a_branch2b          (BatchNorm)\n",
      "res4a_branch2c         (Conv2D)\n",
      "res4a_branch1          (Conv2D)\n",
      "bn4a_branch2c          (BatchNorm)\n",
      "bn4a_branch1           (BatchNorm)\n",
      "res4b_branch2a         (Conv2D)\n",
      "bn4b_branch2a          (BatchNorm)\n",
      "res4b_branch2b         (Conv2D)\n",
      "bn4b_branch2b          (BatchNorm)\n",
      "res4b_branch2c         (Conv2D)\n",
      "bn4b_branch2c          (BatchNorm)\n",
      "res4c_branch2a         (Conv2D)\n",
      "bn4c_branch2a          (BatchNorm)\n",
      "res4c_branch2b         (Conv2D)\n",
      "bn4c_branch2b          (BatchNorm)\n",
      "res4c_branch2c         (Conv2D)\n",
      "bn4c_branch2c          (BatchNorm)\n",
      "res4d_branch2a         (Conv2D)\n",
      "bn4d_branch2a          (BatchNorm)\n",
      "res4d_branch2b         (Conv2D)\n",
      "bn4d_branch2b          (BatchNorm)\n",
      "res4d_branch2c         (Conv2D)\n",
      "bn4d_branch2c          (BatchNorm)\n",
      "res4e_branch2a         (Conv2D)\n",
      "bn4e_branch2a          (BatchNorm)\n",
      "res4e_branch2b         (Conv2D)\n",
      "bn4e_branch2b          (BatchNorm)\n",
      "res4e_branch2c         (Conv2D)\n",
      "bn4e_branch2c          (BatchNorm)\n",
      "res4f_branch2a         (Conv2D)\n",
      "bn4f_branch2a          (BatchNorm)\n",
      "res4f_branch2b         (Conv2D)\n",
      "bn4f_branch2b          (BatchNorm)\n",
      "res4f_branch2c         (Conv2D)\n",
      "bn4f_branch2c          (BatchNorm)\n",
      "res5a_branch2a         (Conv2D)\n",
      "bn5a_branch2a          (BatchNorm)\n",
      "res5a_branch2b         (Conv2D)\n",
      "bn5a_branch2b          (BatchNorm)\n",
      "res5a_branch2c         (Conv2D)\n",
      "res5a_branch1          (Conv2D)\n",
      "bn5a_branch2c          (BatchNorm)\n",
      "bn5a_branch1           (BatchNorm)\n",
      "res5b_branch2a         (Conv2D)\n",
      "bn5b_branch2a          (BatchNorm)\n",
      "res5b_branch2b         (Conv2D)\n",
      "bn5b_branch2b          (BatchNorm)\n",
      "res5b_branch2c         (Conv2D)\n",
      "bn5b_branch2c          (BatchNorm)\n",
      "res5c_branch2a         (Conv2D)\n",
      "bn5c_branch2a          (BatchNorm)\n",
      "res5c_branch2b         (Conv2D)\n",
      "bn5c_branch2b          (BatchNorm)\n",
      "res5c_branch2c         (Conv2D)\n",
      "bn5c_branch2c          (BatchNorm)\n",
      "fpn_c5p5               (Conv2D)\n",
      "fpn_c4p4               (Conv2D)\n",
      "fpn_c3p3               (Conv2D)\n",
      "fpn_c2p2               (Conv2D)\n",
      "fpn_p5                 (Conv2D)\n",
      "fpn_p2                 (Conv2D)\n",
      "fpn_p3                 (Conv2D)\n",
      "fpn_p4                 (Conv2D)\n",
      "In model:  rpn_model\n",
      "    rpn_conv_shared        (Conv2D)\n",
      "    rpn_class_raw          (Conv2D)\n",
      "    rpn_bbox_pred          (Conv2D)\n",
      "mrcnn_mask_conv1       (TimeDistributed)\n",
      "mrcnn_mask_bn1         (TimeDistributed)\n",
      "mrcnn_mask_conv2       (TimeDistributed)\n",
      "mrcnn_mask_bn2         (TimeDistributed)\n",
      "mrcnn_class_conv1      (TimeDistributed)\n",
      "mrcnn_class_bn1        (TimeDistributed)\n",
      "mrcnn_mask_conv3       (TimeDistributed)\n",
      "mrcnn_mask_bn3         (TimeDistributed)\n",
      "mrcnn_class_conv2      (TimeDistributed)\n",
      "mrcnn_class_bn2        (TimeDistributed)\n",
      "mrcnn_mask_conv4       (TimeDistributed)\n",
      "mrcnn_mask_bn4         (TimeDistributed)\n",
      "mrcnn_bbox_fc          (TimeDistributed)\n",
      "mrcnn_mask_deconv      (TimeDistributed)\n",
      "mrcnn_class_logits     (TimeDistributed)\n",
      "mrcnn_mask             (TimeDistributed)\n",
      "Epoch 21/30\n",
      "300/300 [==============================] - 778s 3s/step - loss: 1.8139 - rpn_class_loss: 0.2124 - rpn_bbox_loss: 0.5609 - mrcnn_class_loss: 0.3049 - mrcnn_bbox_loss: 0.3201 - mrcnn_mask_loss: 0.4156 - val_loss: 1.5862 - val_rpn_class_loss: 0.1743 - val_rpn_bbox_loss: 0.5192 - val_mrcnn_class_loss: 0.2818 - val_mrcnn_bbox_loss: 0.2238 - val_mrcnn_mask_loss: 0.3871\n",
      "Epoch 22/30\n",
      "300/300 [==============================] - 767s 3s/step - loss: 1.7966 - rpn_class_loss: 0.2104 - rpn_bbox_loss: 0.5534 - mrcnn_class_loss: 0.3033 - mrcnn_bbox_loss: 0.3158 - mrcnn_mask_loss: 0.4137 - val_loss: 1.6261 - val_rpn_class_loss: 0.1764 - val_rpn_bbox_loss: 0.5169 - val_mrcnn_class_loss: 0.2946 - val_mrcnn_bbox_loss: 0.2469 - val_mrcnn_mask_loss: 0.3912\n",
      "Epoch 23/30\n",
      "300/300 [==============================] - 765s 3s/step - loss: 1.7795 - rpn_class_loss: 0.2112 - rpn_bbox_loss: 0.5449 - mrcnn_class_loss: 0.3008 - mrcnn_bbox_loss: 0.3112 - mrcnn_mask_loss: 0.4113 - val_loss: 1.6235 - val_rpn_class_loss: 0.1727 - val_rpn_bbox_loss: 0.5112 - val_mrcnn_class_loss: 0.2988 - val_mrcnn_bbox_loss: 0.2364 - val_mrcnn_mask_loss: 0.4043\n",
      "Epoch 24/30\n",
      "300/300 [==============================] - 759s 3s/step - loss: 1.7640 - rpn_class_loss: 0.2060 - rpn_bbox_loss: 0.5458 - mrcnn_class_loss: 0.2975 - mrcnn_bbox_loss: 0.3046 - mrcnn_mask_loss: 0.4101 - val_loss: 1.5593 - val_rpn_class_loss: 0.1735 - val_rpn_bbox_loss: 0.5077 - val_mrcnn_class_loss: 0.2382 - val_mrcnn_bbox_loss: 0.2439 - val_mrcnn_mask_loss: 0.3960\n",
      "Epoch 25/30\n",
      "300/300 [==============================] - 765s 3s/step - loss: 1.7483 - rpn_class_loss: 0.2055 - rpn_bbox_loss: 0.5376 - mrcnn_class_loss: 0.2946 - mrcnn_bbox_loss: 0.3027 - mrcnn_mask_loss: 0.4079 - val_loss: 1.5700 - val_rpn_class_loss: 0.1708 - val_rpn_bbox_loss: 0.5133 - val_mrcnn_class_loss: 0.2527 - val_mrcnn_bbox_loss: 0.2457 - val_mrcnn_mask_loss: 0.3873\n",
      "Epoch 26/30\n",
      "300/300 [==============================] - 762s 3s/step - loss: 1.7502 - rpn_class_loss: 0.2054 - rpn_bbox_loss: 0.5396 - mrcnn_class_loss: 0.2932 - mrcnn_bbox_loss: 0.3032 - mrcnn_mask_loss: 0.4087 - val_loss: 1.5687 - val_rpn_class_loss: 0.1759 - val_rpn_bbox_loss: 0.5052 - val_mrcnn_class_loss: 0.2663 - val_mrcnn_bbox_loss: 0.2368 - val_mrcnn_mask_loss: 0.3845\n",
      "Epoch 27/30\n",
      "300/300 [==============================] - 753s 3s/step - loss: 1.7081 - rpn_class_loss: 0.1979 - rpn_bbox_loss: 0.5302 - mrcnn_class_loss: 0.2868 - mrcnn_bbox_loss: 0.2904 - mrcnn_mask_loss: 0.4027 - val_loss: 1.5793 - val_rpn_class_loss: 0.1700 - val_rpn_bbox_loss: 0.5072 - val_mrcnn_class_loss: 0.2633 - val_mrcnn_bbox_loss: 0.2382 - val_mrcnn_mask_loss: 0.4005\n",
      "Epoch 28/30\n",
      "300/300 [==============================] - 760s 3s/step - loss: 1.7742 - rpn_class_loss: 0.2078 - rpn_bbox_loss: 0.5424 - mrcnn_class_loss: 0.3015 - mrcnn_bbox_loss: 0.3130 - mrcnn_mask_loss: 0.4095 - val_loss: 1.5665 - val_rpn_class_loss: 0.1719 - val_rpn_bbox_loss: 0.5157 - val_mrcnn_class_loss: 0.2712 - val_mrcnn_bbox_loss: 0.2195 - val_mrcnn_mask_loss: 0.3882\n",
      "Epoch 29/30\n",
      "300/300 [==============================] - 756s 3s/step - loss: 1.7394 - rpn_class_loss: 0.2019 - rpn_bbox_loss: 0.5424 - mrcnn_class_loss: 0.2900 - mrcnn_bbox_loss: 0.2975 - mrcnn_mask_loss: 0.4077 - val_loss: 1.5258 - val_rpn_class_loss: 0.1710 - val_rpn_bbox_loss: 0.4994 - val_mrcnn_class_loss: 0.2650 - val_mrcnn_bbox_loss: 0.2111 - val_mrcnn_mask_loss: 0.3792\n",
      "Epoch 30/30\n",
      "300/300 [==============================] - 751s 3s/step - loss: 1.7041 - rpn_class_loss: 0.1972 - rpn_bbox_loss: 0.5267 - mrcnn_class_loss: 0.2881 - mrcnn_bbox_loss: 0.2884 - mrcnn_mask_loss: 0.4037 - val_loss: 1.5638 - val_rpn_class_loss: 0.1678 - val_rpn_bbox_loss: 0.5038 - val_mrcnn_class_loss: 0.2719 - val_mrcnn_bbox_loss: 0.2255 - val_mrcnn_mask_loss: 0.3948\n"
     ]
    }
   ],
   "source": [
    "model.train(dataset_train, dataset_val, \n",
    "            learning_rate=config.LEARNING_RATE/10, \n",
    "            epochs=30,augmentation=augmentation,\n",
    "            layers='all') #all\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "button": false,
    "collapsed": true,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": true
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "button": false,
    "collapsed": true,
    "deletable": true,
    "jupyter": {
     "outputs_hidden": true
    },
    "new_sheet": false,
    "run_control": {
     "read_only": false
    }
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tensorflow36",
   "language": "python",
   "name": "tensorflow36"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}