post_processing.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "source": [
    "# manually tries different post-processing strategies on the labelled data to find good parameters\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from utils.utils import load_data, load_test_predictions,  handle_small_segments, get_bbox_margin, get_metrics, compute_auc, get_shells, smooth_segments, load_predictions, handle_small_segments_simple, find_central_id, merge_in_rbc, delete_boundary_segments, get_path\n",
    "from post_process import post_process\n",
    "from skimage.measure import regionprops\n",
    "from skimage.morphology import ball, dilation, erosion\n",
    "import os"
   ],
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "execution_count": null
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "# on labelled data"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [],
   "source": [
    "data_path = os.path.join(get_path(), \"RBC_labelled\")\n",
    "\n",
    "datasets = [\"valid1\", \"valid2\", \"valid3\"]"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [],
   "source": [
    "#load data\n",
    "data_dict = {}\n",
    "for dataset in datasets:\n",
    "    imgs, gt, file_names = load_data(data_path, dataset)\n",
    "    data_dict[dataset] = {\"imgs\": imgs,\n",
    "                          \"gt\": gt,\n",
    "                          \"file_names\": file_names}"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [],
   "source": [
    "# check out size of segments\n",
    "all_counts = []\n",
    "for dataset in datasets:\n",
    "    for gt in data_dict[dataset][\"gt\"]:\n",
    "        _, counts = np.unique(gt, return_counts=True)\n",
    "        all_counts.append(counts[1:])\n",
    "all_counts = np.concatenate(all_counts)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "outputs": [],
   "source": [
    "seed = 0\n",
    "n_folds = 10\n",
    "eval_str_3D = \"data_valid1_valid2_valid3_by_stack_mode_3D_iso_min_train_masks_0_seed_0_epoch_499_aniso_3.2\"\n",
    "for dataset in datasets:\n",
    "    test_pred_3D, test_idx_3D = load_test_predictions(data_path,\n",
    "                                     dataset,\n",
    "                                     data_dict[dataset][\"file_names\"],\n",
    "                                     eval_str_3D,\n",
    "                                     seed=seed,\n",
    "                                     n_folds=n_folds)\n",
    "    reverse_idx = np.argsort(test_idx_3D)\n",
    "    test_pred_3D = test_pred_3D[reverse_idx]\n",
    "    data_dict[dataset][\"pred_3D\"] = test_pred_3D"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "outputs": [],
   "source": [
    "counts_3D = []\n",
    "for dataset in datasets:\n",
    "    for seg in data_dict[dataset][\"pred_3D\"]:\n",
    "        _, counts = np.unique(seg, return_counts=True)\n",
    "        counts_3D.append(counts[1:])\n",
    "counts_3D = np.concatenate(counts_3D)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "bins = np.exp(np.arange(100)/10)\n",
    "plt.hist(counts_3D, bins=bins, alpha=0.5, label=\"pred 2d\")\n",
    "plt.hist(all_counts, bins=bins, alpha=0.5, label=\"gt\")\n",
    "plt.legend()\n",
    "plt.xscale(\"log\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "outputs": [
    {
     "data": {
      "text/plain": "28"
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# there are small segments\n",
    "(counts_3D <2000).sum()"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "outputs": [],
   "source": [
    "# test handling of small segments (deprecated)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 650 ms, sys: 576 µs, total: 651 ms\n",
      "Wall time: 649 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "for seg in data_dict[\"valid2\"][\"pred_3D\"]:\n",
    "    handle_small_segments(seg, 2000, 3)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "outputs": [],
   "source": [
    "sample_id = 12\n",
    "sample_seg = data_dict[\"valid2\"][\"pred_3D\"][sample_id]"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[583095  57290   4283   5316    633    370    529    210    387    203\n",
      "     46    138]\n"
     ]
    }
   ],
   "source": [
    "_, c = np.unique(sample_seg, return_counts=True)\n",
    "print(c)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "outputs": [],
   "source": [
    "merged_seg = handle_small_segments(sample_seg, 2000, 3)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[583095  59806   4283   5316]\n"
     ]
    }
   ],
   "source": [
    "_, c = np.unique(merged_seg, return_counts=True)\n",
    "print(c)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.image.AxesImage at 0x7fd23970c940>"
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 3 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "z = 8\n",
    "\n",
    "fig, ax = plt.subplots(1, 3)\n",
    "ax[0].imshow(sample_seg[z], interpolation=\"none\")\n",
    "ax[1].imshow(merged_seg[z], interpolation=\"none\")\n",
    "ax[2].imshow(data_dict[\"valid2\"][\"gt\"][sample_id][z], interpolation=\"none\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "outputs": [],
   "source": [
    "# quantitative eval"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 4min 18s, sys: 2.56 s, total: 4min 21s\n",
      "Wall time: 4min 21s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "min_sizes = np.arange(0, 20)*100\n",
    "ngbh_sizes = [0, 1, 2]\n",
    "merged_aucs = np.zeros(shape=(len(min_sizes), len(ngbh_sizes)))\n",
    "merged_shell_aucs = np.zeros(shape=(len(min_sizes), len(ngbh_sizes)))\n",
    "\n",
    "for i, min_size in enumerate(min_sizes):\n",
    "    for j, ngbh_size in enumerate(ngbh_sizes):\n",
    "        for dataset in data_dict.keys():\n",
    "            if \"gt_shells\" not in data_dict[dataset].keys():\n",
    "                gt_shells = np.zeros_like(data_dict[dataset][\"gt\"])\n",
    "                for i, seg in enumerate(data_dict[dataset][\"gt\"]):\n",
    "                    gt_shells[i] = get_shells(seg, 1, 1 )\n",
    "                data_dict[dataset][\"gt_shells\"] = gt_shells\n",
    "            merged_segs = []\n",
    "            merged_segs_shells = []\n",
    "            for seg in data_dict[dataset][\"pred_3D\"]:\n",
    "                merged_seg = handle_small_segments(seg, min_size, ngbh_size)\n",
    "                merged_segs.append(merged_seg)\n",
    "                merged_segs_shells.append(get_shells(merged_seg))\n",
    "\n",
    "            data_dict[dataset][\"merged_3D\"] = np.stack(merged_segs)\n",
    "            data_dict[dataset][\"merged_3D_shells\"] = np.stack(merged_segs_shells)\n",
    "\n",
    "        merged_aucs[i,j] = compute_auc(data_dict, stage=\"merged_3D\")\n",
    "        merged_shell_aucs[i,j] = compute_auc(data_dict, stage=\"merged_3D_shells\", gt_stage=\"gt_shells\")\n"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%time\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(merged_aucs.T)\n",
    "_ = plt.xticks(np.arange(len(min_sizes))[::2], min_sizes[::2])"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC before merging 0.8068270980913027\n",
      "Best combination is min_size=700 and nghb_size=1\n",
      "Best AUC after merging 0.8225773358585858\n"
     ]
    }
   ],
   "source": [
    "max_auc = merged_aucs.max()\n",
    "\n",
    "idx = np.argwhere(merged_aucs == max_auc)[0]\n",
    "\n",
    "\n",
    "print(f\"AUC before merging {compute_auc(data_dict)}\")\n",
    "print(f\"Best combination is min_size={min_sizes[idx[0]]} and nghb_size={ngbh_sizes[idx[1]]}\")\n",
    "print(f\"Best AUC after merging {merged_aucs.max()}\")\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(merged_shell_aucs.T)\n",
    "_ = plt.xticks(np.arange(len(min_sizes))[::2], min_sizes[::2])"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC before merging 0.6261137103111535\n",
      "Best combination is min_size=700 and nghb_size=0\n",
      "Best AUC after merging 0.6360090863997114\n"
     ]
    }
   ],
   "source": [
    "max_auc = merged_shell_aucs.max()\n",
    "\n",
    "idx = np.argwhere(merged_shell_aucs == max_auc)[0]\n",
    "\n",
    "for dataset in data_dict:\n",
    "    pred_3D_shells = np.zeros_like(data_dict[dataset][\"gt\"])\n",
    "    for i, seg in enumerate(data_dict[dataset][\"pred_3D\"]):\n",
    "        pred_3D_shells[i] = get_shells(seg, 1, 1 )\n",
    "    data_dict[dataset][\"pred_3D_shells\"] = pred_3D_shells\n",
    "\n",
    "print(f\"AUC before merging {compute_auc(data_dict, stage='pred_3D_shells', gt_stage='gt_shells' )}\")\n",
    "print(f\"Best combination is min_size={min_sizes[idx[0]]} and nghb_size={ngbh_sizes[idx[1]]}\")\n",
    "print(f\"Best AUC after merging {merged_shell_aucs.max()}\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "outputs": [],
   "source": [
    "# --> filtering small segments helps, but merging small to larger cells does not help much and not at all in terms of shells. But it also does not hurt."
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "outputs": [],
   "source": [
    "best_min_size = 700\n",
    "best_nghb_size = 1\n",
    "for dataset in data_dict.keys():\n",
    "    merged_segs = []\n",
    "    for seg in data_dict[dataset][\"pred_3D\"]:\n",
    "        merged_segs.append(handle_small_segments(seg, best_min_size, best_nghb_size))\n",
    "    data_dict[dataset][\"merged_3D\"] = np.stack(merged_segs)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "outputs": [],
   "source": [
    "# get rid or jaggered patterns in z-direction"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "outputs": [],
   "source": [
    "# qualitative eval"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "outputs": [],
   "source": [
    "closed_sample_seg = smooth_segments(merged_seg, 2, False)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.image.AxesImage at 0x7fd23831ed60>"
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 1000x1000 with 3 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "z = 20\n",
    "\n",
    "fig, ax = plt.subplots(1, 3, figsize=(10,10))\n",
    "ax[0].imshow(merged_seg[z], interpolation=\"none\")\n",
    "ax[1].imshow(closed_sample_seg[z], interpolation=\"none\")\n",
    "ax[2].imshow(data_dict[\"valid2\"][\"gt\"][sample_id][z], interpolation=\"none\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "outputs": [],
   "source": [
    "# quantitative analysis"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "outputs": [],
   "source": [
    "dim_setting = [True, False]\n",
    "radii = [0, 1, 2, 3, 4, 5]\n",
    "smoothed_aucs = np.zeros(shape=(len(dim_setting), len(radii)))\n",
    "smoothed_shell_aucs = np.zeros(shape=(len(dim_setting), len(radii)))\n",
    "\n",
    "for i, do_3D in enumerate(dim_setting):\n",
    "    for j, radius in enumerate(radii):\n",
    "        for dataset in data_dict.keys():\n",
    "            if \"gt_shells\" not in data_dict[dataset].keys():\n",
    "                gt_shells = np.zeros_like(data_dict[dataset][\"gt\"])\n",
    "                for i, seg in enumerate(data_dict[dataset][\"gt\"]):\n",
    "                    gt_shells[i] = get_shells(seg, 1, 1 )\n",
    "                data_dict[dataset][\"gt_shells\"] = gt_shells\n",
    "\n",
    "            new_segs = []\n",
    "            new_shells = []\n",
    "            for seg in data_dict[dataset][\"merged_3D\"]:\n",
    "                new_segs.append(smooth_segments(seg, radius, do_3D))\n",
    "                new_shells.append(get_shells(new_segs[-1], 1, 1))\n",
    "            data_dict[dataset][\"smoothed_3D\"] = np.stack(new_segs)\n",
    "            data_dict[dataset][\"smoothed_3D_shells\"] = np.stack(new_shells)\n",
    "        smoothed_aucs[i,j] = compute_auc(data_dict, stage=\"smoothed_3D\")\n",
    "        smoothed_shell_aucs[i,j] = compute_auc(data_dict, stage=\"smoothed_3D_shells\", gt_stage=\"gt_shells\")"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%time\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.image.AxesImage at 0x7fd2267e2130>"
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(smoothed_aucs)\n",
    "#_ = plt.xticks(np.arange(len(dim_setting)), radii)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0.76053898 0.75219781 0.76417416 0.77177641 0.77750107 0.7853068 ]\n",
      " [0.82257734 0.82357165 0.82108248 0.8194591  0.81735886 0.81761251]]\n"
     ]
    }
   ],
   "source": [
    "print(smoothed_aucs)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC before merging 0.8225773358585858\n",
      "Best combination is do_3D=False and radius=1\n",
      "Best AUC after merging 0.8235716540404041\n"
     ]
    }
   ],
   "source": [
    "max_auc = smoothed_aucs.max()\n",
    "\n",
    "idx = np.argwhere(smoothed_aucs == max_auc)[0]\n",
    "\n",
    "\n",
    "print(f\"AUC before merging {compute_auc(data_dict, stage='merged_3D')}\")\n",
    "print(f\"Best combination is do_3D={dim_setting[idx[0]]} and radius={radii[idx[1]]}\")\n",
    "print(f\"Best AUC after merging {max_auc}\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.image.AxesImage at 0x7fd227486730>"
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(smoothed_shell_aucs)\n",
    "#_ = plt.xticks(np.arange(len(dim_setting)), radii)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0.48730441 0.47606872 0.50632835 0.51697336 0.53016555 0.54300713]\n",
      " [0.63600909 0.63389644 0.63199856 0.63046762 0.63046762 0.62837302]]\n"
     ]
    }
   ],
   "source": [
    "print(smoothed_shell_aucs)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC before merging 0.6360090863997114\n",
      "Best combination is do_3D=False and radius=0\n",
      "Best AUC after merging 0.6360090863997114\n"
     ]
    }
   ],
   "source": [
    "max_auc = smoothed_shell_aucs.max()\n",
    "\n",
    "idx = np.argwhere(smoothed_shell_aucs == max_auc)[0]\n",
    "\n",
    "for dataset in data_dict:\n",
    "    pred_3D_shells = np.zeros_like(data_dict[dataset][\"gt\"])\n",
    "    for i, seg in enumerate(data_dict[dataset][\"merged_3D\"]):\n",
    "        pred_3D_shells[i] = get_shells(seg, 1, 1 )\n",
    "    data_dict[dataset][\"merged_3D_shells\"] = pred_3D_shells\n",
    "\n",
    "print(f\"AUC before merging {compute_auc(data_dict, stage='merged_3D_shells', gt_stage='gt_shells' )}\")\n",
    "print(f\"Best combination is do_3D={dim_setting[idx[0]]} and radius={radii[idx[1]]}\")\n",
    "print(f\"Best AUC after merging {smoothed_shell_aucs.max()}\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# parasites"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "outputs": [],
   "source": [
    "# load rbc and parasite predictions\n",
    "\n",
    "data_path = os.path.join(get_path(), \"parasite_labelled\")\n",
    "\n",
    "datasets = [\"mask-ts\"]\n",
    "\n",
    "seed = 0\n",
    "n_folds = 10\n",
    "\n",
    "eval_str_para = \"data_mask-ts_by_stack_mode_3D_iso_min_train_masks_0_seed_0_epoch_499_aniso_3.2\"\n",
    "eval_str_rbc = \"data_valid1_valid2_valid3_by_stack_mode_3D_iso_min_train_masks_0_seed_0_fold_0_of_1_epoch_499_aniso_3.2\"\n",
    "\n",
    "\n",
    "data_dict = {}\n",
    "\n",
    "for dataset in datasets:\n",
    "    data_dict[dataset] = {}\n",
    "\n",
    "    imgs, para_gt, file_names = load_data(data_path, dataset)\n",
    "\n",
    "    data_dict[dataset][\"imgs\"] = imgs\n",
    "    data_dict[dataset][\"para_gt\"] = para_gt\n",
    "    data_dict[dataset][\"file_names\"] = file_names\n",
    "\n",
    "    para_pred, test_idx = load_test_predictions(data_path,\n",
    "                                     dataset,\n",
    "                                     file_names,\n",
    "                                     eval_str_para,\n",
    "                                     seed=seed,\n",
    "                                     n_folds=n_folds)\n",
    "    reverse_idx = np.argsort(test_idx)\n",
    "    para_pred = para_pred[reverse_idx]\n",
    "    data_dict[dataset][\"para_pred\"] = para_pred\n",
    "\n",
    "    rbc_pred = load_predictions(data_path, dataset, file_names, eval_str_rbc)\n",
    "    data_dict[dataset][\"rbc_pred\"] = rbc_pred"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "outputs": [
    {
     "data": {
      "text/plain": "(23, 29, 150, 150)"
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_dict[\"mask-ts\"][\"para_pred\"].shape\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "outputs": [],
   "source": [
    "# postprocess rbcs\n",
    "\n",
    "for dataset in data_dict:\n",
    "    data_dict[dataset][\"rbc_pp\"] = post_process(data_dict[dataset][\"rbc_pred\"], stage=\"rbc_old\", radius=1, do_3D=False, nbhd_size=1)\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.image.AxesImage at 0x7f174f613cd0>"
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 4 Axes>",
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAigAAACcCAYAAABVyVDPAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAA9hAAAPYQGoP6dpAABTNElEQVR4nO29a5Acd3k9fObWM909t71Iu1rLF8kWMWCHixJMKBybAE75LSCEBFJQRUhVUuA4gehPKArKHzDEsQlVcfjgkEuFCqRSYBcpDKmQSrB5EwOvXwgxJq+wgZggIVnWarWX2Z6Z7p5rvx8259lnVhdrpV3tzO5zqrasne3p6W79rN/Z85znPKkkSRIYDAaDwWAwDBHSW30BBoPBYDAYDGthBMVgMBgMBsPQwQiKwWAwGAyGoYMRFIPBYDAYDEMHIygGg8FgMBiGDkZQDAaDwWAwDB2MoBgMBoPBYBg6GEExGAwGg8EwdDCCYjAYDAaDYehgBMVgMBgMBsPQYUsJyqc+9Sns27cPhUIBBw8exDe+8Y2tvBzDFsDWgMHWgAGwdWA4E1tGUB566CEcOnQId911F5588kncfPPNuP3223Hs2LGtuiTDZYatAYOtAQNg68BwdqS2aljgTTfdhJe//OX4i7/4C3nthS98Id785jfjvvvuO+97+/0+nnvuOZRKJaRSqc2+VMMlIkkS1Ot1zMzMIJ1e5cSXsgYAWwejhrOtA1sDOwv2b4HhXGvgbMhepmsaQLvdxhNPPIEPfehDA6/fdtttePzxx884vtVqodVqyfcnTpzAi170ok2/TsPG4vjx49i7dy+A9a8BwNbBdgHXga2BnQv7t8Cg18C5sCUEZX5+Hr1eD1NTUwOvT01NYXZ29ozj77vvPnz0ox894/U//uM/hud5SJIEnU4HuVxO2DOFoSRJkE6n0el0kE6n0e/3kc1mkU6nkUql5Of9fh+9Xg+5XA4A0Ol0kMlkkM1m0el00Ov10Ov1AABxHMN1XXS7XWQyGeTzeflvKpVCJpNBoVAAADiOI9frOA7a7bZ8z3O2Wi25hkwmg16vJ98XCgV0u125n3a7LdfB6/J9H9VqFQCwuLiIJ554Aq1WC/l8HqVSCcViEZ7nIZvNIpvNynPh/fNa+JmdTgdJkiAMQ2QyGSwuLiJJEgRBgDAMEccxwjBEp9NBv99Ho9FAJpOReywUCti1axfK5TJKpRJ838cf/dEfoVQqXfQaON86+LV/fBtyfu6M12df28D014qYfW3jrOeb/lrxrK/zvdsBD//34a2+BEHQ6OPqlx+VdbCRa+DV+L+QxZlrwDBc6KKDb+KfN+3fAlsHw4+zrYFzYUsICrFWiuOmuRYf/vCH8f73v1++D4IAV155JcrlMjzPQ6fTQaFQQBzH6Pf7cBxHSAsAZLNZRFEkm7PjOOj1ekJWOp0OstmsvJ5KpZDP5wFAiEe73Ua/3wcAFAoFeW8qlYLjOEilUsjlcshkMuh2uygUCsjlcnItPFcul0OSJHLOTqcjxIaEJZvNyjna7TYKhQKiKJJn1uv1kCSJEJ5UKoVicWWzbTabQto8z4PneahUKhgbG5PnxOfc7XblmfCcJC3tdhvlchm9Xk/ud2JiAnNzc2i326jX63JNPEe/30cul5P75rPmtZ3t7/ZC18D51kHOz8EpOgPHPvfKOrKpHOZf10I2deY/WDPfOv//HFd9exzPvbJ+3mNGAeXS8DXqrf373Yg1kEXurH/PhiHD/xoKNuvfAlsHI4DzrIG12BKCMjk5iUwmcwY7npubO4NFA0A+nxfCoNFqtUQdaLVaol5QDUmSRFQJ1rqy2awoEKlUCul0ekDx4HvS6TRyuRziOJYHmc/n0el0EEUR4jhGoVBAr9cTEtLr9ZDJZOC6rhAPrYzkcjkhHbxeqi+dTkeup91ui+pDRYNkgsQqk8kgDEO57lQqhTiO0ev10O125R4zmYw8l3Q6LdfY7/fl+qgi8fmQ9LRaLVGU0uk0oijCzMwMut0uwjBEGIbodruoVqtCCLPZLDKZDCYmJuA4DiYmJoSgXMoaON86IJ57Zf15icfFHGvYHGzGGjCMHmwdGM6FLSEojuPg4MGDeOSRR/Crv/qr8vojjzyCX/mVX7ng82SzWdkYkyRBkiSyoZJo8Cufz6PX66HdbouKQVJRLpfRaDRkI+cXVRcSB278LJXEcQwAWF5eRpIkKJfLQgLK5bJ8rud5osy0Wi24rivKzNLSkpw/iiL5HH3tqVRKSlQkECRQfD0MQ7k/x3Hks13XRaFQQDqdFjLU7/eFlKRSKbTb7YFSFlEqlZAkCcbGxoQwkfDw+yRJ0O12hYjxc/ksU6mUKE+bsQaAlXIMf2taj+pxPpKyHdSTYcdGrgHD6MLWgeFc2LISz/vf/368853vxM/93M/hF37hF/DXf/3XOHbsGO64444LPkcmk0GSJEIoqJzwN3lulgDE58HSAzfUdDqNRqMhG3S324XjOOh2u0ilUoiiCMViUZQHAIiiCKlUSjbfdruNJEmk5MGSD0kJlY5cLjfgfQnDUM6hX9MKiPaixHEspRuWkqIoQr/fRxRF6HQ6cBxHPp9EIZVKiZfF9305P0kaS0UkbrlcTogYiRCVJpIzqjk8VyaTkXOwPMVz1etn3+w3Yg1cKs5FUma+VTKSchkwDGvAsPWwdWA4G7aMoPzGb/wGFhYW8LGPfQwnT57EDTfcgH/+53/G1VdffcHnoO8jjmMhK+l0Gq7rSuklnU7Lb/ss1VChYDmFmy2NtlQlaGgNgkCIQiqVEhUDgBzX7/elbMPyDr0b/X5fPpPv5bnoRO90Omg2m7LJx3GMbDYrZSWaT7Xfg/fIshMNtvl8Xso9VEXa7baQExIeGm7T6bQoPiQi9J6QUFERIfErlUry3Gn0BYAwDIXw8HXttt/oNbCZIHEZZaLyyzMvxb8+972tvoxzYtjXgOHywNaB4WzYUpPsnXfeiTvvvPOi309ioTdQKhHaX5IkCbLZrJhNdScL/wysKB9UAKIoks2fvg8AQgq0ysEumyRJEMcxisWilJ4ajYaYVjOZDFqtFrLZLBqNhry31+shiiIhCyQbLL+4rivki54Rfs/yCj+P16l9JbxmkpdcLod+vw/P80Qp4nMiKaHPhQoLz0lCMjc3J88uiqIz7oPXBUDudTPWgGH0YWvAANg6MJyJLSUoGwEaOrmZcyPudrsAIOQinU4jjuMBRUV383CDJeHh5u+6LsIwlE4c3SpMhYTdQ81mE4VCAWEYivpBRUQH0rTbbSnROI6DZrMpP+druh2a15fP5+W/mUxGSlP8fE0mWKbh/VEpyufzAy3LmtjxvSQ0uVxOjul2u/K99p5Q+SE5YmsyiQ+woqoMM6iQaMXESjwGg8GwtRhpgtLv91EsFuW3dSoU3NSpdACQ1l3+1s/Nlt01BFUZqgf0U+gNF8BAxwxBkkQyBEAUk36/L6UnqhssgVAR0d0+AOC6LuI4hu/7A1059Mv4vo9OpwPXdREEgbQH832NRkOICa+L5EQbgnV2DNuEaSCml4TEhmpRs9lEJpNBvV4XwqZLR8AK2Wq1WgiCYOP/8jcB2o9i5MRgMBi2FiNNULTiwfKINqICkO+58fI3fYa66bZillLoAaGKks/nB8pJ3KRZ7qDRlabcdrstITQkJPxsKh46+I0GU9d1AUDUDJZkSBAIfi7Po6+Fn5nNZiWXhSSEnUyFQkFMxfp4EpxutyuErN/vS5dTLpdDq9UayGhh1opWfrrdLuI4Rq1Wk2c4KjBiYjAYDMOBkSYoBBUI3VkDYKBMQiLBTRaAJMey3MP36Y4WXTKiMZVtwjTTAquhM/SgMCSOr+VyObiuK0SJGSM0kZZKJSmd0CPCMgvVDJIUkinf98XzwQ4ivq/ZbMp7G40GOp0OgiDAxMQEWq2WdPuQ4LAUxIRcPk+adPma9vXQW0MS2Ol0hNiFYYh8Po9Wq3XWNmPD5cOwG2UNBoPhbBhpgsJSCI2t9Etwg6f6wE0YWDWP0nOiPR7cbEly+F6CZISqiy4TARCfBklMLpdDqVRCJpOB53mikORyOVFaqNpQuaCPRifM0syrW6p1sFuSJGi1Wuh0OtKJlM/n8dxzz0mZ5dSpU/B9X0pIa42w2ndDlYclHd3C7TgOSqUSPM8bMNfys6MoQrvdRrvdljJQuVy+nMvCcBYYSTEYDKOG4cvBXgfYrhvHsXgsdLlHB51pg6iOrOfGzPeQqLDNV5MPbtLtdhtxHIt6oQdCU8Uh2aEnhqSgUCiIMuJ5npRh+PkkR3yd/yWhATDgVdE+GQCi0ACQjh2WkPg8+Gx0WJ0mcAxk034YKiblchnZbBau60qKL5WTWq2Ger2ORqOBOI7leet5RIatwy/PvHSrL8FgMBguGCOtoJBAsLOGRINKBjM5dDcL1QGqLsxKobcEgCS20vRK3wo3c01caFpliYXzblgK8TwPxWIRruti9+7dorgwYI1+DUbVa5OuTr6NogiFQkHKSixPsVRDIkH/iOM4Yo4NgkDKPCzZMOyNygzvhYSNz8BxHFFymExLAhcEAeI4xqlTp4SwUUFhpxFJmcFgMBgM68FIExQAYuYkeaDXBIAYOUk2SAjonWDbLhUDYDW5lb4UloxICqi+sDOHZReSB13uoGrBTZr/JRHSZlsSJqoZLPdks1m0Wi0pr5CwsJOH98RnwBbfbDaLiYkJMdIyBK5er0vZplgsCmHja1RleK38LBIaZsnweTMzhqZdkkBgdTiieVCGB1bqMRgMo4KRJijcVNeqACxpMNqe6ghzTahy6AGDazdS3/elNMMZOQCkw4flnkwmIxs8SzMsJ7muKx4UekV0dwxLSs1mUxQUXifTYGk0TafT0q7L4YhsW2Z30NrofapL1WoVURRJCYYEhu3LvFYdr8/SDJUYdhaxzMRzRVGEIAjkOmjk1WUd7eMxGAwGg+FCMNIEBcCAYVRPFNaKBEshfA2AqBPA6pweEh7G1euNlaqKTq2lokLFhsSExIjkpFqtDsTp0wcThiEWFxelXEI/TTabxdLS0hmkhtdLkrC8vIzx8XEhSTTTkrCl02n4vi/Thl3XxdGjRxHHMZaXl9Hr9eD7PjzPk+smGWEph/eVJAlqtRqWl5dRr9exsLCAer2OVquFOI4HTL1aUTrXsEDD1sFUFIPBMAoYaYLCjY/lGJ18qoPZ6BsBVn6zZ8eLHoC3NuSN6gkJA89L0sI2YRIgmly5qetcFCocLP+k02kJOAuCQHwo9KJEUYR8Pi/3x2F/YRhKl5E279JDo0PX6J+hojMzMwPHcXD69GksLS0JsfI8T6YV0ytCFURPgOYQw8XFRdRqNczPz4tyUigUBhJ4HccRBUcPUTQMD4ykGAyGYcdIExR24fi+L2UXGjw5cZgbOJULHeBGrwjJDP0VJBX0eJCUkPi0Wi1JWNVeE5KGfD6PsbGxM8LSms0mut0uwjDE0tISGo2GkCX6UfTUZPpj9ORgJseSTDFanmSAZaYwDFEsFtFoNJBKpbB3715Uq1Xk83nMzs5iYWEBx44dExK1uLiIUqmEycnJAQ8JSRlJFM22muTQo0P1hr6YtUMXDcMFdvUYUTEYDMOIkSYourxCZYFmTW7YOmmV3hT6T9jtoqPe2RpLNYTkRQ8EJOHRng1d3mGpg4oE5+Zw4nG73ZbPZ9T9WsWHQw3pDdGtyvTD8L8MhaP6o1umSdbq9Tqq1SquvfZauK4Lx3EwPz+PZrOJIAjQ6XSE0DAEDoAYbDl5uV6vS9mJhITgM2NuCu/XPCjDDVNTDAbDMGLkCQpLGFRMqJTw59z8uWGy+4XGVWDQg0Kio0PX9NRj3c5MPwqPi6JIUmRZYmGSq+d5iKJICIrjOGK+pdqgO5D0BGNg1XTKz+O96uvSxIzXSBWD6sfY2BgmJyeRJAlOnz6NU6dOSdswz1er1VCpVORz6/U60um0kCod3tZoNCRvhc+cnUTAKmkyDDdMTTEYDMOGkSYoLNOsLbHQW6IVEpZ3aHblxs8ofP7mD0AUAmDVHOu6rpATKgM8Jw2oWkGhqhIEgQwt1J4Sek+YusowNmClgyhJEjGaApBrpqeD5SaaU9kerAkLiVq9vjJfJgxD8ZxMTk5ibGwMs7OzOH78OA4fPowgCPDDH/4QwGq+i35+Ovp/bGwMACSoLUkSLC0tiWmZhIqlLcNowIiKwWAYFow0QaHnhEoGfSS6BMO5OyydMAYewEBHDRUBPTxQd+fQJ0ISQ9Or7hKiQZQm0bMltVKF0KUpzr9hFL6O4qdxttlsimIUBIEoJZpA8JqKxaIoRgBESWm321heXka320W1WsXU1BSAFZLRaDRw8uRJzM7OiprSarXQbDYHEmF1W3ehUBACp/09/DySlVEaFmhYgREVg8Gw1RhpgsJgNhIBejSoHnADJTmgiqFnyHBjpTmVvg3mkOgofBIV3W4MQLwYnucBWPVisNTCa9A+FgByPF9j4FkYhuj1ehIZz2wTlo1o4tWEgIoQgIE5QPTG1Ot1Mf1mMhns2rUL1157Lfbv349UKoU9e/bg1KlTOHr0KJ577jkEQYD5+XmZEUTCpgmYLh9pgzI7kbRh2TCa2Mh4/G7SAfCTDTufwWDY3hhpgsJSAjdFlkNINPSQu3w+L8qJJibc2PVUX7YWkwAAGPBRrPVVaAJEPwqD1PiaVk0ymQxKpZJ4Ung8N/0oiqT8wz/rOUO8Pm3e5dwfZqKwFEOPztjYGLrdLhqNBnzfR7vdxvz8vOSg+L6PyclJBEGAQqGAWq0Gx3EQhqF0H/HeWJLSM4H0JGhtkiURNGwswl+96ayvew9/+zJficFgMGwORpqgsLyhY+A5VE8P/gNWO150qzBLEfy5JhC6PEFio9UUfiYJgM4hyeVyEiOvlQ2qDAAQxzHCMBwgKCxLhWEIYJX4cKPnZ5JM6bwXllRIEMIwRKfTQalUEnXI8zz5PggCUZwAoFgsiuISBAGCIIDrumg0GlhaWhLCFMexXBuJoDbE6sh9Xr918Vw6zkVIzneckRWDwTDKGGmCwhIDSQbLGTpFlj8jaQEgEfdUWdj9wvKQnrvDc+nMEWCFHLEriF8kJwAGhg+yJVd303CAHycxkwCwo0ZPZGbcPRUWno+fo8PZGPCWJAmazSaazSaSJMHi4iI8z8PExATK5TI8z0M6nUaz2US1WhUlZHx8HNPT02g0GpienkYURZibm5M5PkyOJQEiwSL5oiGYhEpfn2F9uFBSciHvN7JiMBhGDSNNUPTkYpY5qDZQJaHiQHVFZ4ewLMJNXZdSNNlgmch1XYRhODDxl4RFB5Zps2scx2Lc1fknjUYDYRjKRk/FgZ1J7DjSSazc+Hms3vj1tWovCNUWRuuTkFUqFRQKBfHB+L4P3/dRqVSkBAVAslyYy1Kv11EoFMQo22g0RCHh3wefOY3JWkUyPD8ulZic75xGVAwGw6hgpAkKwVZbKhb0o7DDRBtg9c8ACAFwXXegzNNqtQaGEOoNHlid1Mv2YIafMXOFygvj99lu22g0JByNJRNmjJAAkKCwREX1RYei6cF9mpjRoEufCksyjKpnND0D66iwuK6LOI7FYMsWZybm5vN5FItFzM3NDfhRWEqj+ZjmWD5fEhvD82MziMm5PsOIisFgGHaM9M5BZYQkgJskVQmqIgBkWvDaOTN6CjD/zNd1aixbmDmUD1ghLY7jwPO8gfZblmtYKmLWyezsrBCQIAjO6CBqNpsD+SH8M7tmqK6QiJTLZbl2qj68Nga9VSoVIXCZTAZBEMjcn1arBd/3Ua/XRVVJpVIoFovwPA+7du1CHMcYGxuTichUnRqNBrrdLsrlsvhnWq3WwDXRdMwcFsOZuByk5Pk+18iKwWAYRow0QeGmzZJLkiSiFACQtlgqGix1ABDlQnfg8D26u4e+EBIcAEIQHMdBoVCQmUA8zvd9UXSomkRRJOUQfjavhZ/FcguNtwAGyIUeirjW0EsFhb4crcqQMOjgOnYLkajwHjhd2XEc+WyqK1Raut0ufN9HGIYDilEul5N7JJnK5/Ni+jUYDAaD4UIx8gSFZIJekrN15XAacb/fl9LFWuMmyYtuRaangh4Q+kyy2SzK5bIcq42y9KW0Wi3UajXUajWEYShkgORCd/5okHDooYRUU0hktLGXZIkdRDocjgZWekR03D+JQ5IkqNVqOHr0qPhLSIR4P8Vi8YzuJe1xyWazEufPkDg+TwblGVaxVarJuWBlH4PBMIzYcPfi3XffLSoEv6anp+XnSZLg7rvvxszMDFzXxa233oqnnnrqoj5LKws0czLtVBtZaYjVJII/p0+FnhL6KTgzR3tA8vk8SqUSXNeVe6NhlISDJILTiqmgsBxEI67OWeFnFAoFuK4r6kmpVBIFA1ht1yWByWaz8udisYh8Po9qtQrf91EsFmXoH/+89qtYLMrzarfbOH36NGZnZzE/Pz+Q0cJWYpqM2aHDa85kMtJW7TgOvvCFL+DXf/3X8ba3vQ3vfOc78e53v3vg720j14Bh47DRxOl/kqfwaPIP8vXv+PLAz20dbH/YGjBcCjZFQXnxi1+MRx99VL7XKsEnPvEJ3H///fjMZz6DF7zgBbjnnnvw+te/Hj/60Y+kc+RCQZ8HE1aB1U08m80il8sNDBFcm9uhszxarRb6/b6UMugdAVaJkJ6eTF8IyRDbm1k6qdVq6Ha7qNfrA6FuVHl0Wy6VFBIhPUmZIXMAJIgtlUpJqaVarQJY6cLhsbzHYrGIZrOJdDqNUqkkHhx6RrTqQvVkYWFB7pPn5rPs9/uSdqvj+PWYAGBFabryyivxsY99DJlMBrVaDe973/s2ZQ2MGoZNPdls+Cjj5fhFAEAXHfy/+Ff52U5eBzsJtgYMF4tNISjZbHZANSGSJMEnP/lJ3HXXXXjLW94CAPjsZz+LqakpfO5zn8N73vOedX3O6dOnUalUpISjMziYwko1giSBcfUABjbYMAzFAErDJxUCGl919gk7eDjbh2WdbreLIAhk5g0/A1glOsBqGi2/z+fzosDQHAtAyitafQFWCVK5XEa/34fv+wMmWpKyXq+HZrOJer0+QCBYsqHxF1jxu8zOzopnxnVdlEoleJ4npI3PgPH7zEBhp5Ie3FgoFAbC4DZjDYwSRoGcbHS5J4UU8qkVdTOTrP6ispPXwWbgYmYmbeQYg/PB1sDWgmvjcv19byQ2haA888wzmJmZQT6fx0033YR7770X+/fvx5EjRzA7O4vbbrtNjs3n87jlllvw+OOPn3NBtlqtAS9DEAQAVjtzCG64NGlyE06n08jlcqjX60II9DBBHseZMiQhAGRInjbj6tIK/SRUIRiO1m630Wg05DhgJVNEkyJ9nVQ8WDphGy/LO2dLreXrLG8BGJgxxAA4PUmZrdY8B3/GgYAMsctms6jVakin02g0GigUCgP5JmxjZoAdny8/++TJk7jzzjuRzWaxf/9+ub6LXQPnWwejgFEgJ5uBEA18PfknpJFGCWPy+kb/W7BTcSnDHC/XxrXRawCwdXA2PN9aWPvzUSAsG05QbrrpJvzd3/0dXvCCF+DUqVO455578KpXvQpPPfUUZmdnAUCm6BJTU1P46U9/es5z3nffffjoRz96xus6xp6qA8suJAZMZdXzac42t4cbOJUMqh30bwAQ4yjVAZIBlojq9ToWFhYkwp7g8VqF0QqK9q7k83k4jgPf95HNZuG67kALM5UiEilt5l2bKsuAORIiGmDZccNykOd5aDabKBQKmJ+fl2nGrusiiiLxtujPZvcOc2B4fK/Xw4EDB/C+970PU1NTaDQa+PznPw8AWFxcvOg1cL51YNh4bISSUsE4Xoyfh48SWojxEzwN4NLWga2BVWzUpGl9no3etDZjDQC2DoBL//s/2/uHjbRsuEn29ttvx6/92q/hxhtvxOte9zp85StfAbAi3RFn66A5Xxz6hz/8YSwvL8vX8ePHAax6IzzPE0Oo53nwPE/IRKVSkYTUcrksxtPJyUlUKhWMjY2hXC5jcnIS4+Pj8H0f4+Pj2L17NyYmJuD7PhzHkfOUy+WB0gUJDskDDbbssvE8T/wbJA1azaAiwc4aDvbzfV/+6/u+PCeSJZ2zwkySdrstHUMMa1teXhazbhRF0k7s+z4KhQJKpRLK5bKYabXyVK/XUavVpHTFMDteux5sSHS7XbziFa/AzTffjCuvvBIvfvGLxX/yuc997qLXwPnWgWE4MZnag6nUXhRTFUykpvCzeCWAS1sHtgZWNpaNIiebfe7NWAOArYPN/PsfJmx6Brnv+7jxxhvxzDPPiC+FzJmYm5s7g0Vr5PN5lMvlgS8AqFar8DxPJhWzM6VUKkmMOztc2KrLzhtuyNVqVcgJiQo7dXS3C9WMdDoN3/elxMJ5P0yFpWLDkgmw2vJMzwrLRSzbkLQwfG3tsEF2E5H40Pgax7GoN2u7apaXl6XNudFoyBcJy+LiIprNpqgvJD68nk6ng3q9PkBwlpaWAOCMIYg0FLOjiZH+VLOY7/I///M/F70GzrcOhh2jXN7ZyGvP/K9geynrYFTXwKhhszaqjVgDgK2DzcQwkZRNz0FptVr4wQ9+gJtvvhn79u3D9PQ0HnnkEbzsZS8DsKIEPPbYY/iTP/mTdZ97enoaruuKiVMPy3McR3wSnU5HSiTsjGEyLEkEsBrAxpZilk+KxaKEq9EYyth7hrF1Oh0Z/kczK1uGeW00k7I0QkMvsGKG9X0frutifHxcOoharZZkqOhQNtZcScRIglqtlsTX04/TbDZx7NgxNJtN/OQnPxESlclksGfPHlxxxRV49atfjSuuuAL5fB5zc3NCTFqtFubn56UdWQfa6e6kVCqFRqMhxIulJa2uTE9Pb/gaMIwO+lgxi9s6uHhczs3jX5/73oZL/rYGLh2XYw1sZtlvPdhwgvKBD3wAb3zjG3HVVVdhbm4O99xzD4IgwLve9S6kUikcOnQI9957Lw4cOIADBw7g3nvvhed5eMc73rHuz6pUKnBdV7wUOvad5ljdvstuGmBF2dEBZ47jSM4HlRZdzqCHhN4U3/clNp6pq7o7iMSIxzOvheZYdsQwDRdYJRvASqosFRM9xZhzcRzHGQincxwHYRhKiUl3L4VhiIWFBZw8eRJHjhyRkg07oPbu3TvQ0uy6rqg0/X5fykPVahVhGEpLtR7MqH09f/M3f4OXvvSlmJycxPLyMr74xS8CAN7+9rdv+BowbD7CX73porwo/538F3ZhBgV4aCv/ga2Di8Mw/WZ7obA1YLgUbDhBefbZZ/H2t78d8/Pz2LVrF175ylfiW9/6Fq6++moAwAc/+EFEUYQ777wTS0tLuOmmm/DVr371onreOQeH5QTGufO3dibMkmjozA4ek8lkxACayWSk1OG6rigEVEK0d6TZbAJYIRW1Wg31el2mFPf7fUlVZQcRc1a0ahJFkRAlmnNZXiHZiKJI2p719GROHKb35LnnnhMCs3fvXhQKBTz77LM4fvw4jh8/jsOHD0smDLuBqtWqdFuxw8nzPFSrVTEAx3GM+fl5yZq5/vrrhbhwnpEeD9BqtbCwsIAHHngAQRCgXC7jmmuuAQBcddVVG74Ghh2jXN7RuBiS0kKEw/g2OmjBQV46OHbiOhhVXKqKYmtg9LEZStqFIpXoGseIIAgCVCoV/MM//IMoDlQwuJEy1ZSttrrrhIpEPp+X9lmaQ+lrYVcQQ9KodpCE1Go1SYk9efIkarUafvKTn6DX6yFJEkl25WfQVMo2XJZkaFqdmJiQWirViHa7LS3LDG8rFosAIGm2QRAIQYrjGL7vS6nm8OHDQlCOHTsm76fCVK1WMT4+jj179uD666+H53lifK3X6zh16hSWlpaQyWRQqVQwMzODF77whRgfH0ehUEAYhpLUy5ZtKj3NZhPZbBbdbhdhGOKOO+7A8vLyhtaKuQ5uxa8gm8pt2Hk3EtuFoACXno3STTr4d3x5Q9fBKKyBjcBWqycbtUFtxhoAds46ALZmLWwkQVnPGhj5WTxsiWXLLdt72WJMLwaNqDR0ep4n5lOWafL5vKgMPB9zUKhUsM2YRISlJZZx6FthxolOs9VlJioPnCjM6+Z5eQ5tRmXmyNjYmKglHA5I7wvzV5rNJmq1GhYXF1Gv15HL5VAoFLBnz56B9FxmyVDdoQm4UCggiiLpDArDEEtLS2IGZgcV/TUkYAxuo9qkJywbDIb1Y6vJiWF4sFVrYatUlJEmKMeOHZPNVKeh0o9BlWJ5eVnIA/0ia2fw5PN5NJtN8XXk83n4vo9arSYzcnSsPs/HklEURRK4xi4hbvqe54nJFlglIfx5s9kUTwdLTzSYarJC4y6zSTKZjHTX8H6Xl5fRbrcRBAGefvppBEEgGSfT09O44YYbJK/k9OnTaLVayOVyWFxcFMLC/Jfdu3fD93384Ac/QK1WQ5IkWFpaguM4qFarAyZYdknxmgHINdHwaxhtXKwXxWAwbAx+eealO4qkjDRBocJRr9flNT3nhuUHqiG5XE7CxrjxA6tkgIpHq9WS0hF9H1RiaAilWqGHCrKUxEwUtiOvVV20b4NKhAYJD//L9zYaDQlxAyCKB9uM+/0+isUi6vX6Gam2unTE97bbbfl65pln5Bz0oJCE0JDbaDSwtLQE13UlQ4Z5KfS2MIOFxO1sk6N3EryHv72tyjyGnYmt9CEYdi42PQdlM0GCoP0P7J5hpwqJiiYnLEWwxAKs5pnobpk4joWAtNtt1Ot1OYb5KvSr5PN5mQ7MllwA0nJLLwo3cpZ72HXEUhU3dKo1zExJpVIS+sbSD7uPxsfHJbCOagv9K2EYiuJD0kDPTRzHCIIAx44dw9NPP43Dhw/jyJEjOHbsGObm5mTGDwPk6L9pNBpSDiMZ4XXq8g69M2sJ2E7DdlIdjGxdfhgxMOxUjLSCoifo0kSaSqVQLBYl0p2v604ZDvZLp9MoFouigHDYno60Z6dQpVIBADHe0vNBglKpVMTYSgWHrb6+74tBllHxergelSCWqajUaB8NPR6FQkG6aHhukiGWjU6dOoUgCDA3N4f5+XnJctm9ezcqlYrc5/LyMk6dOoVGoyHqDskGlaN0Oo1du3YhiiK0Wi2cOHEC6XQaQRAIISIB4TMnMWTZSsf+71RokjLqm7yVenYmTEUxXO41MNIERbcTk2Dk83m0Wi24ritKiJ6ZA0Bi5amkUB2gYsHyhx7OR28J/SidTge+70uAGq+DCgc3d272hUJB8kl4frY1c5NnZgnPReWExIdEi6UfKjfASiZMq9XC8vKyKCNBEMhE52KxKGbh+fl5LC4uIggCSa/dtWsXCoUCKpUK9uzZI11M9Ny4roswDKUcRvXH8zy5Fv6dUJnic9eBbobtRVYMlwdb6T0wGLYKI01QJicn4fv+QADb2pZidu5o4kA1g0oGzass5ehSheM4ohYUi0Xs2rVLovCZRVKtVsVnEYahEB9d3sjlcnI9VHm0ebTX68lGTxWHygrfx7JQrVYbKJswS4XJuFEUIQgC8dLQRDw7O4v5+XmZ16PHA/zsz/4syuWymGrpYWFpSifl0ovC1F7XdQcIVBiGch8kd4azg2TFiIrBYDAMYqQJSi6XQ7lcHsgXYekFgJRzSES4aQIrhICqBhWQOI5FHaFHhUqE67ooFApot9vYvXu3KCPASkx9t9uVTR+AkAmqDfw8fhZJEBUHKjW6XZoBdGsj+jnvhnkmvDcqLsViEZVKBbt27Rrw3ywsLIgCw1Tcfr+PXC4nQwn5rMIwRKPRkGGJYRiK5wXAgC+FLdskdCSF2kdjMBgMhkvHTlLTRp6gxHEsZRAaYhltT7WCxlkAZ8zmIRHh+TqdDlzXRRRFQiyYA8Khg61WSwZdATijU6VUKsm8G+aAkJjoOUFUVQBI+YTQZIbnpxrB+wIgpRYSFMdxMDk5KWoRsKIqLS4uIooiNJtNUYe63a60aZMo8bypVArlclnOy8/pdrsysJDdPnye7XZbuqZ08i6fveHcsG4fg8EwCricPpSRJigMTmOZhp07NLKStOjf6PlfJp2yJMEMk3Q6jUajAQDyZ76P82xc18X09LRMROb0YQAyu0a3HqfTaQlr08ZYduuwTFKv10WtoEeFs4F0CzMVFT2EkK2+AHDdddfh2muvxYkTJ7CwsCCR92EYYn5+XkoxbIHOZrO48sor5R5YGmLq7NLSkqglnA/E++RcHs4IIllhLswoh7T9+M9eiev+z7cGvgcw8JrBsJNgRlmDmWQvECQWzNugakCCQJMmyzh6onA2m5Xpu3qYn057ZVy+nmLMPJB2u41isYhqtSpTlavVqpRMmPLabDYHlAkqNDTnMhyNfhTmn7CcRP8IfSDMTeF96+h83icVjFQqJX4TqkBUdKiK8M9hGArZ44wfdgwtLi4OtA/z2XOGkb6ver0ufx80HLObaZhB8nEhr+vXNpKsmIpiOB92krRvOD92yloY/p3jPGDHCg2v3GwZGFYoFABASjXcgKkGcBNlngeVCia2AqtdLMz54GewQ4YzeTRR8X0fqVQKjUZD/DGaOJBIrPWfUPU52/wgJtwyjC2OY4yNjQ2UinzfF8WHCbOcF0SwdJTNZjE5OYk9e/ZgampKDMd8LiwlUa0hyaA/JQgCmTdElYiESptkh1VFORchudjzmKpiMBguJ7aKpFxOFW2kg9qSJEEYhgO+jE6nI7HycRyLF0TPzSHRYEnC8zzphNEmVpZYuBEDkLRZKi0MeKO6whIOSQqNtFQ6SEj4Z05KJokhCWHWCk2/VH94PsdxZKIyiQxVEJpTabTlNReLRRSLRRlKuHfvXlx77bW4/vrrUalUxPjKe6dHh0ZfYNVPohUWkhc+l1QqhXw+P+AJGhb8+M9euWHkZO15NwKWL2IwGC4U273cNjw7x0Wg1WqJiZMdPNwkudHqYXbc4NlyzKGB/A2fpRLP86QkRLC0whZgEp1er4fFxUVJWG00GvA8DxMTE5iamgIA1Go1ISF6uCA7c0haSIh0yzIj97XvhPfB+6PSwQ6aVCqFUqmEq666Cvl8HpVKRZ4PO34cx8HMzAxmZmZQKpXkeljGYbs0SR/nFvFa+v0+4jiWgDr+DFjtYKJ6MiwKymYQk7XnNyXFsJnYKdK+4cKxndfESBMUlh64qVYqFZkdo7NE1sbbM1WWmzLLJ71eD67rAoB0xOi5PgAGZurQa9HtdtHtdnHy5ElpzeVnTE5OwnVdxHGMMAyl1KO7aKIoQqlUknP1ej0pXdFnQy8Mr5PBblRKmIHCY/r9Pq655hpcccUVaLVa2L9/P4IgQLfblVk6Bw4cQLVaFa9IvV6XdmmSC95jq9WSDiBNSDgKwPd9IYN8Tkyo3ephgZtNTM72WdudqFiarMEwPLicJOVyqjYjXeLp9/vI5/My+4Ykgr4NADIwMJVKoVAoSMmChlMA4i9hJw+9IgDEK6JNt7pUQxIEQDpc6vU6giCQ7JOxsTE5jrODdAsxI+O1akM/DMs8PLbb7cpx7KBhhL8ekMiYfpZ0pqensWfPHlx77bU4cOAA9u3bh3K5LGSLgwV1Mm4qlRJ/DsmKVnvYMaXTckkWoyiS50cv0OXGZpVzLvSzLxajsPGPwjUaDDsJ27HcM9IEpVqtwvM8lEollEol5PN5lMvlAa+Gni3DycAkKwAGJgtTVWGwGDdq13Wl8yaTycg8G27UExMTUjZpt9uo1Wo4fvw4fvzjH0vZadeuXXBdF6VSaSAllvki/KKqQz+M7j4iQQFWO4x4PMtYjPfXLdjAirp05ZVXYvfu3ahWq8jn82g2m9KVRC8PPTG604g/AzDQzcPnQzUHwMCUaJaztgJbRUw24hpGoZNnFK7RYDBsLC43CRppgqLzTdhFwrkxBEsmVEx09giNqSQJa3NT2ALMziDGvrOVOJvNyiDAYrGI8fFxGTR46tQpnDhxArOzs1hcXEQqlcLMzIwcA0CIEj0pnFbMNmV6THgf7PBZa87luWhMpWdEZ73wnngeKjlsh9Zkg8MKeT7t8+H1kawx0ZbkKZvNyswiJtOSLF4uDAM5IS7mWkydMBgMF4PtpqKMNEFhaUYbYLkRAxgwaVId4cYOQN5HlYKqBIfr0SyriQ3PQVUml8uhVCpJeYgKBj0sOsEWwIBKAqyUhXiMViEADJRtqERQHeExruvKNfOL7yOx0OqHNujqbqFutyvPk6qSVnB04q4OveNzpDeGnh6SJKo/OxnDRJgMo4+t3IS22wZoGG6MNEEhAdGbOImDNnnSY6HVFh7L7hhu5FQlNOlhKSWXy0krMQft6TIKpwG7rotcLod6vY6jR4/ixIkTOHr0KHzfx9TUFHbt2gVgVQHiHBtgZfBfuVweKAUBGGg7DoIAtVptIN+FigdJBe+h1WpJVxPJkU61XdsJpAmJHnBI/w2w2jLd6XSQz+fFlMsuI5aHgFWzsWF7wVSerYURBcNOwEgTFB1m5nmemEVZ7qCSwe8BDBhL2eGzNshN+yyoBACQLhw9m6bT6aBQKEiGCKcDs0xSr9dx5MgRHD9+HEEQCOlhYJrukmFYGzd/Egv+mffcbrfRarXkekgm9HEkXfSHMJQOWJ05xONoNGZnEYkJr0mbhqnO6PlC7PrRZSgSJZ7zcmFY1Yr1XJf5OwwGw6VguxDYkSYo3CTpu9Dzaphuys2WZIUEhL4Llju42VKF4Aau24J5PD9Xh7tRheBmTLWG59UdOppI6eP1YEGSHvo+9EBDTVbohaESxPvWZSEew1Zk3/cHOpfoa9GkjOfR3UVUkUjy+Ozpo+E5eR1a3TJcOEnxHv62KRQGg+GisF3ICTDiOSi5XE4GBerSgud5iONYYthpaqXCQGMn59fQu0KVQM+qIUHRfhZu/AyK43EkN/V6HaVSSSLhO50OarUagiBALpeTMg7LNiQ4TL11HAetVkvuUXtUdJ4Lzarj4+MDBI3DBrX/g+WpbDYr5lndtsznFMexeGrY5RNFkTzndDqNcrmMSqUiRGctydIDGXu9nrQcXw6cLX9kWFUVg+FSsF2zLwzDia1YAyP9q2232xWvBn0YJAz8rX6tQZSlHP3bPwAp71B9Ac7MHaGyQW8IVQoqLHrQH1ULpsWGYYhTp05heXlZgtXGx8dFTSGR0O3R2lOjBwACK2UVkpjl5eWB6yZRoeJBrws7cADIc9BqDO+V83u0MZfv4fOimpROp+G6rowEYHmHxC9JEpRKpU1cBaMFI0sGg+Fy4JdnXrphpGKrCOpIKyjc+BhDz/ICE2FZAuF8Hh3GttZHwTZk+i50dDw/o9/vy8/ZoUMyEkXRwNA/zr6h8tBut3HixAn5+TXXXIN0Oo1Go4Hl5WXxq7A1F8CAOtNoNOR+SAKiKBKjKlUilp6azaaUsUi++CyowrDLhp1BLG0VCgVRVMIwlNIVjcHlclnuW8/p4WwgfhYJlI7B3wpoVWWUCMKwTje28tPOgqkno41LVdq28u9/3QrK17/+dbzxjW/EzMwMUqkUvvSlLw38PEkS3H333ZiZmYHrurj11lvx1FNPDRzTarXw3ve+VybovulNb8Kzzz677otfO4Om3W5jaWkJABAEAZrN5kACrO5c4Wsse/B8AAa6fJgFQo8JFRQqEjSGUglhiYbqw1oPhlZn+B6Wd5gnQlKj5wnpYDiSJZIknd9CPwlLTgDkPliuYcmo0+nIe3gfAEQJ4vEkKQDgeR4WFhbwyU9+Eu985zvxmte8Bl/72tfEbMvzff7zn8fv/d7v4d3vfjc+/vGPn/F3t1FrwLA1WEpO43vJ/4OvJ/+ER5N/wFxyYuDnSZLgf5Kn8PXkn/B/J1/Efyb/jiaCgWNsDVw6NnvzeL7zr3cdPIlvnnEOWwfDi60mp+smKM1mEy95yUvwwAMPnPXnn/jEJ3D//ffjgQcewHe+8x1MT0/j9a9/Per1uhxz6NAhPPzww3jwwQfxzW9+E41GA294wxvkN/ELBTdpEpVcLgfXdaWcQ5MqI/Fp+qRiwA2aygSNnfRnxHGMVqsl5EYnvrLzRU8a5ufrNFq2ItMbsri4iFqtJufkeXzfl9LR2swUYDWhlWQGgMy60e3RJCt8H7uF6Auh0hTHsYS56cA3moj5rFgG4nvpe9m/fz/+4A/+AMBqSYnhbw899BC+8pWv4Ld+67dw9913Y2xsDAA2ZQ2sF8MwI2eUVJxzoYcuiqjgerzsrD//KX6EY3gG1+NleAVeizwK+C88PnDMVq2B7Yat3ETWuw4crJSIh+Hfgp2Ei1kjW01OgIsgKLfffjvuuecevOUtbznjZ0mS4JOf/CTuuusuvOUtb8ENN9yAz372swjDEJ/73OcArPglPv3pT+NP//RP8brXvQ4ve9nL8Pd///c4fPgwHn300XVdi+6G4UaqE14ZtEavhJ7Zww2bxlOdA8LyhiYIcRzL+UlKdDlEtwRToaE3plwuo1wuA1ghFQsLCwiCQAynu3fvHpgDRALCRFaWU0qlkhAbqhq1Wg0LCwsyMweAkKpCoYAoioTEcLAgCRw9OlR7dDs1s1YWFhbQ6XSEAI6Pj+PgwYN4z3veg1tuuUX+LjjHp9Vq4ctf/jLe/OY348Ybb8RVV12F3/md3wEAfOELX9jwNXAxGCWSMmwdPbyWydQeXJe6AbtTV5xxTJIkOIYfYx+ux+7UFSimKngxfh49rG44W70Gths2YzO5kHOudx288H+JzLD8W7CTcKGelI30rlwqNtQke+TIEczOzuK2226T1/L5PG655RY8/vjKb09PPPEEOp3OwDEzMzO44YYb5Ji1aLVaCIJg4EuDBIWkgnN32BVDQ6sOXmNImw42o1KgA+DYhcLNnCUhqguaHHEj1502AAbOzZ+z7ZelF53dwkwVEh1eu05kpV9EG13jOB5IqG02m8hkMqKWcBSATrxlOYp+Ef48DEOEYSjEhIMHi8UiJiYmUCqVhHTpgYzPPvsslpaW8KIXvWhgRg8A/Md//MdFr4ELWQeGzcWFEqUITbQRYxxT8lo6lUEVk/K9rYGNBzeWjdhgNmKDOus6wMovfPZvwdbhXGtkmIgJsaEm2dnZWQDA1NTUwOtTU1P46U9/Ksc4jiOyvz6G71+L++67Dx/96EfPeJ2qB8mD9npQLWEJh1khnU4H9XpdFBBmdmgVYa0nBcDA5N61IWWVSkUUFqam8toKhYJ8Jrtp4jhGs9mUmPpisYh+vy9BbrrUxM/jMSQmJE+NRkPIjOu6CIJAWpfZhURvDLBKbFiqyWazEnUfx7F4WMIwlNfz+Tx27dqF3bt349prr0WxWITv+3JOEqB6vY7Tp08DWPGqAJCSGwCcOnXqotfA+dbBxYAqynYotwwb2lhRAPMYnGLtIC9/HoY1sN2hN5vL1Y6sca51AAzXvwU7GcNGSNZiU9qM106w1dkh58L5jvnwhz+M5eVl+Tp+/DgAyG//erAdQTVC+1BIZKhK6NwT3aKrPSf8nkFmzWZzQJGpVquSj0LjLaca689nmSaTySAMQywtLWFpaWnAzMqJyFRc9PRiXifbjXVM/trcESbTstylW5+B1RZmdunEcYwwDNFoNOTPURSJ0uK6LjzPQ7FYHGjN5vMmWeI8Hv0ZJEtnWxfrWQPnWwejivWQo60u81yuz99pa+By4ULVlcv1W7T9W2C4EGwoQZmengaAM5jv3NycqCrT09MD3TZnO2Yt8vm8+Di0nwNYDS8DBjdMYHVDp9+CRlYaR3mMTpLVRlqWhFgG0gP14jgWAsNOIUbmU22hisLrJMlxHAeFQgHFYhHdblf8G91uV9qV+Vme5w3MyKEHhmoNy0ls8WUHD8tJzIbRBCqOYzQaDdTrdYRhKN1OJGBUYYAVH8yuXbswNTWF8fFxmTXEzwFWBx4GQSCt31EUyfRnEsHdu3df9Bp4vnVwsRgGP8qwY73khEbI1v/+Bk200ZI/D9Ma2Ik4l8S/kTjXOgCG898Cw/BhQwnKvn37MD09jUceeURea7fbeOyxx/CqV70KAHDw4EHkcrmBY06ePInvf//7csyFgps9I+fZTsuuGc/zpMTA9lwSFAaMkazogXn0Y5BQcPYNyyIcGEhSwnIRSQVLGvy59nGwZFOv11Gv10VB4cbOnBSWXJaXl0VZ8TxP8lV0VD1VkEajIb4RdiqFYShJtqdOncLJkycxOzuL2dlZLC4u4rnnnsPc3ByWl5exuLiIer0urcuTk5PYv38/fuZnfgYvfOELcc0112BsbEyUIZKYOI4lbZYps//1X/8lRIq/Cb3iFa/Y8DWwERgVknK5VZSLNei68OGggEXMyWv9pI8a5uX7YVsDOxUb5Vk5G866DrDyb+Ow/ltgGC6s24PSaDTw4x//WL4/cuQIvve972F8fBxXXXUVDh06hHvvvRcHDhzAgQMHcO+998LzPLzjHe8AAFQqFfz2b/82/vAP/xATExMYHx/HBz7wAdx444143etet65rYUsvu2l0DogOCvM8T/5MdWV8fFyi3PWcHhIRloJIXBiMxjh73eHDsgbLPtyQ+V8qLdqjUqvV4Loupqam4Hme5KJQRQFWNn7P84Tk0DRLw6zjOIiiaCAHpt/vyzRlqjKM1OeAQW2W5bkBiFmXJSnf91GpVFAul88YbHj06FEhYnNzc/jRj34k5On222/HP/7jP2JmZgb79u3DQw89BAB461vfuuFrYKNw3f/5lvlRFJ6PmHSTLiI05PsITdSTGnJwUEh5uCq5DkfxQ3hJER6KOIIfIoMMelgpWQ7jGjCsH+tdBz/B0wCG+98Cw/Bg3QTlP//zP/Ga17xGvn//+98PAHjXu96Fz3zmM/jgBz+IKIpw5513YmlpCTfddBO++tWvDsSd/9mf/Rmy2Sze9ra3IYoivPa1r8VnPvOZdSeOtttteJ4nZRIaRal8sHOGZlAdttbpdFAul8VrkSQJPM9Do9GQoDQA4m2hEbXRaMD3fdmwucGTZLDThwoHVQ7d089SEM2uLIW4rivvZTtwrVYTYsPzkPhw3k6328XJkyfl57w/nkMTFGA19p+Eg8SKalSpVMLExASuu+46aYNmOQoAvvvd7+I973mP3M9f/dVfAQBe/epX43d/93dx++23o9fr4W//9m/RbDZx3XXXAcCmrIGNxCiQFE0cNiNl9kIVkwCL+C6+Lt8/g/8PALAHV+PF+HlcjZ9BDz38EE+iizbKGMfP4lX4T/ybvGcY14BhfVjvOihhxQw77P8WGIYDqYS/Po8QgiBApVKRRazLHfSj0K9BEkG1wHVdtNttAKslmPn5eSRJIoqKNnt6nifnymQycF1Xou4JlmnCMASw6n2hOkMSUKvVJFRt9+7dGBsbww033CD5Kf/93/+NKIpw+vRphGEo103iwSGCa+8TWMkuYaQ9sDq4jyoKv/gelqXWhsrt3btXhhnu378fruuiWq1KSYy5KmEYDnhPeF/0umiTcr/fx1vf+lYsLy9vaK2Y6+BW/AqyqdyGnPNyE5RLLS9tFEm5HOWjbtLBv+PLG7oONmMNGDYPm7EGAFsHo4T1rIGRnsUThiEmJibEO8KyC9UPqhskE+xooWnTcRzZ2Gls1VOBOe2YXSvsmnEcB/V6XfwruVwOQRCckQDreR6azSYKhQKazaaEsVHd0Me2221p3aV/hr4ObvwkPOSUJDYkYfTGsOVZT29mCBzfx64bfu/7PsbHx7F7925Uq1VUKhWMj4+Ll4XlJLYTsxzU6XQkJK7ZbML3/YGYfz7bUcEoqCgaa4kFCctWd/0YDAbDpWKkCQoJAbCaU8LSBf/LQXs6hI2bOt9fKpXQbrfFx0Lfhi4LaR8JsOo3ITnSU4j5OsmG9qto46jjOKhUKkIY6OFYXFwUksFrpL8EWE21DcNQ7okdSgAkM4X3Q4LGMDV+Nk3CrutifHxcSAq7h3it9OJohQiAkLI4jiV0ThNEtjmPGkaNpGgYMTEYDNsFo7d7KLCTxHVdURioCpBQ6LA1ljRYDiJx0NH13HzXxtrTPMpzcjIwB+SxXKQHDpLosFykyzW+70tpiqUjbvYsj9DQyvNx3k0+nxdCRWKmSYGu2vHzSF4AyPGVSkXOy7a/iYkJ+Qx6YljWYeCbVnVIUnjvelAhz82/E4PBYDAYLhQjTVB0aBmwol60Wi1JMaWKwhyQtYZQPZEYwIC5lBsxP2fthGGaUVnOoPeChIBEhtOIqYDws2hcdRxHWqNZ4mFdrtPpCJnQHhqSMt6zjsnXAwjX5sJQPWI7drVaFbLFFux+v4+JiQkJbisUClIi4nUDkFlAfJ4Ms6NXR48cKBaLm7sQDAaDwbDtMNIEBYB4HEg66OdgSBlVC6ocjuMgjmMZuMdyie/7qNfrqFQq8jqJBX0fJBv8b6vVkpwVKgrsuNHmXGCVSAAYiNYHIAbbSqUi90LvjOd5CMNwIGOl0+mI2ZcDE3XgHHNg9BwfqkwkEFRe6C2ZnJwUEsV8Ft0JRFLG8hjLSwSJEUs6hUJBIvqfLzVyGDHKZR6DwWDYDtiUqPvLCe3N4G/xJANr80h05wrTXxmgxvZeHSymJxIDq+3BwKoqQbOr7uphWivB6+GgQc6euOKKKwbUHnbq0EhL5cT3fWnzzWQy8H1fhgryvfl8XtqfdTIu80x0ci7n//D9ruvK5/E+dQIvn4kmGlSu9PweqkgMwtPkaRQxKgFuBoPBsB0x0gqKVkcAiLFVt9qyDMTSDNtz9cA/HdueSqVk5gzVFT2UkGoKu31c1x0IV8vlcoiiSIyyLA1xoF6/34fv+3BdF6VSSQiDjrYngQnDULwq3W5XVAm2QZMwsOMolUpJqi6JBb00VJiY18JyjFZBoigSLw2JFwmdHgxIlYeqEUtd9NQQJDXa/2JYhREgg8FgODdGmqBw49VdLHrmDOfR6OAzPSCQv/VTqQAgs29ofuUgQe1X4bE0rerSDdUCqh0ApKREcqQ/n9cIrHpE2EHDLhyen0mzHJAIQMhNoVAQQqaNwPxiiUeXokg+9LXr/+qhhfoZkQSSfLBbiV1C2hTM0pPBYDAYDOvBSO8cOm+D/gj9G7yeZEzwz9pPAkDKHVRNtPmVxlJG6pMEMTRNZ7DEcTzQ1cLNXifbssRSLpdRKBQkgp5dO9zUSU5YLtH5JZ1ORzZ+Rv7zeyoXmnzpbh6qOyRNJCAkc1R9+OxI8FKplMT96+fIa2WKLw26qVRKhiIaBmHqicFgMJwfI+1B4WbOgDLd0UNVhRHv2hOhA8u4kTIVlRv62tRVhrexu4ekg6WOOI4Rx7GoN1RLGo0G4jhGrVZDv98Xr0ilUhGjbqvVkm4eqiY6jI1lmiRJhIywFMTrJ5EgMWEJimoQu26CIJDPZHKubo9mCzEAIX2celyv18Ugy2emJyfrZ0R1So8NGEVsBpEwcmIwGAzPj9HdORSoXNCXQc+DDlZb6z3RXhKWOnR2Cv0cjuOI0ZStvOzOYacLN38SG27QtVoNURQNJMBqZYXtvTTO8uckHJoIMcWVxKXX6wkh4vwfEg+en11MDKrjNbdaLelS0iUmGo4ZUEeCw+vgc+OzyufzaDabaDabZzw/PdBQK1ijCCMUBoPBcPkx0gRFz6Sh8kA1gyRDTybWMfb0TujsErYG08zKrhs9RVhv0DxOG2zb7baQkjiOEYahbOK5XA4TExPYtWuXTAgmOarX6wAgGzrJAQ2rVEpIVKj46NA0TSY4QJFKEP/M4YgABoLkdI6LVmBItvQ8IF4XAMlU4XUnSQLf96UEpfNYRhkbRVKM7BgMBsOFYaR3Dp3PwS4a13UHYudZhuAm67ouGo2G+FVYJuJ/+dt+oVCQLhcSEm7gOl9FJ7iSOMRxjEajgSiK0Gw2Rakol8viPdFtyVQu4jhGEATSKUO/Cq+DhIokhLOCqLroMDiqIq1WayBDhaoNlQ6SNe25YYicJmD6OfGz+T6SmXw+Lx4WKjuc42MwcmIwGAzrwUibZKMokswNYHWYH2fNsAyhjarNZnMgep3KA9uFdb4Jg9CoFmj1hJtuPp8XVYKlFJIGlnZIMHR4WrvdlsnIWsWg54QgKdFqD1UhZrdQ0dDlIW2OZblFq0n80vN+2GZMwyufgb5fYLVziMZZdhcBq8ZlYDWDZVRzUDYSRk4MBoNhfRhpgsJANB1Clkql4HmekAQqJwBko9WR9zpwTLcSc5PnscwOWZuLEkWRJM/qsgg7f8rlMlzXRaVSwd69e1GtVlEqlWRTp9pB5YZlJnbHkLywZMKMF/psWP5h+UW3XbMDSGemUPXgMSQaVECoLOmWYp5zbfqtTugFIO8rFApCaNbOBhplaJJxISmzRkoMBoPh4jHSBIXlCs7f0XNvAIj6wTIDg9x0VgczPLi58mdalQFWyU0YhigUCgiCQMyy9JzQC8JWYGAlV6VSqaBarSKXy4lng2WfpaUl5HI5ITnar6KD4nSXkuu6UtLSHhkeS2JGVYj3rg2wus1aB7EBEB9Kq9WSNmiei6UiEpG1AxtJnGj+LRaLEvq2E2CkxGAwGDYGI01QoiiSzVQHn+k8FMbVc5PXRlmqAiQCJCCMvqfSAEBMtd1uF/Pz8+LT6PV6aDQaQoby+Tw8z0OhUIDneahWq6hWq3AcB+Pj43ItzWZTSj2MtqeaooPWSCRIIkgYWJphci1VDM/zEMexECHOCqJfhiUekjXdYkw1iqoLVZJKpQJgxd8ThqGMCKDyw/ZtKiftdhvtdltSa7ejB8WIiMFgMGwuRpqg8Dd5pphqpYGbPTdRKgucvcPyCksxOngtiqKBdNhutyutxDoXhb6RUqkkM3PofymVSnAcR+bg6KGCbNcNgkCukW2/umRCP4suK/F1EpgoioSo8H44NFGDz4Ukjs+Mz1GXvFhOImmj6ZX5K9oTw3PoeUe8HnpvtkMXj8FgMBguL0aaoHAzporCzZcqAYkJN1f9PrbScvOMomig7MNyD42uURRJpw6PYzmIBlSWmUhQONiPxwKQllyqGCQ5ukNGEyBt8mU5hkZfEgrmwJBMFAoFUVGolLA0Q3KkCQaP08MDAUgZqlgsynlIjjgVWqfKssRTKpXErKuj+g0Gg8FguFCMNEFhSimNnvRD0IDKDZVlHyoFHNDHMg4VAnol6PXg5sqkV5ZaSH6KxaJ4PZhUS19MtVoVpSKbzQ4M8ev1eqjX60Ia6F/RPhYSkkajAcdxBrJYSFroJ+EEYRINKkDMgQnDUJQOqh/0rvBczGXhZ/M+NbHiYEBgRVkqlUoAVlN7SVy63a6QNYu5NxgMBsPFYKQJSrPZFDWDv/3TNKuj37PZrMzIoTG0Xq9LKWitF4MbdDqdFl8FN2BNDHTUvOM44rVwXVfIUaFQEAWDpIAeD3o1SJIYMMfP0v4OXhew2nqs24hJuDizh8cBq3H9AIR00GxLBYVqSSaTQT6fl5ITgAElSp+XZTNgtVuHc3s4t4ilL4PBYDAY1oORJih64q5WGlgy0YP8uKlzE2d5Qw8a5MZOE6je1JlZwu8544ZKA3NNdOcMAFEc6MlY21LMsg67cvTEY2BwqCEJBNugCR2JTw8JO5f00EGqMvThsCWb3hyWhYDVIYlUkJiLotN2dSsxiWE6nUapVBow3hpBMRgMBsN6MdIExfM8MZKy5VXHxwMQ82ipVBLvBokL388yDFUF+lL4OskBv2fZiBs9SY6OdueXLpNw7g83f5ZTSErCMJR7Yxvv2gnNwGpCLktEPB8JE/0iACSun1099KzopNdcLoeFhQXpEKJHJQxDOZ5/Zt4KACFVnudJuzRbkYvFInK5nLRMGwwGg8GwHow0QaHhlMFn/A2exIMlECoi9FnQh0GDJ7tVqIrwewCyuZOQ6JION3e+V5d4dJkGgJRy1iohenggiUgYhnBdd8CAC0DOpZUfHcRGD40OadOpubr7hiUcrR4xbI6enFwuh3q9LsSH7cwMdiMJ43RjnovXwuvXxMtgMBgMhgvBSBOUqakpeJ4nvhD6KbgxkoAwRAyAGEOpMLCEwo2biglJAUshzCCJokiG4bFTR3s8OLRQh7VRVdABcNzgqdywRJPJZFAsFqX8Q6IFrJpRtWeE96DJk+u6A8ZZqjTsGOKAQWaa6HlBVHnozQGAxcXFgblEfK5UgDSZi6JIPCye5yEMQ1FaDAaDwWC4UIw0Qcnn8/B9XzZeqg56Hg0zTnTSKsmC3lhZJtF5HyQNHNjX7/fldcbO8/NJbprNpnwWO4e08ZVkRaerkpiwxKI9M9qgyoRYmk+pYGg/jO/7ZwTRAZB74bOiUqI9OsyOoQrC6y8UCgMt2CQ7JHMsN7EVm+RODzc0GAwGg2E9GGmC4rqubIYkDiQTYRhKu7Fu8QUwMBBPm01JQpjwSlLQaDREgWFiLJUNbWqlmsJjS6XSABHiRt7pdOC6LoIgkGugsReAlJDY3cOuIpItKiYsLzEDhfdAgy9JiSYPHGJIogFA7oP3TZC86BC8temxfI1Eia3ejuOI/8RyUAwGg8GwXowkQdHtr/zNnjNyOM+GZZ10Oo3l5WUUCgUhAexuqdVq4r3QM3h0RwzD1AgqNDrgjRs7W4pJJubm5iS0jGRHtxiT/LA0Qi9Is9kcuB6Wr3RJh8+A3hYSCLb48lz1en0gUI6JuCRgJB96QKImSywJUQnRShSfEbDafcSJzs1mE47jSFlM/71t9DroogNsj3mE2xpdrKyRjVwHtgZGC5uxBvT5bB0MP9azBkaSoNTrdQDAb/7mb27xlRjWg3q9LnN9NgILCwsAgG/inzfsnIbNx0auA1sDown7t8BwIWtgJAnKzMwMnn76abzoRS/C8ePHUS6Xt/qShg5BEODKK68ciueTJAnq9TpmZmY29Lzj4+MAgGPHjm3oP3bbBcO0BoDNWQe2Bs6PnbAGAFsHz4dhWgfrWQMjSVDS6TSuuOIKAEC5XN7yBz7MGJbnsxn/aLDkValUhuIehxXDsgaAjV8HtgYuDNt5DQC2Di4Uw7IOLnQN2JhZg8FgMBgMQwcjKAaDwWAwGIYOI0tQ8vk8PvKRjwwkpRpWsROez064x0vBTng+O+EeLwU75fnslPu8WIzq80klG93vZTAYDAaDwXCJGFkFxWAwGAwGw/aFERSDwWAwGAxDByMoBoPBYDAYhg5GUAwGg8FgMAwdjKAYDAaDwWAYOowkQfnUpz6Fffv2oVAo4ODBg/jGN76x1Zd0WfD1r38db3zjGzEzM4NUKoUvfelLAz9PkgR33303ZmZm4Loubr31Vjz11FMDx7RaLbz3ve/F5OQkfN/Hm970Jjz77LOX8S42DjtxHdgaGMROXAOArQMNWwPbdw2MHEF56KGHcOjQIdx111148skncfPNN+P222/HsWPHtvrSNh3NZhMveclL8MADD5z155/4xCdw//3344EHHsB3vvMdTE9P4/Wvf70MVwSAQ4cO4eGHH8aDDz6Ib37zm2g0GnjDG94wMLF5FLBT14GtgVXs1DUA2DogbA1s8zWQjBhe8YpXJHfcccfAa9dff33yoQ99aIuuaGsAIHn44Yfl+36/n0xPTycf//jH5bU4jpNKpZL85V/+ZZIkSVKr1ZJcLpc8+OCDcsyJEyeSdDqd/Mu//Mtlu/aNgK0DWwO2Blawk9eBrYEVbNc1MFIKSrvdxhNPPIHbbrtt4PXbbrsNjz/++BZd1XDgyJEjmJ2dHXg2+Xwet9xyizybJ554Ap1OZ+CYmZkZ3HDDDSP1/GwdnB22BmwNADtnHdgaODe2yxoYKYIyPz+PXq+HqampgdenpqYwOzu7RVc1HOD9n+/ZzM7OwnEcjI2NnfOYUYCtg7PD1sDo3cdmYKesA1sD58Z2WQMjRVCIVCo18H2SJGe8tlNxMc9mVJ+frYOzw9bA6N3HZmCnrANbA+fGqK+BkSIok5OTyGQyZ7C7ubm5M5jiTsP09DQAnPfZTE9Po91uY2lp6ZzHjAJsHZwdtgZG7z42AztlHdgaODe2yxoYKYLiOA4OHjyIRx55ZOD1Rx55BK961au26KqGA/v27cP09PTAs2m323jsscfk2Rw8eBC5XG7gmJMnT+L73//+SD0/Wwdnh60BWwPAzlkHtgbOjW2zBrbGm3vxePDBB5NcLpd8+tOfTp5++unk0KFDie/7ydGjR7f60jYd9Xo9efLJJ5Mnn3wyAZDcf//9yZNPPpn89Kc/TZIkST7+8Y8nlUol+eIXv5gcPnw4efvb357s2bMnCYJAznHHHXcke/fuTR599NHku9/9bvJLv/RLyUte8pKk2+1u1W1dFHbqOrA1sIqdugaSxNYBYWtge6+BkSMoSZIkf/7nf55cffXVieM4yctf/vLkscce2+pLuiz4t3/7twTAGV/vete7kiRZaS37yEc+kkxPTyf5fD75xV/8xeTw4cMD54iiKPn93//9ZHx8PHFdN3nDG96QHDt2bAvu5tKxE9eBrYFB7MQ1kCS2DjRsDWzfNZBKkiS5fHqNwWAwGAwGw/NjpDwoBoPBYDAYdgaMoBgMBoPBYBg6GEExGAwGg8EwdDCCYjAYDAaDYehgBMVgMBgMBsPQwQiKwWAwGAyGoYMRFIPBYDAYDEMHIygGg8FgMBiGDkZQDAaDwWAwDB2MoBgMBoPBYBg6GEExGAwGg8EwdPj/AYstVoPMgRG0AAAAAElFTkSuQmCC\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "img_id = 4\n",
    "z = 15\n",
    "dataset = \"mask-ts\"\n",
    "\n",
    "fix, ax = plt.subplots(1, 4)\n",
    "ax[0].imshow(data_dict[dataset][\"imgs\"][img_id][z], interpolation=\"none\", cmap=\"gray\")\n",
    "ax[1].imshow(data_dict[dataset][\"rbc_pp\"][img_id][z], interpolation=\"none\")\n",
    "ax[2].imshow(data_dict[dataset][\"para_gt\"][img_id][z], interpolation=\"none\")\n",
    "ax[3].imshow(data_dict[dataset][\"para_pred\"][img_id][z], interpolation=\"none\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "outputs": [],
   "source": [
    "# merge everything inside main rbc"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "outputs": [],
   "source": [
    "def post_process_para(para_seg, rbc_pp_seg, radius=0, do_3D=True):\n",
    "    para_seg_merged = merge_in_rbc(para_seg, rbc_pp_seg)\n",
    "    para_seg_smooth = smooth_segments(para_seg_merged, radius, do_3D)\n",
    "    return para_seg_smooth\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "outputs": [],
   "source": [
    "for dataset in data_dict:\n",
    "    data_dict[dataset][\"para_merge\"] = merge_in_rbc(data_dict[dataset][\"para_pred\"],\n",
    "                                                      data_dict[dataset][\"rbc_pp\"])\n",
    "    data_dict[dataset][\"para_gt_merge\"] = merge_in_rbc(data_dict[dataset][\"para_gt\"],\n",
    "                                                         data_dict[dataset][\"rbc_pp\"])"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.6679841821843927\n"
     ]
    }
   ],
   "source": [
    "# only run this for mask_ts  for getting the results of that dataset direct, as requested by Sophia\n",
    "smoothed = []\n",
    "for frame in data_dict[\"mask_ts\"][\"para_merge\"]:\n",
    "    smoothed.append(smooth_segments(frame, radius=0, do_3d=True))\n",
    "data_dict[dataset][\"para_pp\"] = np.stack(smoothed)\n",
    "print(compute_auc(data_dict, datasets=[\"mask-ts\"], stage='para_pp', gt_stage='para_gt' ))"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 6 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "img_id = 1\n",
    "z = 15\n",
    "dataset = \"mask-ts\"\n",
    "\n",
    "fix, ax = plt.subplots(1, 6)\n",
    "ax[0].imshow(data_dict[dataset][\"imgs\"][img_id][z], interpolation=\"none\", cmap=\"gray\")\n",
    "ax[1].imshow(data_dict[dataset][\"rbc_pp\"][img_id][z], interpolation=\"none\")\n",
    "ax[2].imshow(data_dict[dataset][\"para_pred\"][img_id][z], interpolation=\"none\")\n",
    "ax[3].imshow(data_dict[dataset][\"para_merge\"][img_id][z], interpolation=\"none\")\n",
    "ax[4].imshow(data_dict[dataset][\"para_gt\"][img_id][z], interpolation=\"none\")\n",
    "ax[5].imshow(data_dict[dataset][\"para_gt_merge\"][img_id][z], interpolation=\"none\")\n",
    "\n",
    "for axes in ax:\n",
    "    axes.axis(\"off\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.6916995915499601\n",
      "0.6916995915499601\n",
      "0.6956521814519708\n"
     ]
    }
   ],
   "source": [
    "# mask -r and mask-ts\n",
    "print(compute_auc(data_dict, stage='para_pred', gt_stage='para_gt' ))\n",
    "print(compute_auc(data_dict, stage='para_merge', gt_stage='para_gt' ))\n",
    "print(compute_auc(data_dict, stage='para_merge', gt_stage='para_gt_merge' ))\n",
    "# these are not the original numbers anymore"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 268,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.4236110773953525\n",
      "0.45833336223255505\n",
      "0.5075757720253684\n"
     ]
    }
   ],
   "source": [
    "# only mask-r\n",
    "print(compute_auc(data_dict, datasets=[\"mask-r\"], stage='para_pred', gt_stage='para_gt' ))\n",
    "print(compute_auc(data_dict, datasets=[\"mask-r\"], stage='para_merge', gt_stage='para_gt' ))\n",
    "print(compute_auc(data_dict, datasets=[\"mask-r\"], stage='para_merge', gt_stage='para_gt_merge' ))\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.6916995915499601\n",
      "0.6916995915499601\n",
      "0.6956521814519708\n"
     ]
    }
   ],
   "source": [
    "# only mask-ts\n",
    "print(compute_auc(data_dict, datasets=[\"mask-ts\"], stage='para_pred', gt_stage='para_gt' ))\n",
    "print(compute_auc(data_dict, datasets=[\"mask-ts\"], stage='para_merge', gt_stage='para_gt' ))\n",
    "print(compute_auc(data_dict, datasets=[\"mask-ts\"], stage='para_merge', gt_stage='para_gt_merge'))"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 273,
   "outputs": [],
   "source": [
    "radii = [0, 1, 2, 3]\n",
    "do_3D_vals = [False, True]\n",
    "\n",
    "metrics = np.zeros((len(radii), len(do_3D_vals), 3, 2))\n",
    "\n",
    "for i, radius in enumerate(radii):\n",
    "    for j, do_3D in enumerate(do_3D_vals):\n",
    "        for dataset in data_dict:\n",
    "            new_segs = []\n",
    "            for seg in data_dict[dataset][\"para_merge\"]:\n",
    "                new_seg = smooth_segments(seg, radius, do_3D)\n",
    "                new_segs.append(new_seg)\n",
    "            data_dict[dataset][\"para_smooth\"] = np.stack(new_segs)\n",
    "        metrics[i, j, 0, 0] = compute_auc(data_dict, datasets=[\"mask-r\"], stage=\"para_smooth\", gt_stage=\"para_gt\")\n",
    "        metrics[i, j, 0, 1] = compute_auc(data_dict, datasets=[\"mask-r\"], stage=\"para_smooth\", gt_stage=\"para_gt_merge\")\n",
    "        metrics[i, j, 1, 0] = compute_auc(data_dict, datasets=[\"mask-ts\"], stage=\"para_smooth\", gt_stage=\"para_gt\")\n",
    "        metrics[i, j, 1, 1] = compute_auc(data_dict, datasets=[\"mask-ts\"], stage=\"para_smooth\", gt_stage=\"para_gt_merge\")\n",
    "        metrics[i, j, 2, 0] = compute_auc(data_dict, stage=\"para_smooth\", gt_stage=\"para_gt\")\n",
    "        metrics[i, j, 2, 1] = compute_auc(data_dict, stage=\"para_smooth\", gt_stage=\"para_gt_merge\")\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 293,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.4917929389260032\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 2 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# mask-r\n",
    "plt.imshow(metrics[:, :, 0, 0].T - compute_auc(data_dict, datasets=[\"mask-r\"], stage='para_merge', gt_stage='para_gt' ))\n",
    "plt.yticks(do_3D_vals, do_3D_vals)\n",
    "plt.xticks(radii)\n",
    "plt.colorbar()\n",
    "\n",
    "print(metrics[:, :, 0, 0].max())"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 294,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5151515007019043\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 2 Axes>",
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAj4AAAGJCAYAAACKFTn8AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAA9hAAAPYQGoP6dpAAA7FUlEQVR4nO3df1yVZZ7/8fcR5YAOHH8QHJlI0bHUwaaCBqHx10wilJnWJI3zoNwt0lUzRFfz10a24eqUNY2pWQ6W/dDp4VjOZAZuI5MjmvIVSzPXNiaoOKEOgjoKyrm/fxhnO/JD4OB9k+f13Mf9WLnPdV33dR92ls98Ptd13zbDMAwBAAD4gQ5WTwAAAMAsBD4AAMBvEPgAAAC/QeADAAD8BoEPAADwGwQ+AADAbxD4AAAAv0HgAwAA/AaBDwAA8BsEPgAAwG8Q+AAAcIVbsWKFoqOjFRQUpNjYWH3wwQdNts/Pz1dsbKyCgoLUp08frVq1ql6bjRs3auDAgbLb7Ro4cKA2bdrk9Xnv3r1ls9nqHVOnTvW0mThxYr3PBw8e3DY33YiOl3V0AAAgSTp79qxqamraZKzAwEAFBQU1q+2GDRuUkZGhFStW6JZbbtELL7yglJQUffLJJ7rmmmvqtS8uLtZtt92m9PR0vfrqq/rb3/6mKVOm6KqrrtLdd98tSSooKFBqaqqeeOIJjRs3Tps2bdL48eO1Y8cOxcfHS5L27Nmj2tpaz7gHDhzQyJEjdc8993hdLzk5WTk5OV73djnZeEkpAACX19mzZxXd6wdylddeunEzOJ1OFRcXNyv4iY+P10033aSVK1d6zg0YMEBjx47V4sWL67WfM2eONm/erEOHDnnOTZ48Wfv371dBQYEkKTU1VVVVVXr33Xc9bZKTk9WtWze98cYbDc4jIyNDf/7zn3XkyBHZbDZJFzI+J06c0FtvvdWs+24LZHwAALjMampq5Cqv1ReFvRUa4tsqk6qTbvWK/buOHTum0NBQz3m73S673V7vuoWFhXr00Ue9ziclJWnnzp0Njl9QUKCkpCSvc6NGjdKaNWt07tw5derUSQUFBZoxY0a9Ns8++2yDY9bU1OjVV19VZmamJ+ips337doWHh6tr164aNmyYnnzySYWHhzf5HfiCwAcAAJP8IMSmH4TYLt2wCW5d6B8VFeV1/rHHHlNWVpbXuWPHjqm2tlYRERFe5yMiIuRyuRoc3+VyNdj+/PnzOnbsmHr27Nlom8bGfOutt3TixAlNnDjR63xKSoruuece9erVS8XFxVq4cKF+/vOfq7CwsF4Q11YIfAAAMEmt4VatjwtMag23JKm0tLRexqcxF2dZDMOod+5S7S8+35Ix16xZo5SUFEVGRnqdT01N9fw7JiZGcXFx6tWrl9555x3dddddjc7PFwQ+AAB8D4WGhnoFPg0JCwtTQEBAvUxMeXl5vYxNHafT2WD7jh07qkePHk22aWjML774Qtu2bdMf//jHS95Tz5491atXLx05cuSSbVuL7ewAAJjELaNNjuYKDAxUbGys8vLyvM7n5eUpMTGxwT4JCQn12ufm5iouLk6dOnVqsk1DY+bk5Cg8PFy33377Jed7/PhxlZaWqmfPnpds21oEPgAAmMTdRv/TEpmZmXrppZf0+9//XocOHdKMGTNUUlKiyZMnS5Lmzp2r++67z9N+8uTJ+uKLL5SZmalDhw7p97//vdasWaNZs2Z52jzyyCPKzc3VkiVL9Omnn2rJkiXatm2bMjIyvO/X7VZOTo7uv/9+dezoXWQ6deqUZs2apYKCAv3973/X9u3bdccddygsLEzjxo1r4TfbfJS6AAC4gqWmpur48eNatGiRysrKFBMToy1btqhXr16SpLKyMpWUlHjaR0dHa8uWLZoxY4aef/55RUZG6rnnnvM8w0eSEhMTtX79ei1YsEALFy5U3759tWHDBs8zfOps27ZNJSUl+td//dd68woICNDHH3+sV155RSdOnFDPnj01YsQIbdiwQSEhIZfp2+A5PgAAXHZVVVVyOBwq/fSHbbKdPar/V6qsrLzkGh/UR8YHAACTtHSNTmNjoPVY4wMAAPwGGR8AAEzilqFaMj6WIvABAMAklLqsR+ADAIBJag1DtT7uKfK1v79jjQ8AAPAbZHwAADCJ+9vD1zHQegQ+AACYpLYNFjf72t/fUeoCAAB+g4wPAAAmqTUuHL6OgdYj8AEAwCSs8bEepS4AAOA3yPgAAGASt2yqlc3nMdB6BD4AAJjEbVw4fB0DrUepCwAA+A0yPgAAmKS2DUpdvvb3dwQ+AACYhMDHegQ+AACYxG3Y5DZ8XNzsY39/xxofAADgN8j4AABgEkpd1iPwAQDAJLXqoFofiy21bTQXf0WpCwAA+A0yPgAAmMRog8XNBoubfULgAwCASVjjYz1KXQAAwG+Q8QEAwCS1RgfVGj4ubuZdXT4h8AEAwCRu2eT2sdjiFpGPLwh8AAAwCWt8rMcaHwAA4DfI+AAAYJK2WeNDqcsXBD4AAJjkwhofH19SSqnLJ5S6AACA3yDjAwCASdxt8K4udnX5hsAHAACTsMbHepS6AACA3yDjAwCASdzqwAMMLUbgAwCASWoNm2p9fLu6r/39HaUuAADgNwh8AAAwSe23u7p8PVpqxYoVio6OVlBQkGJjY/XBBx802T4/P1+xsbEKCgpSnz59tGrVqnptNm7cqIEDB8put2vgwIHatGmT1+dZWVmy2Wxeh9Pp9GpjGIaysrIUGRmp4OBgDR8+XAcPHmzx/bUEgQ8AACZxGx3a5GiJDRs2KCMjQ/Pnz9e+ffs0ZMgQpaSkqKSkpMH2xcXFuu222zRkyBDt27dP8+bN0/Tp07Vx40ZPm4KCAqWmpiotLU379+9XWlqaxo8fr927d3uN9eMf/1hlZWWe4+OPP/b6fOnSpVq2bJmWL1+uPXv2yOl0auTIkTp58mSL7rElbIbBvjgAAC6nqqoqORwOvfj/YtU5JMCnsf55slbpNxWqsrJSoaGhl2wfHx+vm266SStXrvScGzBggMaOHavFixfXaz9nzhxt3rxZhw4d8pybPHmy9u/fr4KCAklSamqqqqqq9O6773raJCcnq1u3bnrjjTckXcj4vPXWWyoqKmpwXoZhKDIyUhkZGZozZ44kqbq6WhEREVqyZIkmTZp06S+jFcj4AADwPVRVVeV1VFdX12tTU1OjwsJCJSUleZ1PSkrSzp07Gxy3oKCgXvtRo0Zp7969OnfuXJNtLh7zyJEjioyMVHR0tO699159/vnnns+Ki4vlcrm8xrHb7Ro2bFijc2sLBD4AAJjErf/b2dXaw/3tWFFRUXI4HJ6joezNsWPHVFtbq4iICK/zERERcrlcDc7R5XI12P78+fM6duxYk22+O2Z8fLxeeeUVvffee3rxxRflcrmUmJio48ePe8ao69fcubUFtrMDAGCStnmOz4X+paWlXqUuu93eaB+bzXsLvGEY9c5dqv3F5y81ZkpKiuffgwYNUkJCgvr27auXX35ZmZmZrZ6brwh8AAD4HgoNDb3kGp+wsDAFBATUy6CUl5fXy7TUcTqdDbbv2LGjevTo0WSbxsaUpC5dumjQoEE6cuSIZwzpQuanZ8+ezR7HV5S6AAAwSd27unw9miswMFCxsbHKy8vzOp+Xl6fExMQG+yQkJNRrn5ubq7i4OHXq1KnJNo2NKV1YuHzo0CFPkBMdHS2n0+k1Tk1NjfLz85scx1dkfAAAMIlbNrnlWxmnpf0zMzOVlpamuLg4JSQkaPXq1SopKdHkyZMlSXPnztVXX32lV155RdKFHVzLly9XZmam0tPTVVBQoDVr1nh2a0nSI488oqFDh2rJkiW688479fbbb2vbtm3asWOHp82sWbN0xx136JprrlF5ebn+8z//U1VVVbr//vslXShxZWRkKDs7W/369VO/fv2UnZ2tzp07a8KECT59R00h8AEA4AqWmpqq48ePa9GiRSorK1NMTIy2bNmiXr16SZLKysq8nukTHR2tLVu2aMaMGXr++ecVGRmp5557TnfffbenTWJiotavX68FCxZo4cKF6tu3rzZs2KD4+HhPmy+//FK/+tWvdOzYMV111VUaPHiwdu3a5bmuJM2ePVtnzpzRlClTVFFRofj4eOXm5iokJOSyfR88xwcAgMus7jk+z+xNVPAPfMs5nDl1XjPidjb7OT7wRsYHAACTtPaVExePgdbj2wMAAH6DjA8AACZxGza5DR8XN/vY398R+AAAYBJ3G5S6fH0Aor8j8AEAwCStebt6Q2Og9fj2AACA3yDjAwCASWplU62PDzD0tb+/I/ABAMAklLqsx7cHAAD8BhkfAABMUivfS1W1bTMVv0XgAwCASSh1WY9vDwAA+A0yPgAAmKTW6KBaHzM2vvb3dwQ+AACYxJBNbh/X+BhsZ/cJYSMAAPAbZHwAADAJpS7rEfgAAGAS3s5uPQIfAABMUtsGb2f3tb+/49sDAAB+g4wPAAAmodRlPQIfAABM4lYHuX0stvja39/x7QEAAL9BxgcAAJPUGjbV+liq8rW/vyPwAQDAJKzxsR6lLgAA4DfI+AAAYBLD6CC3j09eNnhys08IfAAAMEmtbKr18SWjvvb3d1ds4ON2u/X1118rJCRENhv/RwIAaJhhGDp58qQiIyPVoQPZlCvdFRv4fP3114qKirJ6GgCA74nS0lJdffXVl/UabsP3xcluo40m46eu2MAnJCREknTN8n9Xh2C7xbPBviGvWz0FoF164uhAq6fg96pPn9PzSe95/m5cTu42WOPja39/d8UGPnXlrQ7BdnXoHGTxbBAawn9QgYbYz3Syegr4lhnLItyyye3jGh1f+/s7/hoBAAC/ccVmfAAAaG94crP1CHwAADAJa3ysx7cHAAD8BhkfAABM4lYbvKuLxc0+IfABAMAkRhvs6jIIfHxCqQsAAPgNAh8AAEziNmxtcrTUihUrFB0draCgIMXGxuqDDz5osn1+fr5iY2MVFBSkPn36aNWqVfXabNy4UQMHDpTdbtfAgQO1adMmr88XL16sm2++WSEhIQoPD9fYsWN1+PBhrzYTJ06UzWbzOgYPHtzi+2sJAh8AAExSt6vL16MlNmzYoIyMDM2fP1/79u3TkCFDlJKSopKSkgbbFxcX67bbbtOQIUO0b98+zZs3T9OnT9fGjRs9bQoKCpSamqq0tDTt379faWlpGj9+vHbv3u1pk5+fr6lTp2rXrl3Ky8vT+fPnlZSUpNOnT3tdLzk5WWVlZZ5jy5YtLbq/lrIZhnFFvvWjqqpKDodDvdcs4MnN7cCR4WutngLQLi0oH2T1FPxe9alzWnbLn1VZWanQ0NDLco26v0nj8v5FnboE+jTWudM12jQyR6WlpV7ztdvtstvrv6IpPj5eN910k1auXOk5N2DAAI0dO1aLFy+u137OnDnavHmzDh065Dk3efJk7d+/XwUFBZKk1NRUVVVV6d133/W0SU5OVrdu3fTGG280OO+jR48qPDxc+fn5Gjp0qKQLGZ8TJ07orbfeatmX4AMyPgAAmKQtS11RUVFyOByeo6EgpqamRoWFhUpKSvI6n5SUpJ07dzY4x4KCgnrtR40apb179+rcuXNNtmlsTEmqrKyUJHXv3t3r/Pbt2xUeHq5rr71W6enpKi8vb3SMtsCuLgAATNKW7+pqKONzsWPHjqm2tlYRERFe5yMiIuRyuRoc3+VyNdj+/PnzOnbsmHr27Nlom8bGNAxDmZmZ+tnPfqaYmBjP+ZSUFN1zzz3q1auXiouLtXDhQv385z9XYWFhg/fTFgh8AAAwSWsXJ188hiSFhoY2uzR38QtYDcNo8qWsDbW/+HxLxpw2bZo++ugj7dixw+t8amqq598xMTGKi4tTr1699M477+iuu+5q4o5aj8AHAIArVFhYmAICAuplYsrLy+tlbOo4nc4G23fs2FE9evRosk1DYz788MPavHmz/vrXv+rqq69ucr49e/ZUr169dOTIkUveW2uxxgcAAJOYvZ09MDBQsbGxysvL8zqfl5enxMTEBvskJCTUa5+bm6u4uDh16tSpyTbfHdMwDE2bNk1//OMf9f777ys6OvqS8z1+/LhKS0vVs2fPZt1fa5DxAQDAJG1Z6mquzMxMpaWlKS4uTgkJCVq9erVKSko0efJkSdLcuXP11Vdf6ZVXXpF0YQfX8uXLlZmZqfT0dBUUFGjNmjVeu7UeeeQRDR06VEuWLNGdd96pt99+W9u2bfMqZU2dOlWvv/663n77bYWEhHgyRA6HQ8HBwTp16pSysrJ09913q2fPnvr73/+uefPmKSwsTOPGjfPpO2oKgQ8AAFew1NRUHT9+XIsWLVJZWZliYmK0ZcsW9erVS5JUVlbm9Uyf6OhobdmyRTNmzNDzzz+vyMhIPffcc7r77rs9bRITE7V+/XotWLBACxcuVN++fbVhwwbFx8d72tRtnx8+fLjXfHJycjRx4kQFBATo448/1iuvvKITJ06oZ8+eGjFihDZs2KCQkJDL9n3wHB+Yguf4AA3jOT7WM/M5PiO3TGqT5/jk3fbCZZ3vlYyMDwAAJjHk+9vVr8hshYlY3AwAAPwGGR8AAExixeJmeCPwAQDAJAQ+1qPUBQAA/AYZHwAATELGx3oEPgAAmITAx3oEPgAAmMQwbDJ8DFx87e/vWOMDAAD8BhkfAABM4pbN5wcY+trf3xH4AABgEtb4WI9SFwAA8BtkfAAAMAmLm61H4AMAgEkodVmPUhcAAPAbZHwAADAJpS7rEfgAAGASow1KXQQ+vqHUBQAA/AYZHwAATGJIMgzfx0DrEfgAAGASt2yy8eRmSxH4AABgEhY3W481PgAAwG+Q8QEAwCRuwyYbDzC0FIEPAAAmMYw2WNzM6mafmFLqWrt2rbp27WrGpQAAABrVosBn4sSJstls9Y7PPvvscs0PAIArRt3iZl8PtF6LS13JycnKycnxOnfVVVe12YQAALhSsavLei0uddntdjmdTq/jt7/9rQYNGqQuXbooKipKU6ZM0alTpxodY//+/RoxYoRCQkIUGhqq2NhY7d271/P5zp07NXToUAUHBysqKkrTp0/X6dOnW3eHAAAA32qTNT4dOnTQc889pwMHDujll1/W+++/r9mzZzfa/te//rWuvvpq7dmzR4WFhXr00UfVqVMnSdLHH3+sUaNG6a677tJHH32kDRs2aMeOHZo2bVqTc6iurlZVVZXXAQBAe+L+9l1dvh5ovRaXuv785z/rBz/4gefnlJQUvfnmm56fo6Oj9cQTT+jf/u3ftGLFigbHKCkp0b//+7+rf//+kqR+/fp5PvvNb36jCRMmKCMjw/PZc889p2HDhmnlypUKCgpqcMzFixfr8ccfb+ntAABgGnZ1Wa/Fgc+IESO0cuVKz89dunTRX/7yF2VnZ+uTTz5RVVWVzp8/r7Nnz+r06dPq0qVLvTEyMzP14IMPat26dbr11lt1zz33qG/fvpKkwsJCffbZZ3rttdc87Q3DkNvtVnFxsQYMGNDgvObOnavMzEzPz1VVVYqKimrp7QEAgCtYi0tdXbp00Y9+9CPPUVNTo9tuu00xMTHauHGjCgsL9fzzz0uSzp071+AYWVlZOnjwoG6//Xa9//77GjhwoDZt2iRJcrvdmjRpkoqKijzH/v37deTIEU9w1BC73a7Q0FCvAwCA9uRCxsfXXV1W38X3m88PMNy7d6/Onz+vp59+Wh06XIij/vCHP1yy37XXXqtrr71WM2bM0K9+9Svl5ORo3Lhxuummm3Tw4EH96Ec/8nVqAAC0K+zqsp7Pi5v79u2r8+fP63e/+50+//xzrVu3TqtWrWq0/ZkzZzRt2jRt375dX3zxhf72t79pz549nhLWnDlzVFBQoKlTp6qoqEhHjhzR5s2b9fDDD/s6VQAALGW00YHW8znwueGGG7Rs2TItWbJEMTExeu2117R48eJG2wcEBOj48eO67777dO2112r8+PFKSUnxLEy+/vrrlZ+fryNHjmjIkCG68cYbtXDhQvXs2dPXqQIAAD9nM4wrs1pYVVUlh8Oh3msWqEPnhneCwTxHhq+1egpAu7SgfJDVU/B71afOadktf1ZlZeVlWx9a9zepzyvzFODj36Taf57V5/dlX9b5Xsl4SSkAAGZpi1rVFZmuMI8pLykFAADWWbFihaKjoxUUFKTY2Fh98MEHTbbPz89XbGysgoKC1KdPnwbX7m7cuFEDBw6U3W732p3dkusahqGsrCxFRkYqODhYw4cP18GDB3272Usg8AEAwCxt8YLSFu7q2rBhgzIyMjR//nzt27dPQ4YMUUpKikpKShpsX1xcrNtuu01DhgzRvn37NG/ePE2fPl0bN270tCkoKFBqaqrS0tK0f/9+paWlafz48dq9e3eLrrt06VItW7ZMy5cv1549e+R0OjVy5EidPHmyhV9s87HGB6ZgjQ/QMNb4WM/MNT7ROfN9/pvk/udZFf/Lk82eb3x8vG666Savhw8PGDBAY8eObXAz0pw5c7R582YdOnTIc27y5Mnav3+/CgoKJEmpqamqqqrSu+++62mTnJysbt266Y033mjWdQ3DUGRkpDIyMjRnzhxJF14/FRERoSVLlmjSpEkt/Gaah4wPAADfQxe/n7K6urpem5qaGhUWFiopKcnrfFJSknbu3NnguAUFBfXajxo1Snv37vU8mLixNnVjNue6xcXFcrlcXm3sdruGDRvW6NzaAoEPAAAm8f2pzf/3AMSoqCg5HA7P0VD25tixY6qtrVVERITX+YiICLlcrgbn6HK5Gmx//vx5HTt2rMk2dWM257p1/7slc2sL7OoCAMAsrVij0+AYkkpLS71KXXa7vdEuNpv3NQ3DqHfuUu0vPt+cMduqTVsi8AEA4HuoOe+lDAsLU0BAQL0MSnl5eb1MSx2n09lg+44dO6pHjx5NtqkbsznXdTqdki5kfr77kOKm5tYWKHUBAGCSCy8p9f1orsDAQMXGxiovL8/rfF5enhITExvsk5CQUK99bm6u4uLi1KlTpybb1I3ZnOtGR0fL6XR6tampqVF+fn6jc2sLZHwAADCLBQ8wzMzMVFpamuLi4pSQkKDVq1erpKREkydPliTNnTtXX331lV555RVJF3ZwLV++XJmZmUpPT1dBQYHWrFnj2a0lSY888oiGDh2qJUuW6M4779Tbb7+tbdu2aceOHc2+rs1mU0ZGhrKzs9WvXz/169dP2dnZ6ty5syZMmODjl9Q4Ah8AAExixdvZU1NTdfz4cS1atEhlZWWKiYnRli1b1KtXL0lSWVmZ17N1oqOjtWXLFs2YMUPPP/+8IiMj9dxzz+nuu+/2tElMTNT69eu1YMECLVy4UH379tWGDRsUHx/f7OtK0uzZs3XmzBlNmTJFFRUVio+PV25urkJCQlr79VwSz/GBKXiOD9AwnuNjPTOf43PN6v9ok+f4lDy0iHd1tRIZHwAAzHRFphu+Pwh8AAAwiRWlLnhjVxcAAPAbZHwAADCLBbu64I3ABwAA09i+PXwdA61FqQsAAPgNMj4AAJiFUpflCHwAADALgY/lKHUBAAC/QcYHAACzGLYLh69joNUIfAAAMElL367e2BhoPQIfAADMwhofy7HGBwAA+A0yPgAAmIU1PpYj8AEAwCQ248Lh6xhoPUpdAADAb5DxAQDALCxuthyBDwAAZmGNj+UodQEAAL9BxgcAALNQ6rIcgQ8AAGYh8LEcpS4AAOA3yPgAAGAWMj6WI/ABAMAs7OqyHIEPAAAm4cnN1mONDwAA8BtkfAAAMAtrfCxHxgcAAPgNAh8AAOA3rvhSV+CBzgqwB1k9Db83asINVk8B3/pqTqLVUwDaldrqs5L+bMq1bGqDxc1tMhP/dcUHPgAAtBtsZ7ccpS4AAOA3yPgAAGAWdnVZjsAHAACzEPhYjlIXAADwG2R8AAAwCa+ssB6BDwAAZqHUZTkCHwAAzELgYznW+AAAAElSRUWF0tLS5HA45HA4lJaWphMnTjTZxzAMZWVlKTIyUsHBwRo+fLgOHjzo1aa6uloPP/ywwsLC1KVLF40ZM0Zffvml5/O///3veuCBBxQdHa3g4GD17dtXjz32mGpqarzGsdls9Y5Vq1a16B4JfAAAMEndGh9fj8tlwoQJKioq0tatW7V161YVFRUpLS2tyT5Lly7VsmXLtHz5cu3Zs0dOp1MjR47UyZMnPW0yMjK0adMmrV+/Xjt27NCpU6c0evRo1dbWSpI+/fRTud1uvfDCCzp48KCeeeYZrVq1SvPmzat3vZycHJWVlXmO+++/v0X3SKkLAACztOMnNx86dEhbt27Vrl27FB8fL0l68cUXlZCQoMOHD+u6666rPxXD0LPPPqv58+frrrvukiS9/PLLioiI0Ouvv65JkyapsrJSa9as0bp163TrrbdKkl599VVFRUVp27ZtGjVqlJKTk5WcnOwZt0+fPjp8+LBWrlypp556yuuaXbt2ldPpbPV9kvEBAOB7qKqqyuuorq72abyCggI5HA5P0CNJgwcPlsPh0M6dOxvsU1xcLJfLpaSkJM85u92uYcOGefoUFhbq3LlzXm0iIyMVExPT6LiSVFlZqe7du9c7P23aNIWFhenmm2/WqlWr5Ha7W3SfBD4AAJjFaKNDUlRUlGctjsPh0OLFi32amsvlUnh4eL3z4eHhcrlcjfaRpIiICK/zERERns9cLpcCAwPVrVu3Rttc7H//93/1u9/9TpMnT/Y6/8QTT+jNN9/Utm3bdO+992rmzJnKzs5u3g1+i1IXAAAmacvn+JSWlio0NNRz3m63N9g+KytLjz/+eJNj7tmz58LYtvplNMMwGjzvNaeLPm9On8bafP3110pOTtY999yjBx980OuzBQsWeP59ww03SJIWLVrkdf5SCHwAAPgeCg0N9Qp8GjNt2jTde++9Tbbp3bu3PvroI33zzTf1Pjt69Gi9jE6durU2LpdLPXv29JwvLy/39HE6naqpqVFFRYVX1qe8vFyJiYle43399dcaMWKEEhIStHr16kve2+DBg1VVVaVvvvmm0TlejMAHAACzWPAcn7CwMIWFhV2yXUJCgiorK/Xhhx/qpz/9qSRp9+7dqqysrBeg1ImOjpbT6VReXp5uvPFGSVJNTY3y8/O1ZMkSSVJsbKw6deqkvLw8jR8/XpJUVlamAwcOaOnSpZ6xvvrqK40YMUKxsbHKyclRhw6XXo2zb98+BQUFqWvXrpdsW4fABwAAs7TFdvTLtJ19wIABSk5OVnp6ul544QVJ0kMPPaTRo0d77ejq37+/Fi9erHHjxslmsykjI0PZ2dnq16+f+vXrp+zsbHXu3FkTJkyQJDkcDj3wwAOaOXOmevTooe7du2vWrFkaNGiQZ5fX119/reHDh+uaa67RU089paNHj3quV5dV+tOf/iSXy6WEhAQFBwfrL3/5i+bPn6+HHnqo0TJfQwh8AACAJOm1117T9OnTPTuwxowZo+XLl3u1OXz4sCorKz0/z549W2fOnNGUKVNUUVGh+Ph45ebmKiQkxNPmmWeeUceOHTV+/HidOXNGv/jFL7R27VoFBARIknJzc/XZZ5/ps88+09VXX+11PcO4EOl16tRJK1asUGZmptxut/r06aNFixZp6tSpLbpHm1E34hWmqqpKDodD187IVoA9yOrp+L0fLml8yyLM9dWchlPWgL+qrT6r/3lmniorK5u1ZqY16v4m9VmQrYAg3/4m1Z49q8//8/LO90pGxgcAALPwri7LEfgAAGCSttzOjtbhAYYAAMBvEPgAAAC/QakLAACzsMbHcmR8AACA3yDjAwCASVjcbD0CHwAAzETgYilKXQAAwG+Q8QEAwCwsbrYcgQ8AACZhjY/1KHUBAAC/QcYHAACzUOqyHIEPAAAmodRlPQIfAADMQsbHcqzxAQAAfoOMDwAAZiHjYzkCHwAATMIaH+tR6gIAAH6DjA8AAGah1GU5Ah8AAMxC4GM5Sl0AAMBvkPEBAMAkLG62HoEPAABmodRlOUpdAADAb5DxAQDAJJS6rEfgAwCAWSh1Wc6nUpfNZmvymDhxYhtNEwCAK4DRRgdazaeMT1lZmeffGzZs0H/8x3/o8OHDnnPBwcFe7c+dO6dOnTr5ckkAAIBW8ynj43Q6PYfD4ZDNZvP8fPbsWXXt2lV/+MMfNHz4cAUFBenVV19VVlaWbrjhBq9xnn32WfXu3dvrXE5OjgYMGKCgoCD1799fK1as8GWqAABYztZGB1rvsq/xmTNnjp5++mnl5OTIbrdr9erVl+zz4osv6rHHHtPy5ct14403at++fUpPT1eXLl10//33N9inurpa1dXVnp+rqqra7B4AAGgTrPGx3GUPfDIyMnTXXXe1qM8TTzyhp59+2tMvOjpan3zyiV544YVGA5/Fixfr8ccf93m+AADgynXZA5+4uLgWtT969KhKS0v1wAMPKD093XP+/PnzcjgcjfabO3euMjMzPT9XVVUpKiqq5RMGAOAyYTu79S574NOlSxevnzt06CDD8P6tnTt3zvNvt9st6UK5Kz4+3qtdQEBAo9ex2+2y2+2+ThcAgMuHUpflTH+Oz1VXXSWXyyXDMGSzXViiVVRU5Pk8IiJCP/zhD/X555/r17/+tdnTAwAAVzDTA5/hw4fr6NGjWrp0qX75y19q69atevfddxUaGuppk5WVpenTpys0NFQpKSmqrq7W3r17VVFR4VXOAgDge4eMjaVMf1fXgAEDtGLFCj3//PP6yU9+og8//FCzZs3yavPggw/qpZde0tq1azVo0CANGzZMa9euVXR0tNnTBQCgzdSt8fH1QOvZjIsX3Fwhqqqq5HA4dO2MbAXYg6yejt/74ZKdVk8B3/pqTqLVUwDaldrqs/qfZ+apsrLSq/rQlur+JsU8lK2AQN/+JtXWnNWB1Zd3vlcy3s4OAIBZ2vkrKyoqKpSWliaHwyGHw6G0tDSdOHGiyT6GYSgrK0uRkZEKDg7W8OHDdfDgQa821dXVevjhhxUWFqYuXbpozJgx+vLLL73a9O7du96rrx599FGvNiUlJbrjjjvUpUsXhYWFafr06aqpqWnRPRL4AABgkvZe6powYYKKioq0detWbd26VUVFRUpLS2uyz9KlS7Vs2TItX75ce/bskdPp1MiRI3Xy5ElPm4yMDG3atEnr16/Xjh07dOrUKY0ePVq1tbVeYy1atEhlZWWeY8GCBZ7Pamtrdfvtt+v06dPasWOH1q9fr40bN2rmzJktukfezg4AgFna8Xb2Q4cOaevWrdq1a5fncTIvvviiEhISdPjwYV133XX1p2IYevbZZzV//nzPQ4dffvllRURE6PXXX9ekSZNUWVmpNWvWaN26dbr11lslSa+++qqioqK0bds2jRo1yjNeSEiInE5ng/PLzc3VJ598otLSUkVGRkqSnn76aU2cOFFPPvlks8t+ZHwAAPgeqqqq8jq++9qm1igoKJDD4fB6ht7gwYPlcDi0c2fD6zSLi4vlcrmUlJTkOWe32zVs2DBPn8LCQp07d86rTWRkpGJiYuqNu2TJEvXo0UM33HCDnnzySa8yVkFBgWJiYjxBjySNGjVK1dXVKiwsbPZ9kvEBAMAkbfnk5ovfTvDYY48pKyur1eO6XC6Fh4fXOx8eHi6Xy9VoH+nCM/i+KyIiQl988YWnTWBgoLp161avzXfHfeSRR3TTTTepW7du+vDDDzV37lwVFxfrpZde8oxz8XW6deumwMDARufXEAIfAADM0oalrtLSUq/yTmNvL8jKyrrkuyz37NkjSZ4HC3td7jsPHG7MxZ83p8/FbWbMmOH59/XXX69u3brpl7/8pScL5Mv8vovABwCA76HQ0NBmrWuZNm2a7r333ibb9O7dWx999JG++eabep8dPXq0XqalTt16HJfLpZ49e3rOl5eXe/o4nU7V1NSooqLCK+tTXl6uxMTGH68xePBgSdJnn32mHj16yOl0avfu3V5tKioqdO7cuUbn1xDW+AAAYBYLtrOHhYWpf//+TR5BQUFKSEhQZWWlPvzwQ0/f3bt3q7KystEAJTo6Wk6nU3l5eZ5zNTU1ys/P9/SJjY1Vp06dvNqUlZXpwIEDTQY++/btkyRPQJWQkKADBw6orKzM0yY3N1d2u12xsbHN/j7I+AAAYJL2/Hb2AQMGKDk5Wenp6XrhhRckSQ899JBGjx7ttaOrf//+Wrx4scaNGyebzaaMjAxlZ2erX79+6tevn7Kzs9W5c2dNmDBBkuRwOPTAAw9o5syZ6tGjh7p3765Zs2Zp0KBBnl1eBQUF2rVrl0aMGCGHw6E9e/ZoxowZGjNmjK655hpJUlJSkgYOHKi0tDT95je/0T/+8Q/NmjVL6enpLXqQI4EPAACQJL322muaPn26ZwfWmDFjtHz5cq82hw8fVmVlpefn2bNn68yZM5oyZYoqKioUHx+v3NxchYSEeNo888wz6tixo8aPH68zZ87oF7/4hdauXauAgABJF9YnbdiwQY8//riqq6vVq1cvpaena/bs2Z4xAgIC9M4772jKlCm65ZZbFBwcrAkTJuipp55q0T3yygqYgldWtB+8sgLwZuYrK35yX9u8smL/K7yyorXI+AAAYBKbYcjmY77B1/7+jsXNAADAb5DxAQDALO34lRX+gsAHAACTtOddXf6CwAcAALOQ8bEca3wAAIDfIOMDAIBJKHVZj8AHAACzUOqyHKUuAADgN8j4AABgEkpd1iPwAQDALJS6LEepCwAA+A0yPgAAmIhSlbUIfAAAMIthXDh8HQOtRuADAIBJWNxsPdb4AAAAv0HGBwAAs7Cry3IEPgAAmMTmvnD4OgZaj1IXAADwG2R8AAAwC6UuyxH4AABgEnZ1WY9SFwAA8BtkfAAAMAsPMLQcgQ8AACah1GU9Sl0AAMBvkPEBAMAs7OqyHIEPAAAmodRlPQIfAADMwuJmy7HGBwAA+A0yPgAAmIRSl/Wu2MDH+DYVWFt91uKZQJLOG+esngK+xX8mAG91/5kwzCghsbjZclds4HPy5ElJ0v+uWGTxTCBJ/2P1BPB/nnnb6hkA7dLJkyflcDisngYusys28ImMjFRpaalCQkJks9msnk6rVFVVKSoqSqWlpQoNDbV6On6N30X7wu+j/bgSfheGYejkyZOKjIy87Nei1GW9Kzbw6dChg66++mqrp9EmQkNDv7f/D+VKw++ifeH30X58338XpmV63MaFw9cx0Grs6gIAAH7jis34AADQ7rC42XIEPu2Y3W7XY489JrvdbvVU/B6/i/aF30f7we+iZWxqgzU+bTIT/2UzTNm/BwCA/6qqqpLD4dAttz6ujh2DfBrr/Pmz+tu2x1RZWfm9XldlFdb4AABglrpXVvh6XCYVFRVKS0uTw+GQw+FQWlqaTpw4cYlbMpSVlaXIyEgFBwdr+PDhOnjwoFeb6upqPfzwwwoLC1OXLl00ZswYffnll57Pt2/fLpvN1uCxZ88eT7uGPl+1alWL7pHABwAAk9RtZ/f1uFwmTJigoqIibd26VVu3blVRUZHS0tKa7LN06VItW7ZMy5cv1549e+R0OjVy5EjP8/QkKSMjQ5s2bdL69eu1Y8cOnTp1SqNHj1Ztba0kKTExUWVlZV7Hgw8+qN69eysuLs7rejk5OV7t7r///hbdI2t8AAAwSzte3Hzo0CFt3bpVu3btUnx8vCTpxRdfVEJCgg4fPqzrrruu/lQMQ88++6zmz5+vu+66S5L08ssvKyIiQq+//romTZqkyspKrVmzRuvWrdOtt94qSXr11VcVFRWlbdu2adSoUQoMDJTT6fSMe+7cOW3evFnTpk2r9yy+rl27erVtKTI+AAB8D1VVVXkd1dXVPo1XUFAgh8PhCXokafDgwXI4HNq5c2eDfYqLi+VyuZSUlOQ5Z7fbNWzYME+fwsJCnTt3zqtNZGSkYmJiGh138+bNOnbsmCZOnFjvs2nTpiksLEw333yzVq1aJbfb3aL7JPABAMAkNsNok0OSoqKiPGtxHA6HFi9e7NPcXC6XwsPD650PDw+Xy+VqtI8kRUREeJ2PiIjwfOZyuRQYGKhu3bo12uZia9as0ahRoxQVFeV1/oknntCbb76pbdu26d5779XMmTOVnZ3dvBv8FoFPO7ZixQpFR0crKChIsbGx+uCDD6yekl/661//qjvuuEORkZGy2Wx66623rJ6SX1q8eLFuvvlmhYSEKDw8XGPHjtXhw4etnpZfWrlypa6//nrP05oTEhL07rvvWj2t7wd3Gx2SSktLVVlZ6Tnmzp3b4CWzsrIaXThcd+zdu1eSGnzFk2EYl3z108WfN6dPY22+/PJLvffee3rggQfqfbZgwQIlJCTohhtu0MyZM7Vo0SL95je/afI6FyPwaac2bNigjIwMzZ8/X/v27dOQIUOUkpKikpISq6fmd06fPq2f/OQnWr58udVT8Wv5+fmaOnWqdu3apby8PJ0/f15JSUk6ffq01VPzO1dffbX+67/+S3v37tXevXv185//XHfeeWe9nTy4vOoCz7qjsWcpTZs2TYcOHWryiImJkdPp1DfffFOv/9GjR+tldOrUrbW5OHNTXl7u6eN0OlVTU6OKiopG23xXTk6OevTooTFjxlzyOxg8eLCqqqoanHdjWNzcTi1btkwPPPCAHnzwQUnSs88+q/fee08rV670OZ2JlklJSVFKSorV0/B7W7du9fo5JydH4eHhKiws1NChQy2alX+64447vH5+8skntXLlSu3atUs//vGPLZrV98N3S1W+jNESYWFhCgsLu2S7hIQEVVZW6sMPP9RPf/pTSdLu3btVWVmpxMTEBvtER0fL6XQqLy9PN954oySppqZG+fn5WrJkiSQpNjZWnTp1Ul5ensaPHy9JKisr04EDB7R06VKv8QzDUE5Oju677z516tTpknPet2+fgoKC1LVr10u2rUPg0w7V1NSosLBQjz76qNf5pKSkRheCAf6msrJSktS9e3eLZ+Lfamtr9eabb+r06dNKSEiwejrtXzve1TVgwAAlJycrPT1dL7zwgiTpoYce0ujRo712dPXv31+LFy/WuHHjZLPZlJGRoezsbPXr10/9+vVTdna2OnfurAkTJki68ALYBx54QDNnzlSPHj3UvXt3zZo1S4MGDfLs8qrz/vvvq7i4uMEy15/+9Ce5XC4lJCQoODhYf/nLXzR//nw99NBDLXpyOIFPO3Ts2DHV1tY2uVgM8GeGYSgzM1M/+9nPFBMTY/V0/NLHH3+shIQEnT17Vj/4wQ+0adMmDRw40OppwUevvfaapk+f7tmBNWbMmHpl/sOHD3v+i4ckzZ49W2fOnNGUKVNUUVGh+Ph45ebmKiQkxNPmmWeeUceOHTV+/HidOXNGv/jFL7R27VoFBAR4jb1mzRolJiZqwIAB9ebWqVMnrVixQpmZmXK73erTp48WLVqkqVOntugeeWVFO/T111/rhz/8oXbu3On136CefPJJrVu3Tp9++qmFs/NvNptNmzZt0tixY62eil+bOnWq3nnnHe3YsUNXX3211dPxSzU1NSopKdGJEye0ceNGvfTSS8rPzyf4aUTdKyuG3rKwTV5Z8de/PcErK1qJjE87FBYWpoCAgCYXiwH+6uGHH9bmzZv117/+laDHQoGBgfrRj34kSYqLi9OePXv029/+1lMiQcPa4snLl/PJzf6AXV3tUGBgoGJjY5WXl+d1Pi8vr9EFZsCVzjAMTZs2TX/84x/1/vvvKzo62uop4TsMw/D5AXqAGcj4tFOZmZlKS0tTXFycEhIStHr1apWUlGjy5MlWT83vnDp1Sp999pnn5+LiYhUVFal79+665pprLJyZf5k6dapef/11vf322woJCfFkRB0Oh4KDgy2enX+ZN2+eUlJSFBUVpZMnT2r9+vXavn17vZ13aEBbvGSUFSo+IfBpp1JTU3X8+HEtWrRIZWVliomJ0ZYtW9SrVy+rp+Z39u7dqxEjRnh+zszMlCTdf//9Wrt2rUWz8j8rV66UJA0fPtzrfE5OToOPtcfl88033ygtLU1lZWVyOBy6/vrrtXXrVo0cOdLqqbV7NveFw9cx0HosbgYA4DKrW9w8/Kfz22Rx8/YPn2RxcyuxxgcAAPgNSl0AAJilHT/A0F8Q+AAAYBIrXlkBb5S6AACA3yDjAwCAWdjObjkCHwAAzGJI8nU7OnGPTyh1AQAAv0HGBwAAk7C42XoEPgAAmMVQG6zxaZOZ+C1KXQAAwG+Q8QEAwCzs6rIcgQ8AAGZxS7K1wRhoNQIfAABMwuJm67HGBwAA+A0yPgAAmIU1PpYj8AEAwCwEPpaj1AUAAPwGGR8AAMxCxsdyBD4AAJiF7eyWo9QFAAD8BhkfAABMwnN8rEfgAwCAWVjjYzlKXQAAwG+Q8QEAwCxuQ7L5mLFxk/HxBYEPAABmodRlOQIfAABM0waBjwh8fMEaHwAA4DfI+AAAYBZKXZYj8AEAwCxuQz6Xqljc7BNKXQAAwG+Q8QEAwCyG+8Lh6xhoNQIfAADMwhofy1HqAgAAfoPABwAAs7iNtjkuk4qKCqWlpcnhcMjhcCgtLU0nTpxoso9hGMrKylJkZKSCg4M1fPhwHTx40KvN6tWrNXz4cIWGhspmszU4ZnOuXVJSojvuuENdunRRWFiYpk+frpqamhbdI4EPAABmqSt1+XpcJhMmTFBRUZG2bt2qrVu3qqioSGlpaU32Wbp0qZYtW6bly5drz549cjqdGjlypE6ePOlp889//lPJycmaN29eq69dW1ur22+/XadPn9aOHTu0fv16bdy4UTNnzmzRPdoMg2IhAACXU1VVlRwOh26NnKSOHew+jXXeXa1tX7+g0tJShYaGes7b7XbZ7a0f+9ChQxo4cKB27dql+Ph4SdKuXbuUkJCgTz/9VNddd129PoZhKDIyUhkZGZozZ44kqbq6WhEREVqyZIkmTZrk1X779u0aMWKEKioq1LVr1xZd+91339Xo0aNVWlqqyMhISdL69es1ceJElZeXe30XTSHjAwCAWQy1QcbnwlBRUVGespDD4dDixYt9mlpBQYEcDocn8JCkwYMHy+FwaOfOnQ32KS4ulsvlUlJSkuec3W7XsGHDGu3T2msXFBQoJibGE/RI0qhRo1RdXa3CwsJmX4tdXQAAmKUNd3U1lPHxhcvlUnh4eL3z4eHhcrlcjfaRpIiICK/zERER+uKLL9r02i6Xq951unXrpsDAwEbn1xAyPgAAmMXtbptDUmhoqNfRWOCTlZUlm83W5LF3715Jks1mq9ffMIwGz3/XxZ83p8+lxmhonNbO77vI+AAAcAWbNm2a7r333ibb9O7dWx999JG++eabep8dPXq0XqaljtPplHQhG9OzZ0/P+fLy8kb7NDbOpa7tdDq1e/dur88rKip07ty5Fl2LjA8AAGaxYFdXWFiY+vfv3+QRFBSkhIQEVVZW6sMPP/T03b17tyorK5WYmNjg2NHR0XI6ncrLy/Ocq6mpUX5+fqN9GtKcayckJOjAgQMqKyvztMnNzZXdbldsbGyzr0XgAwCAWdrxdvYBAwYoOTlZ6enp2rVrl3bt2qX09HSNHj3aa0dX//79tWnTJkkXSk8ZGRnKzs7Wpk2bdODAAU2cOFGdO3fWhAkTPH1cLpeKior02WefSZI+/vhjFRUV6R//+Eezr52UlKSBAwcqLS1N+/bt03//939r1qxZSk9Pb/aOLonABwAAfOu1117ToEGDlJSUpKSkJF1//fVat26dV5vDhw+rsrLS8/Ps2bOVkZGhKVOmKC4uTl999ZVyc3MVEhLiabNq1SrdeOONSk9PlyQNHTpUN954ozZv3tzsawcEBOidd95RUFCQbrnlFo0fP15jx47VU0891aJ75Dk+AABcZp7n+HT/F3XsEOjTWOfdNdr2jxxVVla2KNOBC1jcDACASQzDLcPHt6v72t/fUeoCAAB+g4wPAABmMdrgJaOsUPEJgQ8AAGYxDHneOeHTGGgtSl0AAMBvkPEBAMAsbrdk83FxMoubfULgAwCAWSh1WY7ABwAAkxhutwwfMz5sZ/cNa3wAAIDfIOMDAIBZKHVZjsAHAACzuA3JRuBjJUpdAADAb5DxAQDALIYhydft7GR8fEHgAwCASQy3IcPHUpdB4OMTSl0AAMBvkPEBAMAshlu+l7p4jo8vCHwAADAJpS7rUeoCAAB+g4wPAAAmOW9U+1yqOq9zbTQb/0TgAwDAZRYYGCin06kdri1tMp7T6VRgYGCbjOVvbAbFQgAALruzZ8+qpqamTcYKDAxUUFBQm4zlbwh8AACA32BxMwAA8BsEPgAAwG8Q+AAAAL9B4AMAAPwGgQ8AAPAbBD4AAMBvEPgAAAC/8f8B3AA0wn/ru+wAAAAASUVORK5CYII=\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(metrics[:, :, 0, 1].T- compute_auc(data_dict, datasets=[\"mask-r\"], stage='para_merge', gt_stage='para_gt_merge' ))\n",
    "plt.yticks(do_3D_vals, do_3D_vals)\n",
    "plt.xticks(radii)\n",
    "plt.colorbar()\n",
    "\n",
    "print(metrics[:, :, 0, 1].max())\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    " --> different optimal settings depending whether the GT is clipped to the RBC segment. Since these might be faulty, better stick to the values for the orignal GT."
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [
    "# mask-ts"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 296,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.6561264991760254\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 2 Axes>",
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAj4AAAGFCAYAAAD91/mHAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAA9hAAAPYQGoP6dpAAA7D0lEQVR4nO3df3RU1b3//9fwI5OAyQikyZBrhKjIjxWtEGwIrRBaCaEiRdoCN12prIUBCxFD4IICXgLWpCCitZFf6gWrUKhfSotXzU24SColQGAREaS5UmOJkgG0YQYREsic7x+Y+TjmB0lmciYyz8daZ8mc2XuffZK1zHu93/vsYzEMwxAAAEAQ6BToCQAAAJiFwAcAAAQNAh8AABA0CHwAAEDQIPABAABBg8AHAAAEDQIfAAAQNLoEegIAAASDS5cuqba21i9jhYSEKDQ01C9jBRsCHwAA2tmlS5cU1+cGOc7U+WU8u92uiooKgp82IPABAKCd1dbWynGmTv881FcR4b6tMnGdd6tPwseqra0l8GkDAh8AAExyQ7hFN4RbfBrDLd/6BzsCHwAATFJnuFXn4xsy6wy3fyYTpHiqCwAABA0yPgAAmMQtQ275lvLxtX+wI/ABAMAkbrnla6HK9xGCG6UuAAAQNMj4AABgkjrDUJ3hW6nK1/7BjsAHAACTsMYn8Ch1AQCAoEHGBwAAk7hlqI6MT0AR+AAAYBJKXYFH4AMAgElY3Bx4rPEBAABBg4wPAAAmcX91+DoG2o7ABwAAk9T5YXGzr/2DHaUuAAAQNMj4AABgkjrj6uHrGGg7Ah8AAEzCGp/Ao9QFAACCBhkfAABM4pZFdbL4PAbajsAHAACTuI2rh69joO0odQEAgKBB4AMAgEnqvip1+Xq01urVqxUXF6fQ0FAlJCTo3XffbbZ9cXGxEhISFBoaqltuuUVr165t0Gbbtm0aNGiQrFarBg0apO3bt3t9n5eXp7vvvlvh4eGKiorShAkTVF5e7tVm6tSpslgsXsewYcNafX+tQeADAIBJAhH4bN26VVlZWVq0aJEOHz6se+65R2PHjtXJkycbbV9RUaEf//jHuueee3T48GEtXLhQs2fP1rZt2zxtSkpKNHnyZKWnp+u9995Tenq6Jk2apP3793vaFBcXa9asWdq3b5+Kiop05coVpaSk6MKFC17XS01NVVVVled46623WnV/rWUxDN52BgBAe3K5XLLZbNpzNEY3hPuWc/jivFs/iD8lp9OpiIiIa7ZPTEzUkCFDtGbNGs+5gQMHasKECcrLy2vQfsGCBdqxY4eOHz/uOffwww/rvffeU0lJiSRp8uTJcrlcevvttz1tUlNT1aNHD/3hD39odB5nz55VVFSUiouLNWLECElXMz7nzp3Tn//85xbduz+Q8QEA4FvI5XJ5HTU1NQ3a1NbW6tChQ0pJSfE6n5KSor179zY6bklJSYP2Y8aM0cGDB3X58uVm2zQ1piQ5nU5JUs+ePb3O7969W1FRUbr99tuVkZGhM2fONDmGPxD4AABgEn+WumJjY2Wz2TxHY9mbzz77THV1dYqOjvY6Hx0dLYfD0egcHQ5Ho+2vXLmizz77rNk2TY1pGIays7P1gx/8QPHx8Z7zY8eO1aZNm7Rr1y4988wzKi0t1Q9/+MNGgzh/4XF2AABMUqdOqvMx51D31X8rKyu9Sl1Wq7XJPhaL97ogwzAanLtW+2+eb82YmZmZOnLkiPbs2eN1fvLkyZ5/x8fHa+jQoerTp4/efPNNTZw4scn5+YLABwCAb6GIiIhrrvGJjIxU586dG2Rizpw50yBjU89utzfavkuXLurVq1ezbRob85FHHtGOHTv017/+VTfddFOz8+3du7f69OmjDz/8sNl2vqDUBQCASQzDIrePh2G0/KmukJAQJSQkqKioyOt8UVGRhg8f3mifpKSkBu0LCws1dOhQde3atdk2Xx/TMAxlZmbqT3/6k3bt2qW4uLhrzvfzzz9XZWWlevfu3aL7awsyPgAAmKSt+/B8c4zWyM7OVnp6uoYOHaqkpCStX79eJ0+e1MMPPyxJevzxx/Xpp5/q97//vaSrT3Dl5+crOztbGRkZKikp0csvv+z1tNajjz6qESNGaPny5frJT36iv/zlL9q5c6dXKWvWrFnavHmz/vKXvyg8PNyTIbLZbAoLC9MXX3yhnJwc/fSnP1Xv3r318ccfa+HChYqMjNQDDzzg08+oOQQ+AABcxyZPnqzPP/9cy5YtU1VVleLj4/XWW2+pT58+kqSqqiqvPX3i4uL01ltvac6cOXrhhRcUExOj559/Xj/96U89bYYPH64tW7Zo8eLFeuKJJ3Trrbdq69atSkxM9LSpf3w+OTnZaz4bNmzQ1KlT1blzZ73//vv6/e9/r3Pnzql3794aNWqUtm7dqvDw8Hb7ebCPDwAA7ax+H5+3j8Spu4/7+Fw479bYOytavI8PvJHxAQDAJG5Z5PZxea1b5Ct8QeADAIBJArHGB954qgsAAAQNMj4AAJikzuikOsPHDQxZmusTAh8AAExydY2Pb6UqX/sHO0pdAAAgaJDxAQDAJG4/vKuLp7p8Q+ADAIBJWOMTeJS6AABA0CDjAwCASdzqxAaGAUbgAwCASeoMi+pa8Xb1psZA21HqAgAAQYOMDwAAJqnzw1NddZS6fELgAwCASdxGJ7l9fKrLzVNdPiHwAQDAJGR8Ao81PgAAIGiQ8QEAwCRu+f5Ults/UwlaBD4AAJjEP/v4UKzxBT89AAAQNMj4AABgEv+8q4uchS8IfAAAMIlbFrnl6xofdm72BWEjAAAIGmR8AAAwCaWuwCPwAQDAJP7ZwJDAxxf89AAAQNAg4wMAgEnchkVuXzcw9LF/sCPwAQDAJG4/lLrYwNA3BD4AAJjEP29nJ/DxBT89AAAQNMj4AABgkjpZVOfjBoS+9g92BD4AAJiEUlfg8dMDAABBg4wPAAAmqZPvpao6/0wlaBH4AABgEkpdgcdPDwAABA0CHwAATFL/klJfj9ZavXq14uLiFBoaqoSEBL377rvNti8uLlZCQoJCQ0N1yy23aO3atQ3abNu2TYMGDZLVatWgQYO0ffv2Vl/XMAzl5OQoJiZGYWFhSk5O1rFjx1p9f61B4AMAgEkMWeT28TBauUZo69atysrK0qJFi3T48GHdc889Gjt2rE6ePNlo+4qKCv34xz/WPffco8OHD2vhwoWaPXu2tm3b5mlTUlKiyZMnKz09Xe+9957S09M1adIk7d+/v1XXXbFihVatWqX8/HyVlpbKbrdr9OjROn/+fCt/si1nMQzDaLfRAQCAXC6XbDabHisZK+sNXX0aq+aLy/pN0ttyOp2KiIi4ZvvExEQNGTJEa9as8ZwbOHCgJkyYoLy8vAbtFyxYoB07duj48eOecw8//LDee+89lZSUSJImT54sl8ult99+29MmNTVVPXr00B/+8IcWXdcwDMXExCgrK0sLFiy4em81NYqOjtby5cs1Y8aMVv5kWoaMDwAAJvFnqcvlcnkdNTU1Da5XW1urQ4cOKSUlxet8SkqK9u7d2+gcS0pKGrQfM2aMDh48qMuXLzfbpn7Mlly3oqJCDofDq43VatXIkSObnJs/EPgAAGCS+rez+3pIUmxsrGw2m+doLHvz2Wefqa6uTtHR0V7no6Oj5XA4Gp2jw+FotP2VK1f02WefNdumfsyWXLf+v62Zmz/wODsAACap88Pb2ev7V1ZWepW6rFZrk30sFu91QYZhNDh3rfbfPN+SMf3Vxp8IfAAA+BaKiIi45hqfyMhIde7cuUEG5cyZMw0yLfXsdnuj7bt06aJevXo126Z+zJZc1263S7qa+endu3eL5uYPlLoAADCJP0tdLRESEqKEhAQVFRV5nS8qKtLw4cMb7ZOUlNSgfWFhoYYOHaquXbs226Z+zJZcNy4uTna73atNbW2tiouLm5ybP5DxAQDAJG51ktvHnENr+2dnZys9PV1Dhw5VUlKS1q9fr5MnT+rhhx+WJD3++OP69NNP9fvf/17S1Se48vPzlZ2drYyMDJWUlOjll1/2PK0lSY8++qhGjBih5cuX6yc/+Yn+8pe/aOfOndqzZ0+Lr2uxWJSVlaXc3Fz169dP/fr1U25urrp166a0tDSffkbNIfABAOA6NnnyZH3++edatmyZqqqqFB8fr7feekt9+vSRJFVVVXntrRMXF6e33npLc+bM0QsvvKCYmBg9//zz+ulPf+ppM3z4cG3ZskWLFy/WE088oVtvvVVbt25VYmJii68rSfPnz9fFixc1c+ZMVVdXKzExUYWFhQoPD2+3nwf7+AAA0M7q9/H51bsT/bKPz5p7/tTifXzgjYwPAAAmae0anabGQNuxuBkAAAQNMj4AAJjEMDrJ3YaXjH5zDLQdgQ8AACapk0V1rXzJaGNjoO2u28DH7Xbr1KlTCg8Pb9cdIAEA326GYej8+fOKiYlRp05kU653123gc+rUKcXGxgZ6GgCAb4nKykrddNNN7XoNt+H74mQ3z2L75LoNfOr3AJhVOEbW7r49OghcT574zgeBngLQobi+cKvPkI/bde+Yem4/rPHxtX+wu24Dn/rylrV7V5/3TACuJxHh/E8TaIwZyyLcssjt4xodX/sHO/4PCAAAgsZ1m/EBAKCjqTMsqvNxjY+v/YMdgQ8AACZhjU/g8dMDAABBg4wPAAAmccsP7+picbNPCHwAADCJ4YenugwCH59Q6gIAAEGDjA8AACZxG34odfFUl08IfAAAMAlPdQUePz0AABA0yPgAAGASSl2BR+ADAIBJeFdX4BH4AABgEjI+gccaHwAAEDTI+AAAYBIyPoFH4AMAgEkIfAKPUhcAAAgaZHwAADAJGZ/AI/ABAMAkhnx/HN3wz1SCFqUuAAAQNMj4AABgEkpdgUfgAwCASQh8Ao9SFwAACBpkfAAAMAkZn8Aj8AEAwCQEPoFH4AMAgEkMwyLDx8DF1/7BjjU+AAAgaJDxAQDAJG5ZfN7A0Nf+wY6MDwAAJqlf4+Pr0V6qq6uVnp4um80mm82m9PR0nTt3rtk+hmEoJydHMTExCgsLU3Jyso4dO+bVpqamRo888ogiIyPVvXt3jR8/Xp988onn+48//ljTpk1TXFycwsLCdOutt2rJkiWqra31GsdisTQ41q5d26p7JPABAACSpLS0NJWVlamgoEAFBQUqKytTenp6s31WrFihVatWKT8/X6WlpbLb7Ro9erTOnz/vaZOVlaXt27dry5Yt2rNnj7744guNGzdOdXV1kqS///3vcrvdWrdunY4dO6Znn31Wa9eu1cKFCxtcb8OGDaqqqvIcDz74YKvukVIXAAAm6ciLm48fP66CggLt27dPiYmJkqQXX3xRSUlJKi8vV//+/RuZi6HnnntOixYt0sSJEyVJr7zyiqKjo7V582bNmDFDTqdTL7/8sl599VXde++9kqTXXntNsbGx2rlzp8aMGaPU1FSlpqZ6xr3llltUXl6uNWvWaOXKlV7XvPHGG2W329t8n2R8AAAwiT9LXS6Xy+uoqanxaW4lJSWy2WyeoEeShg0bJpvNpr179zbap6KiQg6HQykpKZ5zVqtVI0eO9PQ5dOiQLl++7NUmJiZG8fHxTY4rSU6nUz179mxwPjMzU5GRkbr77ru1du1aud3uVt0ngQ8AAN9CsbGxnrU4NptNeXl5Po3ncDgUFRXV4HxUVJQcDkeTfSQpOjra63x0dLTnO4fDoZCQEPXo0aPJNt/0j3/8Q7/73e/08MMPe51/8skn9frrr2vnzp2aMmWK5s6dq9zc3Jbd4FcodQEAYBJ/lroqKysVERHhOW+1Whttn5OTo6VLlzY7ZmlpqaSri4cbXs9o9PzXffP7lvRpqs2pU6eUmpqqn//853rooYe8vlu8eLHn33fddZckadmyZV7nr4XABwAAkxh+eCqrPvCJiIjwCnyakpmZqSlTpjTbpm/fvjpy5IhOnz7d4LuzZ882yOjUq19r43A41Lt3b8/5M2fOePrY7XbV1taqurraK+tz5swZDR8+3Gu8U6dOadSoUUpKStL69euveW/Dhg2Ty+XS6dOnm5zjN1HqAgDgOhYZGakBAwY0e4SGhiopKUlOp1MHDhzw9N2/f7+cTmeDAKVeXFyc7Ha7ioqKPOdqa2tVXFzs6ZOQkKCuXbt6tamqqtLRo0e9xv3000+VnJysIUOGaMOGDerU6dohyuHDhxUaGqobb7yxxT8PMj4AAJjEkGQYvo/RHgYOHKjU1FRlZGRo3bp1kqTp06dr3LhxXk90DRgwQHl5eXrggQdksViUlZWl3Nxc9evXT/369VNubq66deumtLQ0SZLNZtO0adM0d+5c9erVSz179tS8efN0xx13eJ7yOnXqlJKTk3XzzTdr5cqVOnv2rOd69VmlN954Qw6HQ0lJSQoLC9M777yjRYsWafr06U2W+RpD4AMAgEncssjSgXdu3rRpk2bPnu15Amv8+PHKz8/3alNeXi6n0+n5PH/+fF28eFEzZ85UdXW1EhMTVVhYqPDwcE+bZ599Vl26dNGkSZN08eJF/ehHP9LGjRvVuXNnSVJhYaFOnDihEydO6KabbvK6nvFVpNi1a1etXr1a2dnZcrvduuWWW7Rs2TLNmjWrVfdoMQxfY8+OyeVyyWazKftv42S9oWugpwN0GL+Oej/QUwA6FNd5t3rc/pGcTmeL1sy06Rpf/U268/V56tyt5dmJxtR9WaMjP1/ZrvO9nrHGBwAABA1KXQAAmMRtWGTx8amu9nxXVzAg8AEAwCSG4YfFzdflAhXzmFLq2rhxY6seNQMAAGgPrQp8pk6d2ugr4U+cONFe8wMA4LpRv3OzrwfartWlrtTUVG3YsMHr3He+8x2/TQgAgOtVR347e7BodanLarXKbrd7Hb/97W91xx13qHv37oqNjdXMmTP1xRdfNDnGe++9p1GjRik8PFwRERFKSEjQwYMHPd/v3btXI0aMUFhYmGJjYzV79mxduHChbXcIAADwFb+s8enUqZOef/55HT16VK+88op27dql+fPnN9n+F7/4hW666SaVlpbq0KFDeuyxx9S169W9dt5//32NGTNGEydO1JEjR7R161bt2bNHmZmZzc6hpqZGLpfL6wAAoCNxf/WuLl8PtF2rS13//d//rRtuuMHzeezYsXr99dc9n+Pi4vTkk0/qV7/6lVavXt3oGCdPntR//Md/aMCAAZKkfv36eb57+umnlZaWpqysLM93zz//vEaOHKk1a9YoNDS00THz8vKu+fZZAAACiae6Aq/Vgc+oUaO0Zs0az+fu3bvrnXfeUW5urj744AO5XC5duXJFly5d0oULF9S9e/cGY2RnZ+uhhx7Sq6++qnvvvVc///nPdeutt0qSDh06pBMnTmjTpk2e9oZhyO12q6KiQgMHDmx0Xo8//riys7M9n10ul2JjY1t7ewAA4DrW6lJX9+7dddttt3mO2tpa/fjHP1Z8fLy2bdumQ4cO6YUXXpAkXb58udExcnJydOzYMd13333atWuXBg0apO3bt0uS3G63ZsyYobKyMs/x3nvv6cMPP/QER42xWq2KiIjwOgAA6EiuZnx8faor0Hfx7ebzBoYHDx7UlStX9Mwzz3heIf/HP/7xmv1uv/123X777ZozZ47+/d//XRs2bNADDzygIUOG6NixY7rtttt8nRoAAB0KT3UFns+Lm2+99VZduXJFv/vd7/TRRx/p1Vdf1dq1a5tsf/HiRWVmZmr37t365z//qb/97W8qLS31lLAWLFigkpISzZo1S2VlZfrwww+1Y8cOPfLII75OFQCAgDL8dKDtfA587rrrLq1atUrLly9XfHy8Nm3apLy8vCbbd+7cWZ9//rl++ctf6vbbb9ekSZM0duxYz8LkO++8U8XFxfrwww91zz33aPDgwXriiSfUu3dvX6cKAACCnMUwrs9qocvlks1mU/bfxsl6Q9dATwfoMH4d9X6gpwB0KK7zbvW4/SM5nc52Wx9a/zfplt8vVOdujT+d3FJ1X17SR7/Mbdf5Xs94SSkAAGbxR63qukxXmMeUl5QCAAB0BGR8AAAwiz9eMspTXT4h8AEAwCTs3Bx4lLoAAEDQIOMDAIBJ2MAw8Ah8AAAwi2HxfY0OgY9PKHUBAICgQcYHAACTsLg58Ah8AAAwCxsYBhyBDwAAJmFxc+CxxgcAAAQNMj4AAJiJUlVAEfgAAGASSl2BR6kLAAAEDTI+AACYhae6Ao7ABwAA01i+OnwdA21FqQsAAAQNMj4AAJiFUlfAEfgAAGAWAp+Ao9QFAACCBoEPAABmMSz+OdpJdXW10tPTZbPZZLPZlJ6ernPnzjV/S4ahnJwcxcTEKCwsTMnJyTp27JhXm5qaGj3yyCOKjIxU9+7dNX78eH3yySdebfr27SuLxeJ1PPbYY15tTp48qfvvv1/du3dXZGSkZs+erdra2lbdI4EPAAAmqX87u69He0lLS1NZWZkKCgpUUFCgsrIypaenN9tnxYoVWrVqlfLz81VaWiq73a7Ro0fr/PnznjZZWVnavn27tmzZoj179uiLL77QuHHjVFdX5zXWsmXLVFVV5TkWL17s+a6urk733XefLly4oD179mjLli3atm2b5s6d26p7ZI0PAABm6cBrfI4fP66CggLt27dPiYmJkqQXX3xRSUlJKi8vV//+/RtOxTD03HPPadGiRZo4caIk6ZVXXlF0dLQ2b96sGTNmyOl06uWXX9arr76qe++9V5L02muvKTY2Vjt37tSYMWM844WHh8tutzc6v8LCQn3wwQeqrKxUTEyMJOmZZ57R1KlT9dRTTykiIqJF90nGBwCAbyGXy+V11NTU+DReSUmJbDabJ+iRpGHDhslms2nv3r2N9qmoqJDD4VBKSornnNVq1ciRIz19Dh06pMuXL3u1iYmJUXx8fINxly9frl69eumuu+7SU0895VXGKikpUXx8vCfokaQxY8aopqZGhw4davF9kvEBAMAs/lij81X/2NhYr9NLlixRTk5Om4d1OByKiopqcD4qKkoOh6PJPpIUHR3tdT46Olr//Oc/PW1CQkLUo0ePBm2+Pu6jjz6qIUOGqEePHjpw4IAef/xxVVRU6KWXXvKM883r9OjRQyEhIU3OrzEEPgAAmMRiXD18HUOSKisrvco7Vqu10fY5OTlaunRps2OWlpZeHdvSMCgzDKPR815z+sb3LenzzTZz5szx/PvOO+9Ujx499LOf/cyTBfJlfl9H4AMAwLdQREREi9a1ZGZmasqUKc226du3r44cOaLTp083+O7s2bMNMi316tfjOBwO9e7d23P+zJkznj52u121tbWqrq72yvqcOXNGw4cPb3JOw4YNkySdOHFCvXr1kt1u1/79+73aVFdX6/Lly03OrzGs8QEAwCyGn45WiIyM1IABA5o9QkNDlZSUJKfTqQMHDnj67t+/X06ns8kAJS4uTna7XUVFRZ5ztbW1Ki4u9vRJSEhQ165dvdpUVVXp6NGjzQY+hw8fliRPQJWUlKSjR4+qqqrK06awsFBWq1UJCQkt/nmQ8QEAwCx+XOPjbwMHDlRqaqoyMjK0bt06SdL06dM1btw4rye6BgwYoLy8PD3wwAOyWCzKyspSbm6u+vXrp379+ik3N1fdunVTWlqaJMlms2natGmaO3euevXqpZ49e2revHm64447PE95lZSUaN++fRo1apRsNptKS0s1Z84cjR8/XjfffLMkKSUlRYMGDVJ6erqefvpp/etf/9K8efOUkZHR4ie6JAIfAADwlU2bNmn27NmeJ7DGjx+v/Px8rzbl5eVyOp2ez/Pnz9fFixc1c+ZMVVdXKzExUYWFhQoPD/e0efbZZ9WlSxdNmjRJFy9e1I9+9CNt3LhRnTt3lnR1fdLWrVu1dOlS1dTUqE+fPsrIyND8+fM9Y3Tu3FlvvvmmZs6cqe9///sKCwtTWlqaVq5c2ap7tBhGe26FFDgul0s2m03Zfxsn6w1dAz0doMP4ddT7gZ4C0KG4zrvV4/aP5HQ6W5U5aNU1vvqbFLvqSXUKC/VpLPfFS6rMfqJd53s9I+MDAIBZOvAGhsGCxc0AACBokPEBAMAsZHwCjsAHAACzdOCnuoIFgQ8AACbx587NaBvW+AAAgKBBxgcAALOwxifgyPgAAICgQeADAACCxnVf6jqwMEFduvi2SyZwPRmzq3Ogp4Cv+XRB0y9phDnqai5JWmjKtSzyw+Jmv8wkeF33gQ8AAB0Gj7MHHKUuAAAQNMj4AABgFp7qCjgCHwAAzELgE3CUugAAQNAg4wMAgEl4ZUXgEfgAAGAWSl0BR+ADAIBZCHwCjjU+AAAgaJDxAQDAJKzxCTwCHwAAzMLOzQFHqQsAAAQNMj4AAJiFxc0BR+ADAIBJWOMTeJS6AABA0CDjAwCAWSh1BRyBDwAAZvFDqYvAxzeUugAAQNAg4wMAgFkodQUcgQ8AAGYh8Ak4Ah8AAEzC4+yBxxofAAAQNAh8AABA0KDUBQCAWVjjE3BkfAAAQNAg4wMAgElY3Bx4ZHwAADCT4ePRjqqrq5Weni6bzSabzab09HSdO3eu2T6GYSgnJ0cxMTEKCwtTcnKyjh075tWmpqZGjzzyiCIjI9W9e3eNHz9en3zyief73bt3y2KxNHqUlpZ62jX2/dq1a1t1jwQ+AABAkpSWlqaysjIVFBSooKBAZWVlSk9Pb7bPihUrtGrVKuXn56u0tFR2u12jR4/W+fPnPW2ysrK0fft2bdmyRXv27NEXX3yhcePGqa6uTpI0fPhwVVVVeR0PPfSQ+vbtq6FDh3pdb8OGDV7tHnzwwVbdI6UuAADM0oEXNx8/flwFBQXat2+fEhMTJUkvvviikpKSVF5erv79+zecimHoueee06JFizRx4kRJ0iuvvKLo6Ght3rxZM2bMkNPp1Msvv6xXX31V9957ryTptddeU2xsrHbu3KkxY8YoJCREdrvdM+7ly5e1Y8cOZWZmymKxeF3zxhtv9GrbWmR8AAAwSf0aH18PSXK5XF5HTU2NT3MrKSmRzWbzBD2SNGzYMNlsNu3du7fRPhUVFXI4HEpJSfGcs1qtGjlypKfPoUOHdPnyZa82MTExio+Pb3LcHTt26LPPPtPUqVMbfJeZmanIyEjdfffdWrt2rdxud6vuk8AHAIBvodjYWM9aHJvNpry8PJ/GczgcioqKanA+KipKDoejyT6SFB0d7XU+Ojra853D4VBISIh69OjRZJtvevnllzVmzBjFxsZ6nX/yySf1+uuva+fOnZoyZYrmzp2r3Nzclt3gVyh1AQBgFj+WuiorKxUREeE5bbVaG22ek5OjpUuXNjtk/QLib5aVpKvlrMbOf903v29Jn6bafPLJJ/qf//kf/fGPf2zw3eLFiz3/vuuuuyRJy5Yt8zp/LQQ+AACYxJ+Ps0dERHgFPk3JzMzUlClTmm3Tt29fHTlyRKdPn27w3dmzZxtkdOrVr7VxOBzq3bu35/yZM2c8fex2u2pra1VdXe2V9Tlz5oyGDx/eYMwNGzaoV69eGj9+/DXvbdiwYXK5XDp9+nSTc/wmAh8AAMwSgMXNkZGRioyMvGa7pKQkOZ1OHThwQN/73vckSfv375fT6Ww0QJGkuLg42e12FRUVafDgwZKk2tpaFRcXa/ny5ZKkhIQEde3aVUVFRZo0aZIkqaqqSkePHtWKFSu8b80wtGHDBv3yl79U165drznnw4cPKzQ0VDfeeOM129Yj8AEAABo4cKBSU1OVkZGhdevWSZKmT5+ucePGeT3RNWDAAOXl5emBBx6QxWJRVlaWcnNz1a9fP/Xr10+5ubnq1q2b0tLSJEk2m03Tpk3T3Llz1atXL/Xs2VPz5s3THXfc4XnKq96uXbtUUVGhadOmNZjfG2+8IYfDoaSkJIWFhemdd97RokWLNH369CbLfI0h8AEAwCwd+HF2Sdq0aZNmz57teQJr/Pjxys/P92pTXl4up9Pp+Tx//nxdvHhRM2fOVHV1tRITE1VYWKjw8HBPm2effVZdunTRpEmTdPHiRf3oRz/Sxo0b1blzZ6+xX375ZQ0fPlwDBw5sMLeuXbtq9erVys7Oltvt1i233KJly5Zp1qxZrbpHi2EY1+Xm1y6XSzabTT8YuURduoQGejpAh9Fl16FATwFf8+mCxksIME9dzSX937ML5XQ6W7Rmpi3q/yb1n5Orzlbf/ibV1VxSeTvP93rG4+wAACBoUOoCAMAsHbzUFQwIfAAAMAuBT8BR6gIAAEGDjA8AACbx5waGaBsCHwAAzEKpK+AodQEAgKBBxgcAAJNQ6go8Ah8AAMxCqSvgfCp1WSyWZo+pU6f6aZoAAFwHDD8daDOfMj5VVVWef2/dulX/+Z//qfLycs+5sLAwr/aXL19u0dtWAQAA2oNPGR+73e45bDabLBaL5/OlS5d044036o9//KOSk5MVGhqq1157TTk5Obrrrru8xnnuuefUt29fr3MbNmzQwIEDFRoaqgEDBmj16tW+TBUAgICz+OlA27X7Gp8FCxbomWee0YYNG2S1WrV+/fpr9nnxxRe1ZMkS5efna/DgwTp8+LAyMjLUvXt3Pfjgg432qampUU1Njeezy+Xy2z0AAOAXrPEJuHYPfLKysjRx4sRW9XnyySf1zDPPePrFxcXpgw8+0Lp165oMfPLy8rR06VKf5wsAAK5f7R74DB06tFXtz549q8rKSk2bNk0ZGRme81euXJHNZmuy3+OPP67s7GzPZ5fLpdjY2NZPGACAdsLj7IHX7oFP9+7dvT536tRJhuH9W7t8+bLn3263W9LVcldiYqJXu86dOzd5HavVKqvV6ut0AQBoP5S6As70fXy+853vyOFwyDAMWSxXl2iVlZV5vo+Ojta//du/6aOPPtIvfvELs6cHAACuY6YHPsnJyTp79qxWrFihn/3sZyooKNDbb7+tiIgIT5ucnBzNnj1bERERGjt2rGpqanTw4EFVV1d7lbMAAPjWIWMTUKa/q2vgwIFavXq1XnjhBX33u9/VgQMHNG/ePK82Dz30kF566SVt3LhRd9xxh0aOHKmNGzcqLi7O7OkCAOA39Wt8fD3QdhbjmwturhMul0s2m00/GLlEXbqEBno6QIfRZdehQE8BX/PpguGBnkLQq6u5pP97dqGcTqdX9cGf6v8mxU/PVecQ3/4m1dVe0tH17Tvf6xnv6gIAwCwsbg44Ah8AAEzC4+yBR+ADAIBZyPgEnOmLmwEAAAKFjA8AACah1BV4BD4AAJiFUlfAUeoCAABBg4wPAABmIeMTcAQ+AACYhDU+gUepCwAABA0yPgAAmIVSV8AR+AAAYBKLYcji4ysyfe0f7Ch1AQCAoEHGBwAAs1DqCjgCHwAATMJTXYFHqQsAALMYfjraSXV1tdLT02Wz2WSz2ZSenq5z584128cwDOXk5CgmJkZhYWFKTk7WsWPHvNqsX79eycnJioiIkMViaXTMllz75MmTuv/++9W9e3dFRkZq9uzZqq2tbdU9EvgAAABJUlpamsrKylRQUKCCggKVlZUpPT292T4rVqzQqlWrlJ+fr9LSUtntdo0ePVrnz5/3tPnyyy+VmpqqhQsXtvnadXV1uu+++3ThwgXt2bNHW7Zs0bZt2zR37txW3SOlLgAATNKRS13Hjx9XQUGB9u3bp8TEREnSiy++qKSkJJWXl6t///4N+hiGoeeee06LFi3SxIkTJUmvvPKKoqOjtXnzZs2YMUOSlJWVJUnavXt3m69dWFioDz74QJWVlYqJiZEkPfPMM5o6daqeeuopRUREtOg+yfgAAGAWP5a6XC6X11FTU+PT1EpKSmSz2TyBhyQNGzZMNptNe/fubbRPRUWFHA6HUlJSPOesVqtGjhzZZJ+2XrukpETx8fGeoEeSxowZo5qaGh06dKjF1yLwAQDgWyg2NtazHsZmsykvL8+n8RwOh6Kiohqcj4qKksPhaLKPJEVHR3udj46ObrJPW6/tcDgaXKdHjx4KCQlp1bUodQEAYBJ/lroqKyu9yjtWq7XR9jk5OVq6dGmzY5aWll4d22Jp8J1hGI2e95rTN75vSZ9rjdHYOG2d39cR+AAAYBY/7uMTERHRonUtmZmZmjJlSrNt+vbtqyNHjuj06dMNvjt79myDTEs9u90u6Wo2pnfv3p7zZ86cabJPU+Nc69p2u1379+/3+r66ulqXL19u1bUodQEAcB2LjIzUgAEDmj1CQ0OVlJQkp9OpAwcOePru379fTqdTw4cPb3TsuLg42e12FRUVec7V1taquLi4yT6Nacm1k5KSdPToUVVVVXnaFBYWymq1KiEhocXXIvABAMBE9eWuth7tZeDAgUpNTVVGRob27dunffv2KSMjQ+PGjfN6omvAgAHavn371XuxWJSVlaXc3Fxt375dR48e1dSpU9WtWzelpaV5+jgcDpWVlenEiROSpPfff19lZWX617/+1eJrp6SkaNCgQUpPT9fhw4f1v//7v5o3b54yMjJa/ESXROADAIB5DMM/RzvZtGmT7rjjDqWkpCglJUV33nmnXn31Va825eXlcjqdns/z589XVlaWZs6cqaFDh+rTTz9VYWGhwsPDPW3Wrl2rwYMHKyMjQ5I0YsQIDR48WDt27GjxtTt37qw333xToaGh+v73v69JkyZpwoQJWrlyZavu0WIY1+drXl0ul2w2m34wcom6dAkN9HSADqPLrpY/9on29+mClpcD0D7qai7p/55dKKfT2arMQWvU/00a+rNfq0tX3/4mXbl8SQf/v8XtOt/rGRkfAAAQNHiqCwAAs/B29oAj8AEAwCQW99XD1zHQdpS6AABA0CDjAwCAWSh1BRyBDwAAJunIb2cPFpS6AABA0CDjAwCAWfyxAeH1uf2eaQh8AAAwCaWuwKPUBQAAggYZHwAAzMJTXQFH4AMAgEkodQUegQ8AAGZhcXPAscYHAAAEDTI+AACYhFJX4F23gY/xVSrwypWaAM8E6GCMy4GeAb6mruZSoKcQ9Op/B4YZJSQWNwfcdRv4nD9/XpK072+/CfBMAKAZz/4l0DPAV86fPy+bzRboaaCdXbeBT0xMjCorKxUeHi6LxRLo6bSJy+VSbGysKisrFREREejpBDV+Fx0Lv4+O43r4XRiGofPnzysmJqbdr0WpK/Cu28CnU6dOuummmwI9Db+IiIj41v4P5XrD76Jj4ffRcXzbfxemZXrcxtXD1zHQZjzVBQAAgsZ1m/EBAKDDYXFzwBH4dGBWq1VLliyR1WoN9FSCHr+LjoXfR8fB76J1LPLDGh+/zCR4WQxTnt8DACB4uVwu2Ww2ff/eperSJdSnsa5cuaS/7Vwip9P5rV5XFShkfAAAMAuvrAg4Ah8AAEzC4+yBR+ADAIBZWNwccDzODgAAggYZHwAATGIxDFl8XKPja/9gR8anA1u9erXi4uIUGhqqhIQEvfvuu4GeUlD661//qvvvv18xMTGyWCz685//HOgpBaW8vDzdfffdCg8PV1RUlCZMmKDy8vJATysorVmzRnfeeadnt+akpCS9/fbbgZ7Wt4PbTwfajMCng9q6dauysrK0aNEiHT58WPfcc4/Gjh2rkydPBnpqQefChQv67ne/q/z8/EBPJagVFxdr1qxZ2rdvn4qKinTlyhWlpKTowoULgZ5a0Lnpppv0m9/8RgcPHtTBgwf1wx/+UD/5yU907NixQE8NuCb28emgEhMTNWTIEK1Zs8ZzbuDAgZowYYLy8vICOLPgZrFYtH37dk2YMCHQUwl6Z8+eVVRUlIqLizVixIhATyfo9ezZU08//bSmTZsW6Kl0SPX7+Iy45z/9so/PX99dxj4+bUTGpwOqra3VoUOHlJKS4nU+JSVFe/fuDdCsgI7F6XRKuvoHF4FTV1enLVu26MKFC0pKSgr0dDo+w08H2ozFzR3QZ599prq6OkVHR3udj46OlsPhCNCsgI7DMAxlZ2frBz/4geLj4wM9naD0/vvvKykpSZcuXdINN9yg7du3a9CgQYGeFnBNBD4dmMXi/UYWwzAanAOCUWZmpo4cOaI9e/YEeipBq3///iorK9O5c+e0bds2PfjggyouLib4uRZ2bg44Ap8OKDIyUp07d26Q3Tlz5kyDLBAQbB555BHt2LFDf/3rX3XTTTcFejpBKyQkRLfddpskaejQoSotLdVvf/tbrVu3LsAz69jYuTnwWOPTAYWEhCghIUFFRUVe54uKijR8+PAAzQoILMMwlJmZqT/96U/atWuX4uLiAj0lfI1hGKqpqQn0NIBrIvDpoLKzs/XSSy/pv/7rv3T8+HHNmTNHJ0+e1MMPPxzoqQWdL774QmVlZSorK5MkVVRUqKysjK0FTDZr1iy99tpr2rx5s8LDw+VwOORwOHTx4sVATy3oLFy4UO+++64+/vhjvf/++1q0aJF2796tX/ziF4GeWsdXX+ry9Wgn1dXVSk9Pl81mk81mU3p6us6dO3eNWzKUk5OjmJgYhYWFKTk5ucHWBuvXr1dycrIiIiJksVgajPnxxx9r2rRpiouLU1hYmG699VYtWbJEtbW1Xu0sFkuDY+3ata26R0pdHdTkyZP1+eefa9myZaqqqlJ8fLzeeust9enTJ9BTCzoHDx7UqFGjPJ+zs7MlSQ8++KA2btwYoFkFn/qtHZKTk73Ob9iwQVOnTjV/QkHs9OnTSk9PV1VVlWw2m+68804VFBRo9OjRgZ5ah2dxXz18HaO9pKWl6ZNPPlFBQYEkafr06UpPT9cbb7zRZJ8VK1Zo1apV2rhxo26//Xb9+te/1ujRo1VeXq7w8HBJ0pdffqnU1FSlpqbq8ccfbzDG3//+d7ndbq1bt0633Xabjh49qoyMDF24cEErV670arthwwalpqZ6PttstlbdI/v4AADQzur38Un+3iK/7OOz+8BTft/H5/jx4xo0aJD27dunxMRESdK+ffuUlJSkv//97+rfv3+DPoZhKCYmRllZWVqwYIEkqaamRtHR0Vq+fLlmzJjh1X737t0aNWqUqqurdeONNzY7n6efflpr1qzRRx995Dnnj73UKHUBAPAt5HK5vA5f11iVlJTIZrN5gh5JGjZsmGw2W5N7yFVUVMjhcHjtO2e1WjVy5Eif951zOp2N7tOVmZmpyMhI3X333Vq7dq3c7talwAh8AAAwix83MIyNjfWsxbHZbD7v6u9wOBQVFdXgfFRUVJN7yNWf9/e+c//4xz/0u9/9rsG61ieffFKvv/66du7cqSlTpmju3LnKzc1t1dis8QEAwCT+fDt7ZWWlV6nLarU22j4nJ0dLly5tdszS0tKrYzeyV1xL9pDz575zp06dUmpqqn7+85/roYce8vpu8eLFnn/fddddkqRly5Z5nb8WAh8AAL6FIiIiWrTGJzMzU1OmTGm2Td++fXXkyBGdPn26wXdnz55tcg85u90u6Wrmp3fv3p7zbd137tSpUxo1apSSkpK0fv36a7YfNmyYXC6XTp8+3eLrEfgAAGCWAOzcHBkZqcjIyGu2S0pKktPp1IEDB/S9731PkrR//345nc4m95CLi4uT3W5XUVGRBg8eLOnq+yaLi4u1fPnyVs3z008/1ahRo5SQkKANGzaoU6drr8Y5fPiwQkNDr7lQ+usIfAAAMIshydfH0dvpWeyBAwcqNTVVGRkZnh24p0+frnHjxnk90TVgwADl5eXpgQcekMViUVZWlnJzc9WvXz/169dPubm56tatm9LS0jx96vfdOnHihKSr73oLDw/XzTffrJ49e+rUqVNKTk7WzTffrJUrV+rs2bOevvVZpTfeeEMOh0NJSUkKCwvTO++8o0WLFmn69OlNlvkaQ+ADAAAkSZs2bdLs2bM9T2mNHz9e+fn5Xm3Ky8vldDo9n+fPn6+LFy9q5syZqq6uVmJiogoLCz17+EjS2rVrvdYZjRgxQtL/24ersLBQJ06c0IkTJxq8iqZ+152uXbtq9erVys7Oltvt1i233KJly5Zp1qxZrbpH9vEBAKCd1e/j88PBj6lLZx/38am7pF2Hf+P3fXyCBRkfAADMYsgPa3z8MpOgxT4+AAAgaJDxAQDALAF4qgveCHwAADCLW1Lb9vXzHgNtRuADAIBJ/LlzM9qGNT4AACBokPEBAMAsrPEJOAIfAADMQuATcJS6AABA0CDjAwCAWcj4BByBDwAAZuFx9oCj1AUAAIIGGR8AAEzCPj6BR+ADAIBZWOMTcJS6AABA0CDjAwCAWdyGZPExY+Mm4+MLAh8AAMxCqSvgCHwAADCNHwIfEfj4gjU+AAAgaJDxAQDALJS6Ao7ABwAAs7gN+VyqYnGzTyh1AQCAoEHGBwAAsxjuq4evY6DNCHwAADALa3wCjlIXAAAIGmR8AAAwC4ubA47ABwAAs1DqCjhKXQAAIGiQ8QEAwCyG/JDx8ctMghaBDwAAZqHUFXAEPgAAmMXtluTjPjxu9vHxBWt8AABA0CDjAwCAWSh1BRyBDwAAZiHwCThKXQAAIGiQ8QEAwCzs3BxwZHwAADCJYbj9crSX6upqpaeny2azyWazKT09XefOnbvGPRnKyclRTEyMwsLClJycrGPHjnm1Wb9+vZKTkxURESGLxdLomH379pXFYvE6HnvsMa82J0+e1P3336/u3bsrMjJSs2fPVm1tbavukcAHAABIktLS0lRWVqaCggIVFBSorKxM6enpzfZZsWKFVq1apfz8fJWWlsput2v06NE6f/68p82XX36p1NRULVy4sNmxli1bpqqqKs+xePFiz3d1dXW67777dOHCBe3Zs0dbtmzRtm3bNHfu3FbdI6UuAADMYhi+l6raaXHz8ePHVVBQoH379ikxMVGS9OKLLyopKUnl5eXq379/I1Mx9Nxzz2nRokWaOHGiJOmVV15RdHS0Nm/erBkzZkiSsrKyJEm7d+9udg7h4eGy2+2NfldYWKgPPvhAlZWViomJkSQ988wzmjp1qp566ilFRES06D7J+AAAYJb6p7p8PSS5XC6vo6amxqeplZSUyGazeYIeSRo2bJhsNpv27t3baJ+Kigo5HA6lpKR4zlmtVo0cObLJPs1Zvny5evXqpbvuuktPPfWUVxmrpKRE8fHxnqBHksaMGaOamhodOnSoxdcg4wMAwLdQbGys1+clS5YoJyenzeM5HA5FRUU1OB8VFSWHw9FkH0mKjo72Oh8dHa1//vOfrbr+o48+qiFDhqhHjx46cOCAHn/8cVVUVOill17yXOub1+nRo4dCQkKanF9jCHwAADCL2y1ZfFyc/NXi5srKSq/yjtVqbbR5Tk6Oli5d2uyQpaWlkiSLxdLwcobR6Pmv++b3LenzTXPmzPH8+84771SPHj30s5/9zJMF8mV+X0fgAwCAWQw/PM7+VakrIiKiRetaMjMzNWXKlGbb9O3bV0eOHNHp06cbfHf27NkGmZZ69etxHA6Hevfu7Tl/5syZJvu01LBhwyRJJ06cUK9evWS327V//36vNtXV1bp8+XKrrkXgAwCASQy3W4aPGZ/WPs4eGRmpyMjIa7ZLSkqS0+nUgQMH9L3vfU+StH//fjmdTg0fPrzRPnFxcbLb7SoqKtLgwYMlSbW1tSouLtby5ctbNc9vOnz4sCR5AqqkpCQ99dRTqqqq8pwrLCyU1WpVQkJCi8cl8AEAABo4cKBSU1OVkZGhdevWSZKmT5+ucePGeT3RNWDAAOXl5emBBx6QxWJRVlaWcnNz1a9fP/Xr10+5ubnq1q2b0tLSPH0cDoccDodOnDghSXr//fcVHh6um2++WT179lRJSYn27dunUaNGyWazqbS0VHPmzNH48eN18803S5JSUlI0aNAgpaen6+mnn9a//vUvzZs3TxkZGS1+oksi8AEAwDx+LHW1h02bNmn27Nmep7TGjx+v/Px8rzbl5eVyOp2ez/Pnz9fFixc1c+ZMVVdXKzExUYWFhQoPD/e0Wbt2rdc6oxEjRkiSNmzYoKlTp8pqtWrr1q1aunSpampq1KdPH2VkZGj+/PmePp07d9abb76pmTNn6vvf/77CwsKUlpamlStXtuoeLYbB284AAGhPLpdLNptNP7ROUhdLiE9jXTFqtavmj3I6na3KdOAq9vEBAABBg1IXAABmMQxJvj7OTqHGFwQ+AACYxHAbMiy+BS6sUPENpS4AABA0yPgAAGAWwy3fS10+9g9yBD4AAJiEUlfgUeoCAABBg4wPAAAmuWLU+FyquqLLfppNcCLwAQCgnYWEhMhut2uP4y2/jGe32xUS4ttGiMGKnZsBADDBpUuXVFtb65exQkJCFBoa6pexgg2BDwAACBosbgYAAEGDwAcAAAQNAh8AABA0CHwAAEDQIPABAABBg8AHAAAEDQIfAAAQNP5/+r5LEKFP91kAAAAASUVORK5CYII=\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(metrics[:, :, 1, 0].T- compute_auc(data_dict, datasets=[\"mask-ts\"], stage='para_merge', gt_stage='para_gt' ))\n",
    "plt.yticks(do_3D_vals, do_3D_vals)\n",
    "plt.xticks(radii)\n",
    "plt.colorbar()\n",
    "\n",
    "print(metrics[:, :, 1, 0].max())"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 298,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.6859504092823375\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 2 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(metrics[:, :, 1, 1].T - compute_auc(data_dict, datasets=[\"mask-ts\"], stage='para_merge', gt_stage='para_gt_merge'))\n",
    "plt.yticks(do_3D_vals, do_3D_vals)\n",
    "plt.xticks(radii)\n",
    "plt.colorbar()\n",
    "\n",
    "print(metrics[:, :, 1, 1].max())"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "--> only small improvement, always best without 3D and with aggressive smoothing"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 299,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5644745826721191\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 2 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# mask-r and mask-ts\n",
    "plt.imshow(metrics[:, :, 2, 0].T- compute_auc(data_dict, stage='para_merge', gt_stage='para_gt' ))\n",
    "plt.yticks(do_3D_vals, do_3D_vals)\n",
    "plt.xticks(radii)\n",
    "plt.colorbar()\n",
    "\n",
    "print(metrics[:, :, 2, 0].max())"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 300,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5968379540876909\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 2 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(metrics[:, :, 2, 1].T -compute_auc(data_dict, stage='para_merge', gt_stage='para_gt_merge' ))\n",
    "plt.yticks(do_3D_vals, do_3D_vals)\n",
    "plt.xticks(radii)\n",
    "plt.colorbar()\n",
    "print(metrics[:, :, 2, 1].max())"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "# on relevant data (only qualitatively)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "data_dir = os.path.join(get_path(), \"wholelife\")\n",
    "dataset = \"wholelife\"\n",
    "exp_str_rbc = \"data_valid1_valid2_valid3_by_stack_mode_3D_iso_min_train_masks_0_seed_0_fold_0_of_1_epoch_499_aniso_3.2\""
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [],
   "source": [
    "imgs, _, file_names = load_data(data_dir, dataset)\n",
    "pred_rbc = load_predictions(data_dir, dataset, file_names, exp_str_rbc)\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "outputs": [
    {
     "data": {
      "text/plain": "array(['C2-FCR3_KM_B4_15_03processed_1-1',\n       'C2-FCR3_KM_B4_15_03processed_1-2',\n       'C2-FCR3_KM_B4_15_03processed_1-3',\n       'C2-FCR3_KM_B4_15_03processed_11-1',\n       'C2-FCR3_KM_B4_15_03processed_12-1',\n       'C2-FCR3_KM_B4_15_03processed_13-2',\n       'C2-FCR3_KM_B4_15_03processed_13-3',\n       'C2-FCR3_KM_B4_15_03processed_14-1',\n       'C2-FCR3_KM_B4_15_03processed_15-3',\n       'C2-FCR3_KM_B4_15_03processed_15-4',\n       'C2-FCR3_KM_B4_15_03processed_15-5',\n       'C2-FCR3_KM_B4_15_03processed_16-1',\n       'C2-FCR3_KM_B4_15_03processed_17-1',\n       'C2-FCR3_KM_B4_15_03processed_17-2',\n       'C2-FCR3_KM_B4_15_03processed_2-1',\n       'C2-FCR3_KM_B4_15_03processed_5-1',\n       'C2-FCR3_KM_B4_15_03processed_6-1',\n       'C2-FCR3_KM_B4_18_processed-1-1',\n       'C2-FCR3_KM_B4_18_processed-10-1',\n       'C2-FCR3_KM_B4_18_processed-10-2',\n       'C2-FCR3_KM_B4_18_processed-11-1',\n       'C2-FCR3_KM_B4_18_processed-16-2',\n       'C2-FCR3_KM_B4_18_processed-18-1',\n       'C2-FCR3_KM_B4_18_processed-4-1', 'C2-FCR3_KM_B4_18_processed-5-1',\n       'C2-FCR3_KM_B4_18_processed-6-1', 'C2-FCR3_KM_B4_18_processed-7-1',\n       'C2-FCR3_KM_B4_18_processed-9-1',\n       'C2-FCR3_KM_B4_21_-01processed-1-1',\n       'C2-FCR3_KM_B4_21_-01processed-1-2',\n       'C2-FCR3_KM_B4_21_-01processed-11-1',\n       'C2-FCR3_KM_B4_21_-01processed-11-2',\n       'C2-FCR3_KM_B4_21_-01processed-13-1',\n       'C2-FCR3_KM_B4_21_-01processed-15-1',\n       'C2-FCR3_KM_B4_21_-01processed-17-1',\n       'C2-FCR3_KM_B4_21_-01processed-17-2',\n       'C2-FCR3_KM_B4_21_-01processed-19-1',\n       'C2-FCR3_KM_B4_21_-01processed-21-1',\n       'C2-FCR3_KM_B4_21_-01processed-21-2',\n       'C2-FCR3_KM_B4_21_-01processed-7-1',\n       'C2-FCR3_KM_B4_21_-01processed-8-1',\n       'C2-FCR3_KM_B4_21_-01processed-8-2',\n       'C2-FCR3_KM_B4_21_-01processed-9-1'], dtype='<U34')"
     },
     "execution_count": 85,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "file_names"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "outputs": [
    {
     "data": {
      "text/plain": "(array([37]),)"
     },
     "execution_count": 136,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.where(file_names == \"C2-FCR3_KM_B4_21_-01processed-21-1\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "outputs": [
    {
     "data": {
      "text/plain": "'C2-FCR3_KM_B4_21_-01processed-21-1'"
     },
     "execution_count": 138,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img_id = 37\n",
    "file_names[img_id]"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "outputs": [],
   "source": [
    "seg = pred_rbc[img_id]"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.image.AxesImage at 0x7f1b4c317b80>"
     },
     "execution_count": 140,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "t = 20\n",
    "z = 12\n",
    "\n",
    "plt.imshow(seg[t][z], interpolation=\"none\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5\n"
     ]
    }
   ],
   "source": [
    "id_main = find_central_id(seg[t], 30)\n",
    "print(id_main)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.image.AxesImage at 0x7f1b4c275ca0>"
     },
     "execution_count": 142,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAakAAAGhCAYAAADbf0s2AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAA9hAAAPYQGoP6dpAAAjMUlEQVR4nO3dfXBU5fn/8c+SDUtIk5WEsstKAmEmU5UgarBMIy2xQBjkQYepqCDSqdOBIkga5CGlD9SpidAp0pqKA+OIlSJOR0Dq2EJQDDKxBQOpgK3oNIXwsJM+pLtBYhLI/f2jP86PJQkQ2GTvDe/XzJn57n3uXa6r+cLH6+zJrssYYwQAgIV6xboAAAA6QkgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsFdOQeuGFF5SVlaU+ffooNzdX77//fizLAQBYJmYh9frrr6uwsFDLly/XwYMH9fWvf10TJ07U8ePHY1USAMAyrlh9wOyoUaN01113ae3atc7arbfeqgceeEClpaWXfW5ra6tOnTqllJQUuVyuri4VABBlxhg1NDQoEAioV6+O5yV3N9bkaG5uVlVVlZYtWxaxXlBQoMrKyjb7m5qa1NTU5Dw+efKkbrvtti6vEwDQtWprazVo0KAOz8ckpP71r3/p/Pnz8vl8Ees+n0/BYLDN/tLSUv30pz9tsz5a98mtxC6rEwDQNc6pRXv1tlJSUi67LyYhdcGll+qMMe1evisuLlZRUZHzOBwOKyMjQ24lyu0ipAAg7vy/N5qu9JZNTEKqf//+SkhIaDM11dXVtZmuJMnj8cjj8XRXeQAAS8Tk7r7evXsrNzdX5eXlEevl5eXKy8uLRUkAAAvF7HJfUVGRZs2apZEjR+prX/ua1q1bp+PHj2vu3LmxKgkAYJmYhdRDDz2kf//733r66ad1+vRp5eTk6O2339bgwYNjVRIAwDIx+z2p6xEOh+X1epWv+7lxAgDi0DnTovf0pkKhkFJTUzvcx2f3AQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKxFSAEArEVIAQCsRUgBAKwV9ZAqLS3V3XffrZSUFA0YMEAPPPCAPvnkk4g9xhitWLFCgUBASUlJys/P15EjR6JdCgAgzkU9pCoqKvTEE0/oT3/6k8rLy3Xu3DkVFBTo888/d/asWrVKq1evVllZmfbv3y+/36/x48eroaEh2uUAAOKYyxhjuvIP+Oc//6kBAwaooqJC3/jGN2SMUSAQUGFhoZYuXSpJampqks/n08qVKzVnzpwrvmY4HJbX61W+7pfbldiV5QMAusA506L39KZCoZBSU1M73Nfl70mFQiFJUlpamiSppqZGwWBQBQUFzh6Px6MxY8aosrKy3ddoampSOByOOAAAPV+XhpQxRkVFRRo9erRycnIkScFgUJLk8/ki9vp8PufcpUpLS+X1ep0jIyOjK8sGAFiiS0Nq/vz5+uijj/Taa6+1OedyuSIeG2ParF1QXFysUCjkHLW1tV1SLwDALu6ueuEFCxZo+/bt2rNnjwYNGuSs+/1+Sf+bqAYOHOis19XVtZmuLvB4PPJ4PF1VKgDAUlGfpIwxmj9/vrZs2aJ3331XWVlZEeezsrLk9/tVXl7urDU3N6uiokJ5eXnRLgcAEMeiPkk98cQT2rRpk958802lpKQ47zN5vV4lJSXJ5XKpsLBQJSUlys7OVnZ2tkpKStS3b1/NmDEj2uUAAOJY1ENq7dq1kqT8/PyI9Zdfflnf/va3JUlLlixRY2Oj5s2bp/r6eo0aNUo7d+5USkpKtMsBAMSxLv89qa7A70kBQHyz5vekAAC4VoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFpdHlKlpaVyuVwqLCx01owxWrFihQKBgJKSkpSfn68jR450dSkAgDjTpSG1f/9+rVu3TrfffnvE+qpVq7R69WqVlZVp//798vv9Gj9+vBoaGrqyHABAnOmykDpz5oxmzpyp9evXq1+/fs66MUZr1qzR8uXLNW3aNOXk5OiVV17R2bNntWnTpnZfq6mpSeFwOOIAAPR8XRZSTzzxhCZNmqRx48ZFrNfU1CgYDKqgoMBZ83g8GjNmjCorK9t9rdLSUnm9XufIyMjoqrIBABbpkpDavHmzDhw4oNLS0jbngsGgJMnn80Ws+3w+59yliouLFQqFnKO2tjb6RQMArOOO9gvW1tZq4cKF2rlzp/r06dPhPpfLFfHYGNNm7QKPxyOPxxPVOgEA9ov6JFVVVaW6ujrl5ubK7XbL7XaroqJCv/rVr+R2u50J6tKpqa6urs10BQC4sUU9pMaOHatDhw6purraOUaOHKmZM2equrpaQ4cOld/vV3l5ufOc5uZmVVRUKC8vL9rlAADiWNQv96WkpCgnJydiLTk5Wenp6c56YWGhSkpKlJ2drezsbJWUlKhv376aMWNGtMsBAMSxqIfU1ViyZIkaGxs1b9481dfXa9SoUdq5c6dSUlJiUQ4AwFIuY4yJdRGdFQ6H5fV6la/75XYlxrocAEAnnTMtek9vKhQKKTU1tcN9fHYfAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWjH50kMA3WPHqepu+7MmBO7otj8LNw4mKQCAtQgpAIC1uNwHxLnuvKR3Oe3VwSVAXC8mKQCAtZikgDhiy9R0tS6ul6kK14JJCgBgLUIKAGAtLvcBlou3S3wdudAHl/3QGUxSAABrMUkBluopExRwPZikAADWIqQAANbich9gmZ5+mY/fnUJnMEkBAKzFJAVYoKdPT8C1YpICAFiLkAIAWIvLfUAM3eiX+fgUClwJkxQAwFpMUkA3u9GnJ6AzmKQAANYipAAA1uqSkDp58qQeffRRpaenq2/fvrrjjjtUVVXlnDfGaMWKFQoEAkpKSlJ+fr6OHDnSFaUAAOJY1EOqvr5e99xzjxITE/WHP/xBH3/8sX7xi1/opptucvasWrVKq1evVllZmfbv3y+/36/x48eroaEh2uUAAOJY1G+cWLlypTIyMvTyyy87a0OGDHH+b2OM1qxZo+XLl2vatGmSpFdeeUU+n0+bNm3SnDlzol0SYAVumAA6L+qT1Pbt2zVy5Eg9+OCDGjBggO68806tX7/eOV9TU6NgMKiCggJnzePxaMyYMaqsrGz3NZuamhQOhyMOAEDPF/WQ+vvf/661a9cqOztbO3bs0Ny5c/Xkk0/qN7/5jSQpGAxKknw+X8TzfD6fc+5SpaWl8nq9zpGRkRHtsoEuseNUtXOgY/zvhI5EPaRaW1t11113qaSkRHfeeafmzJmj7373u1q7dm3EPpfLFfHYGNNm7YLi4mKFQiHnqK2tjXbZAAALRT2kBg4cqNtuuy1i7dZbb9Xx48clSX6/X5LaTE11dXVtpqsLPB6PUlNTIw4AQM8X9ZC655579Mknn0SsHT16VIMHD5YkZWVlye/3q7y83Dnf3NysiooK5eXlRbscAEAci/rdfd///veVl5enkpISTZ8+Xfv27dO6deu0bt06Sf+7zFdYWKiSkhJlZ2crOztbJSUl6tu3r2bMmBHtcgAAcSzqIXX33Xdr69atKi4u1tNPP62srCytWbNGM2fOdPYsWbJEjY2Nmjdvnurr6zVq1Cjt3LlTKSkp0S4HiAluALh2fDI6LuYyxphYF9FZ4XBYXq9X+bpfbldirMsB2iCkrh8h1bOdMy16T28qFApd9j4DPrsPAGAtvqoDiBKmJyD6mKQAANYipAAA1iKkAADWIqQAANbixgngOnHDBNB1mKQAANZikgJgpYsnVH6x98bFJAUAsBYhBQCwFpf7gGvAzRJA92CSAgBYi5ACAFiLkAIAWIuQAgBYixsngE7ghgmgezFJAQCsRUgBsN6OU9VMsTcoQgoAYC1CCgBgLW6cAK6Ay0xA7DBJAQCsxSQFIG7w9R03HiYpAIC1CCkAgLUIKQCAtQgpAIC1CCkAgLUIKQCAtbgFHegAv8QLxB6TFADAWoQUAMBahBQAwFqEFADAWtw4AVyEmyUAuzBJAQCsRUgBAKwV9ZA6d+6cfvjDHyorK0tJSUkaOnSonn76abW2tjp7jDFasWKFAoGAkpKSlJ+fryNHjkS7FAA9zITAHc6BG0PUQ2rlypV68cUXVVZWpr/+9a9atWqVfv7zn+v555939qxatUqrV69WWVmZ9u/fL7/fr/Hjx6uhoSHa5QAA4ljUb5z44IMPdP/992vSpEmSpCFDhui1117Thx9+KOl/U9SaNWu0fPlyTZs2TZL0yiuvyOfzadOmTZozZ060SwIQ55icblxRn6RGjx6td955R0ePHpUk/eUvf9HevXt13333SZJqamoUDAZVUFDgPMfj8WjMmDGqrKxs9zWbmpoUDocjDgBAzxf1SWrp0qUKhUK65ZZblJCQoPPnz+uZZ57RI488IkkKBoOSJJ/PF/E8n8+nY8eOtfuapaWl+ulPfxrtUgEAlot6SL3++uvauHGjNm3apGHDhqm6ulqFhYUKBAKaPXu2s8/lckU8zxjTZu2C4uJiFRUVOY/D4bAyMjKiXToAi3CJD1IXhNTixYu1bNkyPfzww5Kk4cOH69ixYyotLdXs2bPl9/sl/W+iGjhwoPO8urq6NtPVBR6PRx6PJ9qlAgAsF/X3pM6ePatevSJfNiEhwbkFPSsrS36/X+Xl5c755uZmVVRUKC8vL9rlAIgz3GKOi0V9kpoyZYqeeeYZZWZmatiwYTp48KBWr16t73znO5L+d5mvsLBQJSUlys7OVnZ2tkpKStS3b1/NmDEj2uUAAOJY1EPq+eef149+9CPNmzdPdXV1CgQCmjNnjn784x87e5YsWaLGxkbNmzdP9fX1GjVqlHbu3KmUlJRolwMAiGMuY4yJdRGdFQ6H5fV6la/75XYlxroc9CB8wGxscHnvxnPOtOg9valQKKTU1NQO9/HZfQAAa/FVHQBihgkKV8IkBQCwFiEFALAWIQUAsBYhBQCwFjdOAOhW3CyBzmCSAgBYi5ACAFiLy30AugyX9nC9mKQAANZikgIQdUxQiBYmKQCAtQgpAIC1uNwH4LpwaQ9diUkKAGAtJikAV42pCd2NSQoAYC0mKeAiF08KN+pXyTMtwSZMUgAAaxFSAABrcbkPuIFxaQ+2Y5ICAFiLSQrowIUpo6fcQMHUhHjEJAUAsBYhBQCwFpf7gB6IS3voKZikAADWYpICriBePoWC6Qk9EZMUAMBahBQAwFpc7gPiEJf2cKNgkgIAWItJCuiEWH8KBRMUbjRMUgAAaxFSAABrcbkPsBSX9gAmKQCAxTodUnv27NGUKVMUCATkcrm0bdu2iPPGGK1YsUKBQEBJSUnKz8/XkSNHIvY0NTVpwYIF6t+/v5KTkzV16lSdOHHiuhoButOEwB3OEe3Xi+brAvGu0yH1+eefa8SIESorK2v3/KpVq7R69WqVlZVp//798vv9Gj9+vBoaGpw9hYWF2rp1qzZv3qy9e/fqzJkzmjx5ss6fP3/tnQAAepxOvyc1ceJETZw4sd1zxhitWbNGy5cv17Rp0yRJr7zyinw+nzZt2qQ5c+YoFArppZde0quvvqpx48ZJkjZu3KiMjAzt2rVLEyZMuI52AAA9SVTfk6qpqVEwGFRBQYGz5vF4NGbMGFVWVkqSqqqq1NLSErEnEAgoJyfH2XOppqYmhcPhiAOwxfVcnuPSHnB5UQ2pYDAoSfL5fBHrPp/PORcMBtW7d2/169evwz2XKi0tldfrdY6MjIxolg0AsFSX3ILucrkiHhtj2qxd6nJ7iouLVVRU5DwOh8MEFaxztV/pweQEXL2oTlJ+v1+S2kxEdXV1znTl9/vV3Nys+vr6DvdcyuPxKDU1NeIAAPR8UQ2prKws+f1+lZeXO2vNzc2qqKhQXl6eJCk3N1eJiYkRe06fPq3Dhw87ewAAkK7hct+ZM2f02WefOY9rampUXV2ttLQ0ZWZmqrCwUCUlJcrOzlZ2drZKSkrUt29fzZgxQ5Lk9Xr1+OOPa9GiRUpPT1daWpqeeuopDR8+3LnbD+iJuMwHdF6nQ+rDDz/Uvffe6zy+8F7R7NmztWHDBi1ZskSNjY2aN2+e6uvrNWrUKO3cuVMpKSnOc5577jm53W5Nnz5djY2NGjt2rDZs2KCEhIQotAQA6ClcxhgT6yI6KxwOy+v1Kl/3y+1KjHU5AIBOOmda9J7eVCgUuux9Bnx2HwDAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqdDqk9e/ZoypQpCgQCcrlc2rZtm3OupaVFS5cu1fDhw5WcnKxAIKDHHntMp06diniNpqYmLViwQP3791dycrKmTp2qEydOXHczAICepdMh9fnnn2vEiBEqKytrc+7s2bM6cOCAfvSjH+nAgQPasmWLjh49qqlTp0bsKyws1NatW7V582bt3btXZ86c0eTJk3X+/Plr7wQA0OO4jDHmmp/scmnr1q164IEHOtyzf/9+ffWrX9WxY8eUmZmpUCikL3/5y3r11Vf10EMPSZJOnTqljIwMvf3225owYcIV/9xwOCyv16t83S+3K/FaywcAxMg506L39KZCoZBSU1M73Nfl70mFQiG5XC7ddNNNkqSqqiq1tLSooKDA2RMIBJSTk6PKysp2X6OpqUnhcDjiAAD0fF0aUl988YWWLVumGTNmOEkZDAbVu3dv9evXL2Kvz+dTMBhs93VKS0vl9XqdIyMjoyvLBgBYostCqqWlRQ8//LBaW1v1wgsvXHG/MUYul6vdc8XFxQqFQs5RW1sb7XIBABbqkpBqaWnR9OnTVVNTo/Ly8ojrjX6/X83Nzaqvr494Tl1dnXw+X7uv5/F4lJqaGnEAAHq+qIfUhYD69NNPtWvXLqWnp0ecz83NVWJiosrLy52106dP6/Dhw8rLy4t2OQCAOObu7BPOnDmjzz77zHlcU1Oj6upqpaWlKRAI6Fvf+pYOHDigt956S+fPn3feZ0pLS1Pv3r3l9Xr1+OOPa9GiRUpPT1daWpqeeuopDR8+XOPGjYteZwCAuNfpkPrwww917733Oo+LiookSbNnz9aKFSu0fft2SdIdd9wR8bzdu3crPz9fkvTcc8/J7XZr+vTpamxs1NixY7VhwwYlJCRcYxsAgJ7oun5PKlb4PSkAiG/W/J4UAADXipACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFiLkAIAWIuQAgBYi5ACAFir0yG1Z88eTZkyRYFAQC6XS9u2betw75w5c+RyubRmzZqI9aamJi1YsED9+/dXcnKypk6dqhMnTnS2FABAD9fpkPr88881YsQIlZWVXXbftm3b9Oc//1mBQKDNucLCQm3dulWbN2/W3r17debMGU2ePFnnz5/vbDkAgB7M3dknTJw4URMnTrzsnpMnT2r+/PnasWOHJk2aFHEuFArppZde0quvvqpx48ZJkjZu3KiMjAzt2rVLEyZM6GxJAIAeKurvSbW2tmrWrFlavHixhg0b1uZ8VVWVWlpaVFBQ4KwFAgHl5OSosrKy3ddsampSOByOOAAAPV/UQ2rlypVyu9168skn2z0fDAbVu3dv9evXL2Ld5/MpGAy2+5zS0lJ5vV7nyMjIiHbZAAALRTWkqqqq9Mtf/lIbNmyQy+Xq1HONMR0+p7i4WKFQyDlqa2ujUS4AwHJRDan3339fdXV1yszMlNvtltvt1rFjx7Ro0SINGTJEkuT3+9Xc3Kz6+vqI59bV1cnn87X7uh6PR6mpqREHAKDni2pIzZo1Sx999JGqq6udIxAIaPHixdqxY4ckKTc3V4mJiSovL3eed/r0aR0+fFh5eXnRLAcAEOc6fXffmTNn9NlnnzmPa2pqVF1drbS0NGVmZio9PT1if2Jiovx+v77yla9Ikrxerx5//HEtWrRI6enpSktL01NPPaXhw4c7d/sBACBdQ0h9+OGHuvfee53HRUVFkqTZs2drw4YNV/Uazz33nNxut6ZPn67GxkaNHTtWGzZsUEJCQmfLAQD0YC5jjIl1EZ0VDofl9XqVr/vldiXGuhwAQCedMy16T28qFApd9j4DPrsPAGAtQgoAYC1CCgBgLUIKAGAtQgoAYC1CCgBgLUIKAGAtQgoAYC1CCgBgLUIKAGAtQgoAYC1CCgBgLUIKAGAtQgoAYC1CCgBgrU5/6aENLnwF1jm1SHH3bVgAgHNqkfT//z3vSFyGVENDgyRpr96OcSUAgOvR0NAgr9fb4fm4/Gbe1tZWnTp1SsYYZWZmqra29rLf7GizcDisjIwMeoixntCD1DP6oAd7dGUfxhg1NDQoEAioV6+O33mKy0mqV69eGjRokMLhsCQpNTU1rv8fQaIHW/SEHqSe0Qc92KOr+rjcBHUBN04AAKxFSAEArBXXIeXxePSTn/xEHo8n1qVcM3qwQ0/oQeoZfdCDPWzoIy5vnAAA3BjiepICAPRshBQAwFqEFADAWoQUAMBahBQAwFpxG1IvvPCCsrKy1KdPH+Xm5ur999+PdUkdKi0t1d13362UlBQNGDBADzzwgD755JOIPcYYrVixQoFAQElJScrPz9eRI0diVPGVlZaWyuVyqbCw0FmLlx5OnjypRx99VOnp6erbt6/uuOMOVVVVOedt7+PcuXP64Q9/qKysLCUlJWno0KF6+umn1dra6uyxrYc9e/ZoypQpCgQCcrlc2rZtW8T5q6m3qalJCxYsUP/+/ZWcnKypU6fqxIkT3djF5ftoaWnR0qVLNXz4cCUnJysQCOixxx7TqVOnrOrjSj+Li82ZM0cul0tr1qyJWO/WHkwc2rx5s0lMTDTr1683H3/8sVm4cKFJTk42x44di3Vp7ZowYYJ5+eWXzeHDh011dbWZNGmSyczMNGfOnHH2PPvssyYlJcW88cYb5tChQ+ahhx4yAwcONOFwOIaVt2/fvn1myJAh5vbbbzcLFy501uOhh//85z9m8ODB5tvf/rb585//bGpqasyuXbvMZ5995uyxvY+f/exnJj093bz11lumpqbG/O53vzNf+tKXzJo1a5w9tvXw9ttvm+XLl5s33njDSDJbt26NOH819c6dO9fcfPPNpry83Bw4cMDce++9ZsSIEebcuXNW9PHf//7XjBs3zrz++uvmb3/7m/nggw/MqFGjTG5ubsRrxLqPK/0sLti6dasZMWKECQQC5rnnnos41509xGVIffWrXzVz586NWLvlllvMsmXLYlRR59TV1RlJpqKiwhhjTGtrq/H7/ebZZ5919nzxxRfG6/WaF198MVZltquhocFkZ2eb8vJyM2bMGCek4qWHpUuXmtGjR3d4Ph76mDRpkvnOd74TsTZt2jTz6KOPGmPs7+HSfxivpt7//ve/JjEx0WzevNnZc/LkSdOrVy/zxz/+sdtqv9jl/oG/YN++fUaS8x/QtvXRUQ8nTpwwN998szl8+LAZPHhwREh1dw9xd7mvublZVVVVKigoiFgvKChQZWVljKrqnFAoJElKS0uTJNXU1CgYDEb05PF4NGbMGOt6euKJJzRp0iSNGzcuYj1eeti+fbtGjhypBx98UAMGDNCdd96p9evXO+fjoY/Ro0frnXfe0dGjRyVJf/nLX7R3717dd999kuKjh4tdTb1VVVVqaWmJ2BMIBJSTk2NlTxeEQiG5XC7ddNNNkuKjj9bWVs2aNUuLFy/WsGHD2pzv7h7i7lPQ//Wvf+n8+fPy+XwR6z6fT8FgMEZVXT1jjIqKijR69Gjl5ORIklN3ez0dO3as22vsyObNm3XgwAHt37+/zbl46eHvf/+71q5dq6KiIv3gBz/Qvn379OSTT8rj8eixxx6Liz6WLl2qUCikW265RQkJCTp//ryeeeYZPfLII5Li52dxwdXUGwwG1bt3b/Xr16/NHlv/3n/xxRdatmyZZsyY4XyCeDz0sXLlSrndbj355JPtnu/uHuIupC5wuVwRj40xbdZsNH/+fH300Ufau3dvm3M291RbW6uFCxdq586d6tOnT4f7bO5B+t9/JY4cOVIlJSWSpDvvvFNHjhzR2rVr9dhjjzn7bO7j9ddf18aNG7Vp0yYNGzZM1dXVKiwsVCAQ0OzZs519NvfQnmup19aeWlpa9PDDD6u1tVUvvPDCFffb0kdVVZV++ctf6sCBA52up6t6iLvLff3791dCQkKbxK6rq2vzX2K2WbBggbZv367du3dr0KBBzrrf75ckq3uqqqpSXV2dcnNz5Xa75Xa7VVFRoV/96ldyu91OnTb3IEkDBw7UbbfdFrF266236vjx45Li42exePFiLVu2TA8//LCGDx+uWbNm6fvf/75KS0slxUcPF7uaev1+v5qbm1VfX9/hHlu0tLRo+vTpqqmpUXl5ecT3MNnex/vvv6+6ujplZmY6f8+PHTumRYsWaciQIZK6v4e4C6nevXsrNzdX5eXlEevl5eXKy8uLUVWXZ4zR/PnztWXLFr377rvKysqKOJ+VlSW/3x/RU3NzsyoqKqzpaezYsTp06JCqq6udY+TIkZo5c6aqq6s1dOhQ63uQpHvuuafN7f9Hjx7V4MGDJcXHz+Ls2bNtvsk0ISHBuQU9Hnq42NXUm5ubq8TExIg9p0+f1uHDh63q6UJAffrpp9q1a5fS09Mjztvex6xZs/TRRx9F/D0PBAJavHixduzYISkGPUT9VoxucOEW9Jdeesl8/PHHprCw0CQnJ5t//OMfsS6tXd/73veM1+s17733njl9+rRznD171tnz7LPPGq/Xa7Zs2WIOHTpkHnnkEatue27PxXf3GRMfPezbt8+43W7zzDPPmE8//dT89re/NX379jUbN2509tjex+zZs83NN9/s3IK+ZcsW079/f7NkyRJnj209NDQ0mIMHD5qDBw8aSWb16tXm4MGDzl1vV1Pv3LlzzaBBg8yuXbvMgQMHzDe/+c1uvwX9cn20tLSYqVOnmkGDBpnq6uqIv+tNTU3W9HGln8WlLr27z5ju7SEuQ8oYY37961+bwYMHm969e5u77rrLuZ3bRpLaPV5++WVnT2trq/nJT35i/H6/8Xg85hvf+IY5dOhQ7Iq+CpeGVLz08Pvf/97k5OQYj8djbrnlFrNu3bqI87b3EQ6HzcKFC01mZqbp06ePGTp0qFm+fHnEP4S29bB79+52/w7Mnj37quttbGw08+fPN2lpaSYpKclMnjzZHD9+3Jo+ampqOvy7vnv3bmv6uNLP4lLthVR39sD3SQEArBV370kBAG4chBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFqEFADAWoQUAMBahBQAwFr/B1/R+huKR+fGAAAAAElFTkSuQmCC\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(seg[t][z] == id_main, interpolation=\"none\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'nb_ids': array([1], dtype=uint16), 'contact': array([789])}\n",
      "Ratio: 0.8436272878535773\n",
      "Merged ratio: 0.8007643756146751\n",
      "Relative change: -0.05080787790536914\n",
      "\n"
     ]
    }
   ],
   "source": [
    "merged = handle_small_segments_simple(seg[t], nbhd_size=1)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "outputs": [
    {
     "data": {
      "text/plain": "<matplotlib.image.AxesImage at 0x7f1b4c1f0940>"
     },
     "execution_count": 144,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "plt.imshow(merged[z], interpolation=\"none\")"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [],
   "source": [],
   "metadata": {
    "collapsed": false
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "name": "conda-env-kahrp-py",
   "language": "python",
   "display_name": "Python [conda env:kahrp]"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}