cotton_regression.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "import scipy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gen_density_map_gaussian(im, points, sigma=4):\n",
    "    \"\"\"\n",
    "    func: generate the density map\n",
    "    \"\"\"\n",
    "    density_map = np.zeros(im.shape[:2], dtype=np.float32)\n",
    "    h, w = density_map.shape[:2]\n",
    "    num_gt = np.squeeze(points).shape[0]\n",
    "    if num_gt == 0:\n",
    "        return density_map\n",
    "    if sigma == 4:\n",
    "        # Adaptive sigma in CSRNet.\n",
    "        leafsize = 2048\n",
    "        tree = scipy.spatial.KDTree(points.copy(), leafsize=leafsize)\n",
    "        distances, _ = tree.query(points, k=4)\n",
    "    for idx_p, p in enumerate(points):\n",
    "        p = np.round(p).astype(int)\n",
    "        p[0], p[1] = min(h-1, p[1]), min(w-1, p[0])\n",
    "        gaussian_radius = sigma * 2 - 1\n",
    "        if sigma == 4:\n",
    "            # Adaptive sigma in CSRNet.\n",
    "            sigma = max(int(np.sum(distances[idx_p][1:4]) * 0.1), 1)\n",
    "            gaussian_radius = sigma * 3\n",
    "        gaussian_map = np.multiply(\n",
    "            cv2.getGaussianKernel(int(gaussian_radius*2+1), sigma),\n",
    "            cv2.getGaussianKernel(int(gaussian_radius*2+1), sigma).T\n",
    "        )\n",
    "        x_left, x_right, y_up, y_down = 0, gaussian_map.shape[1], 0, gaussian_map.shape[0]\n",
    "        # cut the gaussian kernel\n",
    "        if p[1] < gaussian_radius:\n",
    "            x_left = gaussian_radius - p[1]\n",
    "        if p[0] < gaussian_radius:\n",
    "            y_up = gaussian_radius - p[0]\n",
    "        if p[1] + gaussian_radius >= w:\n",
    "            x_right = gaussian_map.shape[1] - (gaussian_radius + p[1] - w) - 1\n",
    "        if p[0] + gaussian_radius >= h:\n",
    "            y_down = gaussian_map.shape[0] - (gaussian_radius + p[0] - h) - 1\n",
    "        gaussian_map = gaussian_map[y_up:y_down, x_left:x_right]\n",
    "        if np.sum(gaussian_map):\n",
    "            gaussian_map = gaussian_map / np.sum(gaussian_map)\n",
    "        print(p)\n",
    "        density_map[\n",
    "            max(0, p[0]-gaussian_radius):min(h, p[0]+gaussian_radius+1),\n",
    "            max(0, p[1]-gaussian_radius):min(w, p[1]+gaussian_radius+1)\n",
    "        ] += gaussian_map\n",
    "    density_map = density_map / (np.sum(density_map / num_gt))\n",
    "    return density_map"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "import cv2\n",
    "import numpy as np\n",
    "from matplotlib import pyplot as plt\n",
    "import pandas as pd\n",
    "import json\n",
    "from os import path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "ann = pd.read_csv(\"/mnt/d/personal/cotton/samples/via_export_csv.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "base_path = \"/mnt/d/personal/cotton/samples/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 310,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2019_1010_123336_234.JPG\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/suba/.virtualenvs/cotton/lib/python3.6/site-packages/pandas/core/indexing.py:845: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  self.obj[key] = _infer_fill_value(value)\n",
      "/home/suba/.virtualenvs/cotton/lib/python3.6/site-packages/pandas/core/indexing.py:966: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  self.obj[item] = s\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2019_1010_123338_235.JPG\n",
      "2019_1010_123339_236.JPG\n",
      "2019_1010_123341_237.JPG\n",
      "2019_1010_123342_238.JPG\n",
      "2019_1010_123344_239.JPG\n",
      "2019_1010_123345_240.JPG\n",
      "2019_1010_123347_241.JPG\n",
      "2019_1010_123348_242.JPG\n",
      "2019_1010_123350_243.JPG\n",
      "2019_1010_123351_244.JPG\n",
      "2019_1010_123353_245.JPG\n"
     ]
    },
    {
     "ename": "KeyError",
     "evalue": "'y'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-310-d426a63b4323>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m     17\u001b[0m     \u001b[0;32mfor\u001b[0m \u001b[0mix\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mval\u001b[0m \u001b[0;32min\u001b[0m \u001b[0msample_ann\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0miterrows\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     18\u001b[0m         \u001b[0mcoords\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mjson\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mloads\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mval\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"region_shape_attributes\"\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 19\u001b[0;31m         \u001b[0mcenter\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcoords\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"y\"\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mcoords\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"height\"\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m/\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcoords\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"x\"\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mcoords\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m\"width\"\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m/\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     20\u001b[0m         \u001b[0msample_ann\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mloc\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mix\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"centery\"\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcenter\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     21\u001b[0m         \u001b[0msample_ann\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mloc\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mix\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m\"centerx\"\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcenter\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyError\u001b[0m: 'y'"
     ]
    }
   ],
   "source": [
    "create_images = True\n",
    "data = []\n",
    "pos_cnt = 1\n",
    "neg_cnt = 1\n",
    "all_data = pd.DataFrame()\n",
    "X = []\n",
    "Y = []\n",
    "for file in ann[\"filename\"].unique():\n",
    "    print(file)\n",
    "    sample_ann = ann[ann[\"filename\"] == file]\n",
    "    # file_name = \"2019_1010_123336_234.JPG\"\n",
    "    # sample_ann = ann[ann[\"filename\"]==file_name]\n",
    "    img = cv2.imread(path.join(base_path, file_name))\n",
    "    points = np.zeros((img.shape[0], img.shape[1]))\n",
    "\n",
    "    # Read all the annotations\n",
    "    for ix, val in sample_ann.iterrows():\n",
    "        coords = json.loads(val[\"region_shape_attributes\"])\n",
    "        center = [int(coords[\"y\"] + (coords[\"height\"] / 2)), int(coords[\"x\"] + (coords[\"width\"] / 2))]\n",
    "        sample_ann.loc[ix, \"centery\"] = center[0]\n",
    "        sample_ann.loc[ix, \"centerx\"] = center[1]\n",
    "        points[center[0]][center[1]] = 1\n",
    "    imgs = []\n",
    "    pts = []\n",
    "    for _x in range(100):\n",
    "        for _y in range(100):\n",
    "            xstart = _x * 30\n",
    "            xend = (_x + 1) * 30\n",
    "            ystart = _y * 40\n",
    "            yend = (_y + 1) * 40\n",
    "            # X.append(img[xstart:xend, ystart:yend])\n",
    "            points_sc = points[xstart:xend, ystart:yend]\n",
    "            pts = []\n",
    "            for ii, jj in zip(np.nonzero(points_sc)[1], np.nonzero(points_sc)[0]):\n",
    "                pts.append([ii, jj])\n",
    "            if len(pts) > 0:\n",
    "                X.append(img[xstart:xend, ystart:yend] * 1/255.0)\n",
    "                Y.append(gen_density_map_gaussian(X[-1], pts, 3))    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 311,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "804 804\n"
     ]
    }
   ],
   "source": [
    "print(len(X), len(Y))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 243,
   "metadata": {},
   "outputs": [
    {
     "ename": "AttributeError",
     "evalue": "'list' object has no attribute 'shape'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-243-bc5f1a0adac8>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mX\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mAttributeError\u001b[0m: 'list' object has no attribute 'shape'"
     ]
    }
   ],
   "source": [
    "X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 211,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([], dtype=int64), array([], dtype=int64))"
      ]
     },
     "execution_count": 211,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.nonzero(points)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "img_sm = img[2500:2700, 3600:3800]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "points_sm = points[2500:2700, 3600:3800]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gen_density_map_gaussian(im, points, sigma=4):\n",
    "    \"\"\"\n",
    "    func: generate the density map\n",
    "    \"\"\"\n",
    "    density_map = np.zeros(im.shape[:2], dtype=np.float32)\n",
    "    h, w = density_map.shape[:2]\n",
    "    num_gt = np.squeeze(points).shape[0]\n",
    "    if num_gt == 0:\n",
    "        return density_map\n",
    "    if sigma == 4:\n",
    "        # Adaptive sigma in CSRNet.\n",
    "        leafsize = 2048\n",
    "        tree = scipy.spatial.KDTree(points.copy(), leafsize=leafsize)\n",
    "        distances, _ = tree.query(points, k=4)\n",
    "    for idx_p, p in enumerate(points):\n",
    "        p = np.round(p).astype(int)\n",
    "        p[0], p[1] = min(h-1, p[1]), min(w-1, p[0])\n",
    "        gaussian_radius = sigma * 2 - 1\n",
    "        if sigma == 4:\n",
    "            # Adaptive sigma in CSRNet.\n",
    "            sigma = max(int(np.sum(distances[idx_p][1:4]) * 0.1), 1)\n",
    "            gaussian_radius = sigma * 3\n",
    "        gaussian_map = np.multiply(\n",
    "            cv2.getGaussianKernel(int(gaussian_radius*2+1), sigma),\n",
    "            cv2.getGaussianKernel(int(gaussian_radius*2+1), sigma).T\n",
    "        )\n",
    "        x_left, x_right, y_up, y_down = 0, gaussian_map.shape[1], 0, gaussian_map.shape[0]\n",
    "        # cut the gaussian kernel\n",
    "        if p[1] < gaussian_radius:\n",
    "            x_left = gaussian_radius - p[1]\n",
    "        if p[0] < gaussian_radius:\n",
    "            y_up = gaussian_radius - p[0]\n",
    "        if p[1] + gaussian_radius >= w:\n",
    "            x_right = gaussian_map.shape[1] - (gaussian_radius + p[1] - w) - 1\n",
    "        if p[0] + gaussian_radius >= h:\n",
    "            y_down = gaussian_map.shape[0] - (gaussian_radius + p[0] - h) - 1\n",
    "        gaussian_map = gaussian_map[y_up:y_down, x_left:x_right]\n",
    "        if np.sum(gaussian_map):\n",
    "            gaussian_map = gaussian_map / np.sum(gaussian_map)\n",
    "        density_map[\n",
    "            max(0, p[0]-gaussian_radius):min(h, p[0]+gaussian_radius+1),\n",
    "            max(0, p[1]-gaussian_radius):min(w, p[1]+gaussian_radius+1)\n",
    "        ] += gaussian_map\n",
    "    density_map = density_map / (np.sum(density_map / num_gt))\n",
    "    return density_map"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[61 91]\n"
     ]
    }
   ],
   "source": [
    "img_gaus = gen_density_map_gaussian(img_sm, [[91, 61]], 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "import scipy.spatial"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 256,
   "metadata": {},
   "outputs": [],
   "source": [
    "for ix in range(50):\n",
    "    cv2.imwrite(path.join(base_path, \"regression\", str(ix) + \".jpg\"), X[ix])\n",
    "    temp = X[ix].copy()\n",
    "    temp[:, :, 2] =  temp[:, :, 2] + Y[ix]\n",
    "    cv2.imwrite(path.join(base_path, \"regression\", str(ix) + \"ann.jpg\"), temp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [],
   "source": [
    "img_sm_cp = img_sm.copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [],
   "source": [
    "img_sm_cp[:, :, 0] = img_sm_cp[:,:, 0] +img_gaus * 25555\n",
    "# img_sm_cp[:, :, 1] = img_sm_cp[:,:, 1] +img_gaus * 25555\n",
    "# img_sm_cp[:, :, 2] = img_sm_cp[:,:, 2] +img_gaus * 25555"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 83,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cv2.imwrite(path.join(base_path, \"zgauss.jpg\"), img_sm_cp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([ 0,  0,  0, ..., 39, 39, 39]), array([ 0,  1,  2, ..., 37, 38, 39]))"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.nonzero(img_gaus)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.33711848, 0.3366986 , 0.33543748, ..., 0.        , 0.        ,\n",
       "        0.        ],\n",
       "       [0.3366986 , 0.3362775 , 0.33501938, ..., 0.        , 0.        ,\n",
       "        0.        ],\n",
       "       [0.33543748, 0.33501938, 0.33376437, ..., 0.        , 0.        ,\n",
       "        0.        ],\n",
       "       ...,\n",
       "       [0.        , 0.        , 0.        , ..., 0.        , 0.        ,\n",
       "        0.        ],\n",
       "       [0.        , 0.        , 0.        , ..., 0.        , 0.        ,\n",
       "        0.        ],\n",
       "       [0.        , 0.        , 0.        , ..., 0.        , 0.        ,\n",
       "        0.        ]], dtype=float32)"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img_gaus"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([91]), array([61]))"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.nonzero(points_sm)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "200"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.squeeze(points_sm).shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "       0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.])"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "points_sm[91]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 366,
   "metadata": {},
   "outputs": [],
   "source": [
    "shutil.move?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(3000, 4000, 3)"
      ]
     },
     "execution_count": 84,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(3000, 4000)"
      ]
     },
     "execution_count": 85,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "points.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([2591]), array([3661]))"
      ]
     },
     "execution_count": 86,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.nonzero(points)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [],
   "source": [
    "imgs = []\n",
    "points_sc = []\n",
    "for _x in range(10):\n",
    "    for _y in range(10):\n",
    "        xstart = _x * 300\n",
    "        xend = (_x + 1) * 300\n",
    "        ystart = _y * 400\n",
    "        yend = (_y + 1) * 400\n",
    "        imgs.append(img[xstart:xend, ystart:yend])\n",
    "        points_sc.append(points[xstart:xend, ystart:yend])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [],
   "source": [
    "pts = []\n",
    "for i in points_sc:\n",
    "    for ii, jj in zip(np.nonzero(i)[0], np.nonzero(i)[1]):\n",
    "        pts.append([ii, jj])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.layers import Dense, Conv2D, Input, MaxPooling2D, concatenate, add, UpSampling2D\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.initializers import RandomNormal"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 336,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_flow = Input(shape=(30, 40, 3))\n",
    "dilated_conv_kernel_initializer = RandomNormal(stddev=0.01)\n",
    "\n",
    "# front-end\n",
    "x = Conv2D(64, (3, 3), strides=(1, 1), padding='same', activation='relu')(input_flow)\n",
    "x = Conv2D(64, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "x = MaxPooling2D(pool_size=(2, 2))(x)\n",
    "\n",
    "x = Conv2D(128, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "x = Conv2D(128, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "# x = MaxPooling2D(pool_size=(2, 2))(x)\n",
    "\n",
    "x = Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "x = Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "# x = Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "# x = MaxPooling2D(pool_size=(2, 2))(x)\n",
    "\n",
    "# x = Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "# x = Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "# x = Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "\n",
    "# x = UpSampling2D((2,2))(x)\n",
    "# x = Conv2D(128, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "x = UpSampling2D((2,2))(x)\n",
    "x = Conv2D(128, (3, 3), strides=(1, 1), padding='same', activation='relu')(x)\n",
    "\n",
    "# # back-end\n",
    "# x = Conv2D(512, (3, 3), strides=(1, 1), padding='same', dilation_rate=2, activation='relu', kernel_initializer=dilated_conv_kernel_initializer)(x)\n",
    "# x = Conv2D(512, (3, 3), strides=(1, 1), padding='same', dilation_rate=2, activation='relu', kernel_initializer=dilated_conv_kernel_initializer)(x)\n",
    "# x = Conv2D(512, (3, 3), strides=(1, 1), padding='same', dilation_rate=2, activation='relu', kernel_initializer=dilated_conv_kernel_initializer)(x)\n",
    "# x = Conv2D(256, (3, 3), strides=(1, 1), padding='same', dilation_rate=2, activation='relu', kernel_initializer=dilated_conv_kernel_initializer)(x)\n",
    "# x = Conv2D(128, (3, 3), strides=(1, 1), padding='same', dilation_rate=2, activation='relu', kernel_initializer=dilated_conv_kernel_initializer)(x)\n",
    "# x = Conv2D(64, (3, 3), strides=(1, 1), padding='same', dilation_rate=2, activation='relu', kernel_initializer=dilated_conv_kernel_initializer)(x)\n",
    "\n",
    "output_flow = Conv2D(1, 1, strides=(1, 1), padding='same', activation='relu', kernel_initializer=dilated_conv_kernel_initializer)(x)\n",
    "model = Model(inputs=input_flow, outputs=output_flow)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 338,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_17\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_19 (InputLayer)        [(None, 30, 40, 3)]       0         \n",
      "_________________________________________________________________\n",
      "conv2d_150 (Conv2D)          (None, 30, 40, 64)        1792      \n",
      "_________________________________________________________________\n",
      "conv2d_151 (Conv2D)          (None, 30, 40, 64)        36928     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_43 (MaxPooling (None, 15, 20, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_152 (Conv2D)          (None, 15, 20, 128)       73856     \n",
      "_________________________________________________________________\n",
      "conv2d_153 (Conv2D)          (None, 15, 20, 128)       147584    \n",
      "_________________________________________________________________\n",
      "conv2d_154 (Conv2D)          (None, 15, 20, 256)       295168    \n",
      "_________________________________________________________________\n",
      "conv2d_155 (Conv2D)          (None, 15, 20, 256)       590080    \n",
      "_________________________________________________________________\n",
      "up_sampling2d_11 (UpSampling (None, 30, 40, 256)       0         \n",
      "_________________________________________________________________\n",
      "conv2d_156 (Conv2D)          (None, 30, 40, 128)       295040    \n",
      "_________________________________________________________________\n",
      "conv2d_157 (Conv2D)          (None, 30, 40, 1)         129       \n",
      "=================================================================\n",
      "Total params: 1,440,577\n",
      "Trainable params: 1,440,577\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 240,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "30.183594"
      ]
     },
     "execution_count": 240,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.sum(YY[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 339,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(optimizer=\"adam\", loss='MSE', metrics=[\"mae\", \"mse\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 340,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 643 samples, validate on 161 samples\n",
      "Epoch 1/10\n",
      "643/643 [==============================] - 20s 30ms/sample - loss: 1.7293 - mae: 0.5014 - mse: 1.7293 - val_loss: 1.7233 - val_mae: 0.5718 - val_mse: 1.7233\n",
      "Epoch 2/10\n",
      "643/643 [==============================] - 19s 30ms/sample - loss: 1.6919 - mae: 0.5881 - mse: 1.6919 - val_loss: 1.7170 - val_mae: 0.5889 - val_mse: 1.7170\n",
      "Epoch 3/10\n",
      "643/643 [==============================] - 20s 31ms/sample - loss: 1.6884 - mae: 0.5999 - mse: 1.6884 - val_loss: 1.7150 - val_mae: 0.5994 - val_mse: 1.7150\n",
      "Epoch 4/10\n",
      "643/643 [==============================] - 20s 31ms/sample - loss: 1.6874 - mae: 0.6017 - mse: 1.6874 - val_loss: 1.7165 - val_mae: 0.6235 - val_mse: 1.7165\n",
      "Epoch 5/10\n",
      "643/643 [==============================] - 20s 31ms/sample - loss: 1.6865 - mae: 0.6044 - mse: 1.6865 - val_loss: 1.7165 - val_mae: 0.5955 - val_mse: 1.7165\n",
      "Epoch 6/10\n",
      "643/643 [==============================] - 21s 32ms/sample - loss: 1.6851 - mae: 0.6055 - mse: 1.6851 - val_loss: 1.7172 - val_mae: 0.5997 - val_mse: 1.7172\n",
      "Epoch 7/10\n",
      "643/643 [==============================] - 23s 35ms/sample - loss: 1.6851 - mae: 0.6018 - mse: 1.6851 - val_loss: 1.7192 - val_mae: 0.6296 - val_mse: 1.7192\n",
      "Epoch 8/10\n",
      "643/643 [==============================] - 23s 36ms/sample - loss: 1.6829 - mae: 0.6060 - mse: 1.6829 - val_loss: 1.7208 - val_mae: 0.5821 - val_mse: 1.7208\n",
      "Epoch 9/10\n",
      "643/643 [==============================] - 20s 31ms/sample - loss: 1.6866 - mae: 0.5978 - mse: 1.6866 - val_loss: 1.7204 - val_mae: 0.6294 - val_mse: 1.7204\n",
      "Epoch 10/10\n",
      "643/643 [==============================] - 20s 31ms/sample - loss: 1.6821 - mae: 0.6051 - mse: 1.6821 - val_loss: 1.7251 - val_mae: 0.6374 - val_mse: 1.7251\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7f1528297128>"
      ]
     },
     "execution_count": 340,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X, Y, epochs=10, validation_split=0.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 273,
   "metadata": {},
   "outputs": [],
   "source": [
    "def show(img):\n",
    "    plt.imshow(img[:, :, ::-1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 268,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_img = cv2.imread(path.join(base_path, \"regression\", \"24.jpg\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 358,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show(X[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 359,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUQAAAD5CAYAAABfyUzZAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nO3da4ykZ5Uf8P9537p29b177jP22MaweE12IIOzCLJik+zKsVYxRCsEH1b+gDKraJEWaaPEIdJCPkRiowDaDxGRCQ5OQljIAsIf0GaJg4LyBTDG9/tljKc9Mz3T90vd6+RDP4OGSf9Ptbtnqtqb/08aTU+drnqfeuqtM9VVp88xd4eIiADZsBcgIrJfKCGKiCRKiCIiiRKiiEiihCgikighiogkhb1c2czuBvDnAHIA/9HdPx99f5aZ59n2OdgRlf9YvI54mW8ffe5IVCJlGY/lhZzGiqX4FCgWg7jtrmSr5734G4L7GV032p8853sAAOy8BIBOp82vaPxBsz4PaJbt/rr8Nvvcz5w/nha8Pmq1WjTWbPIYAHS7XRqL1lsIzlsL9g4ALHhclhc2L7v7gW2PGd5qfMAcwL8H8DsAzgH4qZk97O7PsuvkWYbJ8cq2sbgeMn4hG935XfPdb7gFy3XnJ4cHSQ0Aes6fmFmJ3+7MgVEaO3LTofCYh4/M8GCRHxPGY/XGRnjMbpffz3pzk8babX69qamp8JgjIyM0tnD5Io0VCvwpVOiTnMrlMo2VgsTVC/4/GR0ZD485MT5LY8VClcbOvjZHY6++ejY85sryGo3Vxvi5OTPDz71KNU5d0ePynYcefZ3F9vIj810AXnb3V929BeAvANy7h9sTERmqvSTEYwDeuOrf59JlIiJvS3t6D3EnzOwMgDNA/J6JiMiw7eUV4hyAE1f9+3i67Fe4+wPuftrdT2c34r0+EZHrZC8J8acAbjezW8ysBODjAB6+PssSERm8Xf/I7O4dM/sUgP+BrbKbB939mf7XZDmYf8K6l448u/0Eei+fXOcWfLoYfATtFpejNNtNft2cr7fb4sdcXlwNj5kF5R+lKr/dUpnHsnJ82lUr21ciAEAWlMc0M74/Fj0miB/viQn+CXWr1eA32ovP20pwPwvG96jR4Pfz8qXl8Jgvv8g/LZ4/f5nGLl6o01idf/APACgWeaxU4RUH8+f5fbE+JV9RGVpkT+8huvv3AXx/L7chIrJf6DdVREQSJUQRkUQJUUQkUUIUEUmUEEVEEiVEEZHkhv/q3k5FtYZ7mgwYdK2Jas/idmRA9Fs3UeeUXrtDY60ury8DgAxBqyQP6v66QWurzfh+bmRB/RkPoRDUno1Pxh1ZEHTu6QT3pdfmsc21eG/zoO5vfHyCxtaD86vTCtqGAWi3eN1pq8PX22nx/VldiTsJvfLiGzQ2z5v6oBc0NqqWwkMiKq9t13kwD573/X4NuBu1bAvoFaKISKKEKCKSKCGKiCRKiCIiiRKiiEiihCgikgy+7Ia1RIpaJfUru4k+gd/lZLh+7b+ieB4syIPrZUHpTL94j1fzoNMMptEV4/vZ3ORlEbWclxc1G7wmZ7G9Hh7TwUsmus7vaDSlrViJ23+NVPg+LC3y9UatuFpBDAC8y283z4IyoJExfj3wwVVbcb4P41VeW1MIhl6NjsZlVL3gud2o84l90eTBYvBYA0CxxAdmXcAijekVoohIooQoIpIoIYqIJEqIIiKJEqKISKKEKCKSDLTsxn13nWv2MgEvEq2lEEzHA+JpdGtra7s6Zi9qKQIgCzqyRN15Oh0eK3Xi+1lf42URS8HEvrFxXpJT7sXH7Ab7YAV+3UqRHxPteG8v/IJPnGsFpT4jI3xyXtgiBkB9g4+rq9V4aU2nGHX1CaYAAvBgSdHTLCr5yoKOP0BcQtTJ+N42m7xsqduO80heDc6FgF4hiogkSogiIokSoohIooQoIpIoIYqIJEqIIiLJnspuzOwsgDUAXQAddz8df//uSmiyPcyYimZF7aWcJxxQdYMGZuU57/DR8WBgUTOY8mNxmUa9wYcW9YJOQu12MORnKd6D6gifWpQV+L4vLyzz6/X5r7/Z4eVFtTFeWjM7c4BfrxqU5ABYzpZorFzm1+22+P4tLPDbBICNdV530+VbgCwY/NUNBqcBgAe1Pr0Oj+VR6ZvFJU29XQ6Zuh51iL/t7ryIS0TkbUI/MouIJHtNiA7gr83sZ2Z25nosSERkWPb6I/OH3H3OzA4C+IGZPe/uP7r6G1KiPAPEw91FRIZtT68Q3X0u/T0P4LsA7trmex5w99PufjrLlBBFZP/adUI0s5qZjV35GsDvAnj6ei1MRGTQ9vIj8yEA303lJwUA/83d/+q6rEpEZAh2nRDd/VUAv/GWD2jb19JFLZ/61e5l/YrMdnG9frcZTVQrBrVyI4UijU1MTIfHjPYhapW0uMinjHW7fD0A0A2mopUq/Lr1dT51r1CNj9la422xmsF4wegx61f9OVLjtY/VcT5Vbv0Sn5zXG42n7h2cnaUxC2o8X3j+ZRqbn+d1owBQDIbV9YJYJ6gXrNf5Yw3Ej4uRfAAAHrQV61dmWC5F0wf5+aWyGxGRRAlRRCRRQhQRSZQQRUQSJUQRkUQJUUQkGejUvS3blxOEE7/6lMBEbbF6Pd76qtuN2hYFNQgAKhX+sX5e4CUTo7VRGjty4lB4zNFRft25uTkaa/V4i6+FhZXwmLUav5+FIi+fqVR4+6pGN66ZaLWCUp8iX0/Uks3y+LekajU+pW1ikk/AW1nljZ6awbRDIG5z1gUvc+kE7bSC7QEArAbdwQrBKZ8HHeTyPG7FVQ6en2GpVHA/u8HzGgB6u2yzp1eIIiKJEqKISKKEKCKSKCGKiCRKiCIiiRKiiEgy8LIb1rEl7oixh+l4wcg+iz6atz8Nb7dQ/ASNdbvHaCwqBsjiSh8c5TeLSpXHVlZ5bCQeuoc8f4nGev5dGitkf0ZjWZ+SiDxqJBzEikEZULUWT8CrVPjmh2VApeApVIpLQ+ot3iVmdZ2XQ61vrtHYxHR8PyfG+Xq9x2Obq7xDTKcT308POve48esGjXBQ8Dh17TZn6BWiiEiihCgikighiogkSogiIokSoohIooQoIpIMtOzGzGhpRKHAl9Ltxd1R2m0ez8MuJ8EQm84/Co8ZldYcO8Gvd1MQqwalMwAwymcdYWqGxxaCDifNPsN6lpdup7Esv4fGet3P8lgWD5nK82B4VYvXCVUqvGPNoQMHw2M6eGnN4aN848vBY3bh0pvhMTc3+YAqD8rFStVgIFa1Fh5zfGSKxvKMl+ysBl2RVlZ4GRAQD2TrdnnZTbQH5aBMCgBK5d2lNr1CFBFJlBBFRBIlRBGRRAlRRCRRQhQRSZQQRUQSJUQRkaRvsY6ZPQjg9wDMu/ud6bJpAN8EcBLAWQAfc/eg2u2Xt7WrOsRen/ZCvaAOMWq41erw67Xb7wyPOTHJYx/4EI/92rt4LGrTBQB13i0KE0GN4gc+yGOXFuNjbgbtwRqNO2ksC2rIokmIW4Lpeb2g/VfQP21sbCI8YqHIJ7xNTfGpeyOj/Jj1Dq8zBIBGg5/znaAt1kbwoGw2NsJjWvCUHxvl9Z95id/PvBS32sra/L60OnzyZZ7ztY70Kdotl3mtZmQnrxC/BuDuay67H8Aj7n47gEfSv0VE3tb6JkR3/xGAa19H3AvgofT1QwA+cp3XJSIycLv91b1D7n4+fX0BAJ2wbmZnAJwBgDzXW5Yisn/tOUP51kwA+oaRuz/g7qfd/XQejAkQERm23Waoi2Z2BADS3/PXb0kiIsOx24T4MID70tf3Afje9VmOiMjw7KTs5hsAPgxg1szOAfgsgM8D+JaZfRLA6wA+tqOjuaPb3b68gU3jA4Buj5dE9NPr8Y/163VedlMux7d78hYeu+lmHovKWJZ4hyUAcVnOJu+whNFRHrvpZHzMuTkei2bnlSv81Kr3mfQXnQvVoNwiKt1qNPjUOAAYr/AHvBOUdS0u8get0w4eFAC1cf7AWPDMbARTABt1fr4DQOa8fGZ1fZnf7gav+VrfiOvFPCi7yQv8NVmtyh+TibG47CbP+zyBib4J0d3ZvM2/v6sjiojsU/qUQ0QkUUIUEUmUEEVEEiVEEZFECVFEJBno1D13PmUrqLToW3YTdU+xoOtK9IszU3w4GQDg3e/msWLQaOOFF3hsM+hmAwBNXm2BqCFQ0DQEt9waH/PZZ3isHVR4WMYfM+vzG0vmvHtKqcQ3NyrXWVnlJSUA0OnxTTp3gbcEmprlU+6Onogn/c0cnqWxjQafZLdZ5+U8hVL8XLEuL7tZWQ668wRD7vJi/HhGpTXF4OSsBtMFx8b5hEUAyLMb1+1GROT/C0qIIiKJEqKISKKEKCKSKCGKiCRKiCIiyUDLbswMeb795/fscgDoBoNogLjsppDzEo5qlQ/VmZ4OD4nDx3js4kUeWwuargQNVwAAzWCWViuITc3w2NJCfMxjx3ksGnrVbe++Q1EWlOV0Orz2KMv4edBsRr15gHaHr3dknK/HjJ9f9WiDAMzP8zairS5vCRSd7/0GeHmH70M5aPFULvDYeC0ugcl6/JiNoNYsC+5LHjyvAWB0JO6GQ4+5q2uJiPwNpIQoIpIoIYqIJEqIIiKJEqKISKKEKCKSKCGKiCQDrUOMtIJJYt6N83bmQXuhjNcrFSr8Nm+7LTxk2DpsM6g1jIa/lYP1APGEvKhUc5V3kgLpxvZLJ27isaiG8eJ5XhjZbARFkwAqZX5npmd5y6yDBw/QWG0irkubnOJtvPIyb7e1tHSJxjY2NsJjXr7M24qtrPFJdu1WMKEyLrdE9JSPphYWgjGAUUs2ABgJJiWOj47R2EYwZnJ9Yyk85uzMRBhn9ApRRCRRQhQRSZQQRUQSJUQRkUQJUUQkUUIUEUn6lt2Y2YMAfg/AvLvfmS77HIB/AuBKzcFn3P37/W6r5z00m9uXMEQtnxB3+glbHnWD1kMjJd7SqMg7gwEA2kHlSCMogSnx6g4U+5TdLASVBlH1TIlXjaAQTFMDgI1g0t9msAed4DGbmI37nL3nzlM89r730FjUEmplPe5zNjbO17S8zktrOjbJ1xPVXwFYCB7QV1/htVJBhRqqfc6hoMseKqWgtCbnpTXVavx49op8wRa0bDPnLdn6tfeK2rJFdvIK8WsA7t7m8i+5+6n0p28yFBHZ7/omRHf/EQBeQSoi8jfEXt5D/JSZPWlmD5pZn7HuIiL7324T4pcB3AbgFIDzAL7AvtHMzpjZo2b2aC94P09EZNh2lRDd/aK7d929B+ArAO4KvvcBdz/t7qez4PeKRUSGbVcJ0cyOXPXPjwJ4+vosR0RkeHZSdvMNAB8GMGtm5wB8FsCHzewUAAdwFsAf7uho7uh2t69JsaCbRp7vvilPVJITTRnr96n9+jqPjfNKjLAk5+y5+JgLQeVIdZTHpoL1jAXXA4BGMDwvihWCqohjJw+Fx7z912+hsfFZPuHtjTdep7G5CzwGAKWlaNIfn4AXnUO10XhzR0Z5DValxEtygmoUWDygEs2oAU8pqhfj+1NGXKO2UecHbbZ4aVIhuNliHr+Wazf5Yxbpm2nc/RPbXPzVXR1NRGQf02+qiIgkSogiIokSoohIooQoIpIoIYqIJEqIIiLJQKfumRlK5bd+yH41gYUiz+u9Hu991W7ztkSLi0GxF4BOj/dRqgYDv94Iag2fezk8JIJuZbiZl+ehGrQc6/SZuhfFs1JQjx9cr1SJH9B6a4XGXj/HY5cuX6SxVi9uxdXY4DV4xQp/rCtBwWWpTy+uQwcO0ljzVr6elct8D9ZW4/q7VjCWrxrUVJaD3nSlqGAQgGX8mMUCPxeyvM/JGWBtBvvRK0QRkUQJUUQkUUIUEUmUEEVEEiVEEZFECVFEJBlo2U2WZRgd2b4GpNHg5QL9ym6qZT4RrB7cbmOzTmNv/OJL4TGL8/+Yxo7edCuNjY/z27zzjvCQqAXdpMaCIQ61oCSnz2A4VCp8j44f/9/8djd47VGWxyVN85fm+HXL/GToOR8DODYWbAKAVpufJ+3gdltdXt7hffbWggl4h4/wkpxSxp+2rcZ8eMxeM5iA58EEPOOlM/1eVbHnPADMHpqhsUqV3896I+i/BwC9+Bxj9ApRRCRRQhQRSZQQRUQSJUQRkUQJUUQkUUIUEUkG3u0mz7cvm8iC1JxH9QkAikVeduPOy0ZKQZecdvMz4TG7vX9OY2srvD7m9iO30didNx8Njxlpdvn9bLV45xTr01EkK/AyDbc1Gjt4+CSNra1Fo9+AV86+RmOdoHuRB+PojhyP9/boscM01u7yPajX+XqWFhbDY0adct71rnfT2N8+9V4ae+rJZ8NjPvXzZ2hscTGoE+rwvS32gnZKAJDz+1kd4R12Jid5jZqtxOft0hKfWhjRK0QRkUQJUUQkUUIUEUmUEEVEEiVEEZFECVFEJOlbdmNmJwD8ZwCHADiAB9z9z81sGsA3AZwEcBbAx9w9/KzbvYdOe/syBQumEkXlFGmRNFQJSnJmpvjH+qUK7+4BAHmZr7cYNFapN3lXlfXNuBwlz3n50fLGKo2trfLbrfcZxtPt8L09fOgYjU1OztLYm2++GR7z1ddep7GlVd55psOrizA2HbeeKRR5+cdm0BKIVJEBAGo1Xm4CAO02P69X1xZobGJijMZO3BKXF/WC59LZYN/XVnh3GQ+GSAFAs8fLluYvn6exzQYv62oGzyMAWF3lz4fITl4hdgD8ibvfAeA3AfyRmd0B4H4Aj7j77QAeSf8WEXnb6psQ3f28uz+Wvl4D8ByAYwDuBfBQ+raHAHzkRi1SRGQQ3tJ7iGZ2EsB7AfwYwCF3v/J69wK2fqQWEXnb2nFCNLNRAN8G8Gl3/5Uf0N3dsfX+4nbXO2Nmj5rZo91e/F6DiMgw7SghmlkRW8nw6+7+nXTxRTM7kuJHAGzbu9zdH3D30+5+Os/6zAIQERmivgnRzAzAVwE85+5fvCr0MID70tf3Afje9V+eiMjg7KTbzQcB/AGAp8zs8XTZZwB8HsC3zOyTAF4H8LEbs0QRkcGwrbf/BqNczP34VJ9WQdtwj3/UjlqHWcaL02659TiNjY4Xw2OWRviaarO8pq3tvO6vVOU1k0Bch/jsS8/T2NISr+eyYry3J0++g8YmJ/jEtOkpPjVuYyOutzx79hdhfDemZ/kUQACoVvjer60v01g7qIerVCrhMaPHE0Hpba3G6xDLZX7uAUCnzZ/vufHXR08/wduKra3x8wsASgW+t+1gCmChwNdTjJ70ADY2eO3ok4/Xf+bup7eL6TdVREQSJUQRkUQJUUQkUUIUEUmUEEVEEiVEEZFkoFP30HP0Gtt/zN4M2lCRq/zSOO/ihZlZXubTa63QWNHiMo1izstysh6vmVhc4pPYmsF0PADYbPE9Wt3ksbzE21CNTsX3EwV+P9cavHymvcTbOhX6TFE8epz/WvxYjU80rFb5Y92vdKsXTJWrBPvXa/PHrGfxZLioRVXUcmyzwVugFbN4b6NSoJEK37+DR3g7t83NuNVW0J0PeZGvNwvSUzMo1wGAXnd3qU2vEEVEEiVEEZFECVFEJFFCFBFJlBBFRBIlRBGRZKBlN4VCjtmZyW1jUfOKtXVeHgMAU8H0vJmDvDNIlvOyiM06nzIGAMtBmYuv8lKCxTXeOaXjcZnGxPQUjc3M8JKJ0XG+BzOHD4THrNfrNDZ//gKNra/z/ZuZ4l1yAODY0RM0trDABzu+OfcCjXWikXwAqmVeWnPiJj7J7sRR3jHp0uKl8JjLS3z6oGW8DKhU4t1jLI9f4/SMd7tZ3+Rda5ZX+b73m7pXDro4RZMHuy1+ux7V8gAoFuNuVYxeIYqIJEqIIiKJEqKISKKEKCKSKCGKiCRKiCIiyUDLbirlEt71jpu3X0gwUKbd5YN8AKBc5h+xe5F/rF+v8zKD9Xp8zDzodrMSDFGqjfASoTy4HwAwNsE704xMBIOHgg4nGfp0gQnKIkolPtCoWuIlE4U8Hqa1vspLfc6+9gaNvfgi77DTpzIE09O8bKndCPYg49dbC0qPAMA7fO8nZ3lp0sQkf6y73bgLTNRVaqPJO+xMTvLztlbjJUsAsHCJl5pF5VmZ8edDIRrQBaAUlFFF9ApRRCRRQhQRSZQQRUQSJUQRkUQJUUQkUUIUEUn6JkQzO2FmPzSzZ83sGTP743T558xszsweT3/uufHLFRG5cXZSh9gB8Cfu/piZjQH4mZn9IMW+5O7/bqcHazabeOWVl7aNRe16RsfjmqJmi9etWVAPVx0J2hI5rz0DgDzjW7cW1NFlDd7ia2Q0/v+p2+N1YrXR7duqAUCtymvILi8uhMfcXOMT3ko2QmO9It+fy+d5KykAeHnpFzQW1Sg6L7FD/GgCG6v8fr6+MUdjCxeC+xKXymF0lJ/X1VFe31gJpu5Zn5c47WBK4NISvy9TU9M0djRogQYA7eBxeXOOt0jzDq8Fnhjj5zsAZN6n8JTomxDd/TyA8+nrNTN7DsCxXR1NRGQfe0vvIZrZSQDvBfDjdNGnzOxJM3vQzHj3UhGRt4EdJ0QzGwXwbQCfdvdVAF8GcBuAU9h6BfkFcr0zZvaomT3a7u7uZayIyCDsKCGaWRFbyfDr7v4dAHD3i+7edfcegK8AuGu767r7A+5+2t1PF/P492ZFRIZpJ58yG4CvAnjO3b941eVHrvq2jwJ4+vovT0RkcHbyKfMHAfwBgKfM7PF02WcAfMLMTgFwAGcB/OENWaGIyIDs5FPm/wNs2yPq+2/1YK12F+dImULUzafLh5MBAKJP2N/564d5MGg51mzw8g4AaDV5fGOdlzasbfBYdYy3DQOAkaBMw52/2C/k/HqZx6240OV7tHCZl+xEVUtRWRIAbK7wcotOgz/YleAtmWqZlwgBQK02SmNRi6pzry3SWFCZBQCYmOR732rxNl6dI7M0NjUTf7Y5Ocqv29jgx1xZ4OdmtcDb6AFAJ2ifFk3Wa27yWB6UoG3dcNy+j9FvqoiIJEqIIiKJEqKISKKEKCKSKCGKiCRKiCIiyUCn7jkA1uylyge4odmnVcl7fmP7SX4A8Nv/4O/S2CuvvEhjl1+aD4/pwbS6mVneXabbXaWx5bgJDDY3eblKu3WRxuqbvNQnL8RlN5vrvNxiI5gu2Aiu5734N5ayoE1MMdj3UomXFxX71MBsLPPSGgvOv1pQL9bt0zFpfSkorWnxch4PSlXqQXciADh0+ACNHZg6QWM3HeXdd7qt+JivLvJpiO06fzwtuNlGUNoGAOUsSCgBvUIUEUmUEEVEEiVEEZFECVFEJFFCFBFJlBBFRJKBlt1YDhTHtv+YPSvxIVPTtbg05LZ3v5PGpg7N0NjIIu+AMn2ID9UBgHe+gx+z1eT/z/z8Md42cuV5XjoDAJ1gWE8hr9BYe5OXNiysxkOmVlZ4mVAnKP/IgsqasZG4JKKQ8VKWdpvXYmz1KibrKcelPqXgtUEWdEWaqo3RWMvjcpRGg3dkceN7u7HIT4S1lbg11KXzvJzn4KFDNPb+9/8dGrM+WaToQbelLn9ud7q8bKnXiV/LVaZ49yKAl4TpFaKISKKEKCKSKCGKiCRKiCIiiRKiiEiihCgikighiogkA61D7DnQICPy8gLPzTfdxtt7AcD4zASNvXmRtx5qtHkdWKsXT+2qjPC6vwOzvPbxsZ88QWNBuduWYLpgY5PXvFWKfG87neBGAfQ6QXum4P/Tcs7rSqP6xa0b5vWEhajdVoe3hFpv8lZbAFDK+eYXSzzWbfDbbbTiyXBW4PclC1qkNZvReRu3HFu4yNf05hyvUTx/jterTo7zdncAsLrE6/46zeBc6PI9yPq0c6uvB0W7Ab1CFBFJlBBFRBIlRBGRRAlRRCRRQhQRSZQQRUQSc1IGc0MOZnYJwOtXXTQL4PLAFtCf1hPbb+sB9t+atJ7YfljPze6+7fjBgSbE/+fgZo+6++mhLeAaWk9sv60H2H9r0npi+20919KPzCIiiRKiiEgy7IT4wJCPfy2tJ7bf1gPsvzVpPbH9tp5fMdT3EEVE9pNhv0IUEdk3hpIQzexuM3vBzF42s/uHsYZr1nPWzJ4ys8fN7NEhreFBM5s3s6evumzazH5gZi+lv6eGvJ7Pmdlc2qfHzeyeAa7nhJn90MyeNbNnzOyP0+VD2aNgPUPZIzOrmNlPzOyJtJ5/nS6/xcx+nJ5r3zSzeITlYNb0NTN77ao9OjWoNfXl7gP9AyAH8AqAWwGUADwB4I5Br+OaNZ0FMDvkNfwWgPcBePqqy/4tgPvT1/cD+LMhr+dzAP7ZkPbnCID3pa/HALwI4I5h7VGwnqHsEQADMJq+LgL4MYDfBPAtAB9Pl/8HAP90H6zpawB+fxjnUb8/w3iFeBeAl939VXdvAfgLAPcOYR37irv/CMC1DenuBfBQ+vohAB8Z8nqGxt3Pu/tj6es1AM8BOIYh7VGwnqHwLevpn8X0xwH8PQB/mS4f9DnE1rRvDSMhHgPwxlX/PochnkiJA/hrM/uZmZ0Z8lqudsjdr3S4vQCATxIfnE+Z2ZPpR+qB/Qh/NTM7CeC92HrFMfQ9umY9wJD2yMxyM3scwDyAH2DrJ7Fld7/SOXfgz7Vr1+TuV/bo36Q9+pKZlQe5pog+VNnyIXd/H4B/COCPzOy3hr2ga/nWzx3D/t/1ywBuA3AKwHkAXxj0AsxsFMC3AXza3Vevjg1jj7ZZz9D2yN277n4KwHFs/ST2a4M6NnPtmszsTgD/Eltrez+AaQD/YohL/BXDSIhzAE5c9e/j6bKhcfe59Pc8gO9i62TaDy6a2REASH/PD3Mx7n4xneA9AF/BgPfJzIrYSj5fd/fvpIuHtkfbrWfYe5TWsAzghwA+AGDSzK702x/ac+2qNd2d3m5wd28C+E/YP8+3oSTEnwK4PX36VQLwcQAPD2EdAAAzq5nZ2JWvAfwugKfjaw3MwwDuS1/fB+B7Q1zLlYRzxUcxwH0yMwPwVQDPufsXrwoNZY/Yeoa1R2Z2wMwm09dVAL+Drfc1fwjg99O3DfQcImt6/qr/wAxb72nul+fb4D9lTp843YOtT+VeAfCvhvmpErY+7c1/t6EAAACgSURBVH4i/XlmWOsB8A1s/YjVxtZ7PZ8EMAPgEQAvAfifAKaHvJ7/AuApAE9iKxEdGeB6PoStH4efBPB4+nPPsPYoWM9Q9gjA3wLw83TcpwH8abr8VgA/AfAygP8OoDzAx4yt6X+lPXoawH9F+iR6P/zRb6qIiCT6UEVEJFFCFBFJlBBFRBIlRBGRRAlRRCRRQhQRSZQQRUQSJUQRkeT/AqvEBnzYySNGAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "temp = X[1].copy()\n",
    "temp[:, :, 0] = temp[:,:, 0] + Y[1]\n",
    "show(temp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 349,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred = model.predict(np.array([test_img]) * 1 / 255.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 355,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "temp = test_img.copy()\n",
    "temp[:, :, 0] = temp[:,:, 0] + pred[0][:, :, 0] * 100\n",
    "show(temp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 357,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2.0406428527832032"
      ]
     },
     "execution_count": 357,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.sum(pred) / 200"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 333,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = np.array(X)\n",
    "Y = np.array(Y) * 200"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 246,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9998\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "1999.9998\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "1999.9998\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0002\n",
      "2000.0002\n",
      "2000.0\n",
      "1999.9999\n",
      "1999.9999\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "1999.9998\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "1999.9998\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "1999.9998\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0002\n",
      "2000.0001\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "1999.9998\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0002\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0001\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "1999.9999\n",
      "2000.0002\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "1999.9998\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0001\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0001\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0002\n",
      "2000.0\n",
      "1999.9999\n",
      "2000.0\n",
      "2000.0\n",
      "2000.0\n"
     ]
    }
   ],
   "source": [
    "for i in Y:\n",
    "    print(np.sum(i))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 224,
   "metadata": {},
   "outputs": [],
   "source": [
    "YY = []\n",
    "for i in Y:\n",
    "    YY.append(cv2.resize(i, (50, 37)) * 1000)\n",
    "YY = np.array(YY)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 225,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1957.5862 2.0\n",
      "3885.1108 4.0\n",
      "2966.4258 3.0000002\n",
      "2042.5576 2.0\n",
      "2070.1453 2.0\n",
      "2088.4058 2.0\n",
      "4033.766 4.0\n",
      "5075.481 5.0\n",
      "2063.2598 2.0\n",
      "1919.2952 2.0\n",
      "2066.2856 2.0\n",
      "2732.1338 3.0\n",
      "5081.0376 4.9999995\n",
      "2082.2898 2.0\n",
      "2133.7048 2.0\n",
      "6858.475 7.000001\n",
      "1939.9275 2.0\n",
      "2023.5555 2.0\n",
      "2091.1428 2.0\n",
      "2088.8027 2.0\n",
      "2057.3955 2.0\n",
      "2091.702 2.0\n",
      "2065.7512 2.0\n",
      "6225.318 6.0\n",
      "1986.8865 2.0\n",
      "4125.5034 4.0\n",
      "2157.9783 2.0000002\n",
      "3067.9995 3.0\n",
      "4018.5437 4.0\n",
      "1930.8595 2.0\n",
      "2075.8389 2.0\n",
      "1828.7651 2.0\n",
      "1947.6935 2.0\n",
      "2017.5442 2.0000002\n",
      "2085.7634 2.0\n",
      "11073.673 11.0\n",
      "2064.3914 2.0\n",
      "1965.2642 2.0000002\n",
      "1947.4069 2.0\n",
      "1848.6885 2.0000002\n",
      "3016.612 3.0\n",
      "1952.1895 2.0\n",
      "1969.1041 2.0\n",
      "9071.793 9.0\n",
      "2038.098 2.0\n",
      "2103.1643 2.0\n",
      "2075.2375 2.0\n",
      "2053.1255 2.0\n",
      "5830.7896 5.9999995\n",
      "4958.625 5.0\n",
      "2149.3667 2.0\n",
      "2008.615 2.0\n",
      "3110.1724 3.0\n",
      "2943.8125 3.0\n",
      "2983.1343 3.0\n",
      "1915.0321 2.0000002\n",
      "1849.3538 2.0\n",
      "4126.6597 4.0000005\n",
      "2088.318 2.0\n",
      "1926.629 2.0\n",
      "2998.199 2.9999998\n",
      "2082.8523 2.0\n",
      "2054.7273 2.0\n",
      "2040.5591 2.0\n",
      "2120.995 2.0000002\n",
      "2066.003 2.0\n",
      "3866.8367 4.0\n",
      "3960.8364 4.0\n",
      "1939.0271 2.0\n",
      "4881.4336 5.0\n",
      "1993.0237 2.0\n",
      "3035.7395 3.0\n",
      "3107.8032 3.0\n",
      "2932.0732 2.9999998\n",
      "4009.6682 4.0000005\n",
      "2069.9177 2.0\n",
      "1959.4519 2.0\n",
      "4883.5884 5.0\n",
      "4839.175 4.9999995\n",
      "9859.062 10.0\n",
      "2082.047 2.0\n",
      "4873.773 5.0\n",
      "2060.2417 2.0\n",
      "1893.5348 2.0\n",
      "1909.52 2.0\n",
      "2131.3613 2.0\n",
      "1969.3588 2.0\n",
      "1917.7471 2.0000002\n",
      "4869.5117 4.9999995\n",
      "2033.8402 2.0\n",
      "2068.0227 2.0\n",
      "2040.0745 2.0\n",
      "2104.276 2.0\n",
      "1935.744 2.0\n",
      "2048.4963 2.0\n",
      "5026.2686 5.0\n",
      "1741.6575 2.0\n",
      "2074.0823 2.0\n",
      "1882.9304 2.0\n",
      "6965.8447 7.0\n",
      "1952.7793 2.0\n",
      "4051.9165 4.0\n",
      "3060.623 3.0\n",
      "2081.711 2.0000002\n",
      "1949.9391 2.0\n",
      "2033.3359 2.0\n",
      "4026.3438 4.0\n",
      "2058.7188 2.0\n",
      "2000.4037 2.0\n",
      "2009.4978 2.0\n",
      "2068.3347 2.0\n",
      "1877.469 2.0000002\n",
      "3920.7317 4.0\n",
      "6003.787 6.0\n",
      "6778.0596 7.000001\n",
      "5156.4814 4.9999995\n",
      "3074.3884 3.0\n",
      "2065.751 2.0\n",
      "7072.17 7.0\n",
      "3043.003 3.0\n",
      "1956.6586 2.0\n",
      "1980.5358 2.0000002\n",
      "2103.7998 2.0\n",
      "3089.2483 3.0\n",
      "1917.7467 2.0\n",
      "2069.4653 2.0\n",
      "2061.83 2.0\n",
      "1911.0964 2.0\n",
      "3160.3438 3.0\n",
      "4064.733 3.9999998\n",
      "2078.5513 2.0\n",
      "1994.0455 2.0000002\n",
      "1951.2258 2.0\n",
      "2060.9956 2.0\n",
      "1953.9595 2.0\n",
      "2965.1218 3.0\n",
      "4997.7896 5.0\n",
      "5005.0747 5.0\n",
      "4092.1047 4.0\n",
      "2084.2146 2.0\n",
      "4146.7227 4.0\n",
      "2002.4729 2.0\n",
      "2099.316 2.0\n",
      "4019.0974 4.0\n",
      "2088.803 2.0\n",
      "2067.508 2.0\n",
      "4165.7603 4.0\n",
      "4050.4531 4.0000005\n",
      "2068.8032 2.0\n",
      "2005.9948 2.0\n",
      "3115.8206 3.0\n",
      "2060.8728 2.0\n",
      "3962.2441 4.0\n",
      "1911.5944 2.0\n",
      "1882.9299 2.0\n",
      "1984.3354 2.0\n",
      "3910.3188 4.0000005\n",
      "1943.5258 2.0\n",
      "2117.0342 2.0000002\n",
      "2056.5098 2.0\n",
      "1935.4825 2.0\n",
      "1818.107 2.0\n",
      "2064.7546 2.0000002\n",
      "2077.6858 2.0\n",
      "1956.8031 2.0\n",
      "1903.0569 2.0\n",
      "5826.7144 6.0\n",
      "3105.6726 3.0\n",
      "2071.4363 2.0\n",
      "3964.168 4.0000005\n",
      "3098.1724 3.0\n",
      "4057.406 4.0\n",
      "2998.1912 3.0000005\n",
      "1941.8645 2.0\n",
      "2061.8262 2.0\n",
      "3011.0386 3.0\n",
      "4124.037 4.0\n",
      "1941.0403 2.0\n",
      "2047.9703 2.0\n",
      "2101.5027 2.0\n",
      "2086.5623 2.0\n",
      "2025.7936 2.0\n",
      "2086.6328 2.0\n",
      "6121.786 6.0\n",
      "3012.342 3.0\n",
      "2896.257 3.0\n",
      "2103.9585 2.0\n",
      "3042.3062 3.0000002\n",
      "2074.86 2.0\n",
      "2004.408 2.0\n",
      "1935.7439 2.0\n",
      "2046.6163 2.0\n",
      "7077.406 6.9999995\n",
      "2045.8896 2.0\n",
      "1975.987 2.0\n",
      "3976.1636 4.0\n",
      "3812.8 3.9999995\n",
      "1975.9869 2.0\n",
      "2127.8022 2.0000002\n",
      "1928.7249 2.0\n",
      "1976.3298 2.0000002\n",
      "2070.2664 2.0000002\n",
      "5145.884 4.9999995\n",
      "2862.8298 3.0000005\n",
      "1946.7588 2.0\n",
      "3028.2288 3.0000005\n",
      "4888.955 5.0000005\n",
      "6020.477 6.0\n",
      "3952.9683 4.0\n",
      "1872.0342 2.0\n",
      "1871.5226 2.0\n",
      "1951.7483 2.0\n",
      "1975.9883 2.0\n",
      "2055.2053 2.0\n",
      "4745.206 5.0\n",
      "5973.427 6.0\n",
      "1927.8119 2.0\n",
      "2001.0625 2.0\n",
      "2096.4011 2.0\n",
      "3102.4294 3.0\n",
      "5122.6963 5.0\n",
      "3054.2085 2.9999998\n",
      "4015.5151 4.0\n",
      "5153.7 5.0\n",
      "4933.559 5.0\n",
      "4015.7903 4.0\n",
      "5117.8657 5.0\n",
      "3935.0625 4.0\n",
      "1867.0106 2.0\n",
      "2026.6139 2.0\n",
      "2028.3933 2.0\n",
      "2074.8765 2.0\n",
      "1827.2434 2.0000002\n",
      "2009.8586 2.0\n",
      "2085.8298 2.0\n",
      "2149.4185 2.0\n",
      "2873.9583 3.0000002\n",
      "3065.1926 3.0000002\n",
      "2036.4575 2.0\n",
      "2009.8591 2.0000002\n",
      "1934.5781 2.0\n",
      "1989.6228 1.9999999\n",
      "2990.1343 3.0\n",
      "2070.176 2.0\n",
      "5016.2266 5.0\n",
      "3041.9824 3.0\n",
      "1860.9175 2.0\n",
      "2092.6494 2.0\n",
      "1936.7778 2.0000002\n",
      "2008.105 2.0\n",
      "1983.2396 2.0\n",
      "2069.8403 2.0\n",
      "2069.2698 2.0\n",
      "2012.8756 2.0000002\n",
      "2045.8867 2.0\n",
      "3003.2817 3.0\n",
      "5038.697 5.0\n",
      "2155.853 2.0\n",
      "2057.087 2.0\n",
      "1931.6957 2.0\n",
      "3013.2231 3.0\n",
      "2068.1792 2.0000002\n",
      "1919.358 2.0\n",
      "1984.0999 2.0\n",
      "1982.4064 2.0\n",
      "2063.1953 2.0\n",
      "3903.6785 4.0\n",
      "2051.2803 2.0\n",
      "1848.6884 2.0000002\n",
      "1926.2644 2.0\n",
      "4926.4004 4.999999\n",
      "1877.0598 2.0\n",
      "3021.0762 3.0\n",
      "1823.5 2.0\n",
      "2094.4443 2.0\n",
      "1955.4658 2.0\n",
      "2084.093 2.0\n",
      "2066.462 2.0000002\n",
      "1872.0046 2.0\n",
      "2956.7979 3.0\n",
      "5997.4834 6.0\n",
      "4030.0806 4.0\n",
      "5122.1094 5.0\n",
      "2009.9473 2.0\n",
      "4909.8584 5.0\n",
      "2074.5137 2.0\n",
      "2133.3381 2.0\n",
      "1935.7439 2.0000002\n",
      "3118.6528 3.0000002\n",
      "1959.4031 2.0\n",
      "3018.0906 3.0000005\n",
      "1908.5348 2.0\n",
      "1935.7096 2.0\n",
      "2034.4316 2.0\n",
      "2075.2375 2.0\n",
      "2007.7057 2.0\n",
      "2099.113 2.0\n",
      "2034.8168 2.0\n",
      "7017.92 7.0\n",
      "1947.5819 2.0\n",
      "2061.8303 2.0\n",
      "3085.8484 3.0\n",
      "4113.459 4.0\n",
      "1967.3602 2.0\n",
      "6007.5894 6.0\n",
      "3109.0576 3.0\n",
      "2975.535 3.0\n",
      "5018.431 4.999999\n",
      "6130.0576 6.0\n",
      "3086.64 3.0\n",
      "4985.108 4.9999995\n",
      "2052.5896 2.0\n",
      "4127.727 4.0\n",
      "2039.3601 2.0\n",
      "2057.8916 2.0\n",
      "2032.7423 2.0\n",
      "1932.6765 2.0\n",
      "1818.1073 2.0\n",
      "2078.872 2.0\n",
      "4179.286 4.0\n",
      "1914.5144 2.0\n",
      "2089.8855 2.0\n",
      "2054.3635 2.0\n",
      "4944.191 5.0\n",
      "2085.8298 2.0\n",
      "3175.7405 2.9999998\n",
      "1950.7515 2.0\n",
      "1951.1581 2.0\n",
      "1950.7513 2.0\n",
      "1923.9409 2.0\n",
      "4878.656 4.9999995\n",
      "2065.9932 2.0\n"
     ]
    }
   ],
   "source": [
    "for ix in range(len(YY)):\n",
    "    rez = cv2.resize(YY[ix], (400, 300))\n",
    "    print(np.sum(rez), np.sum(Y[ix]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 363,
   "metadata": {},
   "outputs": [
    {
     "ename": "AttributeError",
     "evalue": "'numpy.ndarray' object has no attribute 'length'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-363-2dc026ff9f16>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwhere\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtemp\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlength\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mAttributeError\u001b[0m: 'numpy.ndarray' object has no attribute 'length'"
     ]
    }
   ],
   "source": [
    "np.where(temp==-1)[0].length"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}