masks_from_mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(28, 28, 60000)"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from keras.datasets import mnist\n",
    "digits = mnist.load_data()[0][0]\n",
    "import scipy.ndimage\n",
    "digidt2 = scipy.ndimage.zoom(digits, (1, 3, 3), order=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "count = 1000\n",
    "patch_size = 128\n",
    "out = np.zeros((count, patch_size, patch_size))\n",
    "smol = digits[np.random.randint(0, len(digits), count)]\n",
    "out[:, :28, :28] = smol / 2\n",
    "out = np.roll(out, np.random.randint(0, patch_size), 1)\n",
    "out = np.roll(out, np.random.randint(0, patch_size), 2)\n",
    "smol = digits[np.random.randint(0, len(digits), count)]\n",
    "out[:, :28, :28] = smol / 2\n",
    "out = np.roll(out, np.random.randint(0, patch_size), 1)\n",
    "out = np.roll(out, np.random.randint(0, patch_size), 2)\n",
    "medium = digidt2[np.random.randint(0, len(digits), count)]\n",
    "out[:, :84, :84] += medium / 2\n",
    "out = np.roll(out, np.random.randint(0, patch_size), 1)\n",
    "out = np.roll(out, np.random.randint(0, patch_size), 2)\n",
    "medium = digidt2[np.random.randint(0, len(digits), count)]\n",
    "out[:, :84, :84] += medium / 2\n",
    "out = np.roll(out, np.random.randint(0, patch_size), 1)\n",
    "out = np.roll(out, np.random.randint(0, patch_size), 2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQUAAAD8CAYAAAB+fLH0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAEVRJREFUeJzt3X2sHNV5x/HvrzbYNSkxhsjyC6pd4SaiKAFkgRFVhXAi\nA6WYSgiBUOKAJasSLSSNlNrlD1SpkYIShVAppb0CglshXuLQ2qI0DjhEUf/AxQaLgB2Dw+s1NiaK\nIVGowE6f/jFzy57LXu7endfd/X0kyzuzs3eee+7uM885M7NHEYGZ2YTfaToAM2sXJwUzSzgpmFnC\nScHMEk4KZpZwUjCzhJOCmSUqSwqSLpG0X9IBSRur2o+ZlUtVXLwkaRbwAvA5YBx4Crg2IvaWvjMz\nK9Xsin7uecCBiHgJQNIDwFqga1I4UXNiLidVFEo7/eGn3206hEa88Oy8pkMYWb/m6C8i4hPTbVdV\nUlgCvN6xPA6c37mBpA3ABoC5zON8ra4olHbavn1P0yE0Ys3is5sOYWQ9Hlte7WW7qpLCtCJiDBgD\nOFkLRu4GDH84rK2qGmg8CJzesbw0X2dmLVdVUngKWCFpuaQTgWuAbRXty8xKVEn3ISKOS/pLYDsw\nC7gnIp6vYl9mVq7KxhQi4lHg0ap+vplVw1c0mlnCScHMEk4KZpZwUjCzhJOCmSWcFMws4aRgZgkn\nBTNLOCmYWcJJwcwSTgpmlnBSMLOEk4KZJZwUzCzhpGBmCScFM0s4KZhZwknBzBJOCmaWcFIws4ST\ngpklnBTMLOGkYGYJJwUzSzgpmFmi76Qg6XRJT0jaK+l5STfn6xdIekzSi/n/p5QXrplVrUilcBz4\nSkScCawCbpR0JrAR2BERK4Ad+bKZDYi+k0JEHIqIp/PHvwb2AUuAtcDmfLPNwJVFgzSz+pQywayk\nZcA5wE5gYUQcyp86DCyc4jUbgA0Ac5lXRhhmVoLCA42SPgZ8H/hSRPyq87mICCC6vS4ixiJiZUSs\nPIE5RcMws5IUSgqSTiBLCPdFxMP56jclLcqfXwQcKRaimdWpyNkHAXcD+yLiWx1PbQPW5Y/XAVv7\nD8/M6lZkTOFC4PPATyXtydf9LfB14CFJ64FXgauLhWhmdeo7KUTEfwGa4unV/f5cM2uWr2g0s4ST\ngpklnBTMLOGkYGYJJwUzSzgpmFnCScHMEk4KZpZwUjCzhJOCmSWcFMws4aRgZgknBTNLOCmYWcJJ\nwcwSTgpmlnBSMLOEk4KZJZwUzCxRymQwo2r7G3um3whYs/jsiiMxK4+TQh96TQaTt3dysEHg7oOZ\nJVwpzNBMq4TpXuvqwdrGlYKZJQpXCpJmAbuAgxFxuaTlwAPAqcBu4PMR8X7R/TStSIXQy891xWBt\nUUalcDOwr2P5NuD2iDgDOAqsL2EfQ2/7G3v+/59Zk4rOOr0U+FPgrnxZwMXAlnyTzcCVRfZhZvUq\n2n34NvBV4Pfy5VOBtyPieL48DiwpuI+R4S6EtUGRqegvB45ExO4+X79B0i5Ju47xXr9hmFnJik5F\nf4Wky4C5wMnAHcB8SbPzamEpcLDbiyNiDBgDOFkLokAcrTD5KL/9jT1d130UDzpaG/RdKUTEpohY\nGhHLgGuAH0XEdcATwFX5ZuuArYWjNLPaVHGdwt8Afy3pANkYw90V7KM2/Z4R+Kij/XSVgM9CWJNK\nuaIxIn4M/Dh//BJwXhk/d1D0Wu53bjfx2B9+axtf0WhmCScFM0s4KZhZwknBzBK+dbohHmC0tnKl\nYGYJJ4VprFl8tq8wtJHipGBmCSeFEnh8wIaJBxpLMpObmaZLIu6uWJNcKZhZwpVCj3q9V6GzYphc\nPbhCsEHgSsHMEopo/vtNTtaCOF+rmw6jZ1UMLLpKKJ8rs9TjsWV3RKycbjtXCmaW8JjCCCur4hnU\nI66//q47dx8KKOND1cQbsqrrKobhdxnmBOHug5n1xd2HAj7qqDKKg1x1luNVT+M3YRj/TtNxpWBm\nCY8pjJAm7tGo4kjb1L0mg1419Dqm4O7DiGjqg1Rml6LpG89G5WyFuw9mlnClMOSaPrq2QS+T79gH\nXCmYWaJQpSBpPnAXcBYQwA3AfuBBYBnwCnB1RBwtFKVZD/rt6/dyF+uwjyN0Klop3AH8ICI+BXwG\n2AdsBHZExApgR75sNRu1+ShH6UNbtb6TgqSPA39CPoFsRLwfEW8Da4HN+WabgSuLBmlm9SlSKSwH\n3gK+K+kZSXdJOglYGBGH8m0OAwuLBmmDa1i+DXuUqq4iSWE2cC5wZ0ScA/yGSV2FyK6M6np1lKQN\nknZJ2nWM9wqEYWZlKjLQOA6MR8TOfHkLWVJ4U9KiiDgkaRFwpNuLI2IMGIPsisYCcVjDhqESsA/0\nXSlExGHgdUmfzFetBvYC24B1+bp1wNZCEZpZrYpevPRXwH2STgReAq4nSzQPSVoPvApcXXAf1nKj\ncvnvqCiUFCJiD9DtBgvf3dSQJgfEut123C0eJ4928xWNZpbwvQ9Wmamqll6qmZlWE9vf2OMKpCSu\nFMws4UrBWqmzmui1AqhyPGWUqhAnhSHT6xR1g6iJ322UksEEdx/MLOFKYcgMY4XQxHUQo1ghTHCl\nYGYJVwpDYBirg266DT52VhGDOmNX27hSMLOEKwUbaJ1Hdh/ly+GkMMBGpdvQjW/Cqo67D2aWcFIY\nUKNcJXQatS+orYOTgpklnBQGjI+M3blNyuOkYEPDCbMcTgpmlvApyQHQ9NHPE7SOFlcKZpZwpdBi\nTR+Be70wqG3f4eALm4pxUmihpj9cVc7eXCcnh/64+2BmCVcKLdKWI2xRnUfmNvxO/qbnmXGlYGaJ\nQklB0pclPS/pOUn3S5orabmknZIOSHown1LOzAZE30lB0hLgJmBlRJwFzAKuAW4Dbo+IM4CjwPoy\nAjWzehTtPswGflfSbGAecAi4mGxaeoDNwJUF92EDbM3is2fUn5/p9la+vgcaI+KgpG8CrwH/A/wQ\n2A28HRHH883GgSWFoxwBbRiQg+pO3/mDPjiKdB9OAdYCy4HFwEnAJTN4/QZJuyTtOsZ7/YZhZiUr\nckrys8DLEfEWgKSHgQuB+ZJm59XCUuBgtxdHxBgwBnCyFkSBOAZaWyqEYTe5nV25TK3ImMJrwCpJ\n8yQJWA3sBZ4Arsq3WQdsLRaimdWpyJjCTklbgKeB48AzZEf+/wAekPT3+bq7ywh02LStQvCR0yYU\nuqIxIm4Fbp20+iXgvCI/16zq+yh8lePUfEWjmSWcFBrQtq5Dm/loXj8nBTNLOClY6/kqx3r51mkD\nus/obKPJlYKZJVwp1GhQBhhnEqeriuHjSsHMEq4UajAoFUI/PBYxfFwpmFnClUKFhrlC6MZfqT4c\nnBSsdO5SDDZ3H8ws4UrBKlVGl2LUumFNc6VgZgknBavF9jf29HXEd5VQPycFq1WvH/J+k4gV56Rg\nZgknBatdG6oAnyqdmpOCmSV8SrJCVXz5aK9HuKaPxL3wFZDt5KRQg6LJoe4PTdXfpDxZnd+s7AQ0\nPXcfzCzhSqFGMz0CN31Uq7NiqLor0XRbDhJXCmaWmLZSkHQPcDlwJCLOytctAB4ElgGvAFdHxNF8\nTsk7gMuAd4EvRsTT1YQ+uAbtqNUZb9VVwyAMkA67XiqFe/nwFPMbgR0RsQLYkS8DXAqsyP9tAO4s\nJ0wzq8u0SSEifgL8ctLqtcDm/PFm4MqO9f8SmSfJpqVfVFawVo+PurjIczAMv37HFBZGxKH88WFg\nYf54CfB6x3bj+TobQE4Oo6nwQGNEBBAzfZ2kDZJ2Sdp1jPeKhmFmJek3Kbw50S3I/z+Srz8InN6x\n3dJ83YdExFhErIyIlScwp88wrA7TVQw2XPpNCtuAdfnjdcDWjvVfUGYV8E5HN8PMBkAvpyTvBy4C\nTpM0DtwKfB14SNJ64FXg6nzzR8lORx4gOyV5fQUxWw/qvPCo7suirVrTJoWIuHaKp1Z32TaAG4sG\nZWbN8RWNZpbwvQ9DrszSflC7Bx4MnRlXCmaWcFIYEaN4tPQFVv1xUjCzhMcUbOi4OijGlcIIcTlt\nvXBSMLOEuw82NFwFlcOVgpklnBTMLOHuwwgaphuY3GUonysFM0s4KYywQT7K+vRqdZwUzCzhMYUR\nN/loOwjjDK4QquWkYIk6J36ZKSeDerj7YGYJJwWbkgfzRpOTgpklnBRsWq4YRouTgpklfPbBetZr\ntdC2sxY2M04KM9TGN3zVpX0bf2erjrsPZpboZdq4e4DLgSMRcVa+7hvAnwHvAz8Hro+It/PnNgHr\ngd8CN0XE9opir8UgHCUHIcYiPMhZr14qhXuBSyateww4KyI+DbwAbAKQdCZwDfBH+Wv+UdKs0qI1\ns8pNmxQi4ifALyet+2FEHM8XnySbch5gLfBARLwXES+TTTR7Xonx2ohxlVC/MgYabwAezB8vIUsS\nE8bzdWYz4mTQnEJJQdItwHHgvj5euwHYADCXeUXCMLMS9Z0UJH2RbABydT4FPcBB4PSOzZbm6z4k\nIsaAMYCTtSC6bWNm9evrlKSkS4CvAldExLsdT20DrpE0R9JyYAXw38XDNLO69HJK8n7gIuA0SePA\nrWRnG+YAj0kCeDIi/iIinpf0ELCXrFtxY0T8tqrgzax80yaFiLi2y+q7P2L7rwFfKxKUmTXHVzSa\nWcL3Plir+FRk81wpmFnClcI0hmk2pbZyddAurhTMLOFKoUf+gpHyuUJoJyeFkrm7MT0ng3Zz98HM\nEvrgtoUGg5DeAn4D/KLpWIDTcBydHEdqkOP4/Yj4xHQbtSIpAEjaFRErHYfjcBzNxuHug5klnBTM\nLNGmpDDWdAA5x5FyHKmhj6M1Ywpm1g5tqhTMrAVakRQkXSJpv6QDkjbWtM/TJT0haa+k5yXdnK9f\nIOkxSS/m/59SUzyzJD0j6ZF8ebmknXmbPCjpxBpimC9pi6SfSdon6YIm2kPSl/O/yXOS7pc0t672\nkHSPpCOSnutY17UNlPmHPKZnJZ1bcRzfyP82z0r6N0nzO57blMexX9KaIvtuPCnk80J8B7gUOBO4\nNp8/omrHga9ExJnAKuDGfL8bgR0RsQLYkS/X4WZgX8fybcDtEXEGcJRsgp2q3QH8ICI+BXwmj6fW\n9pC0BLgJWJlPPjSLbC6RutrjXj48z8lUbXAp2VcOriD7EuI7K46jnvlWIqLRf8AFwPaO5U3Apgbi\n2Ap8DtgPLMrXLQL217DvpWRvtouBRwCRXZgyu1sbVRTDx4GXyceZOtbX2h5kUwK8Diwguwz/EWBN\nne0BLAOem64NgH8Gru22XRVxTHruz4H78sfJZwbYDlzQ734brxT44E0wofa5IiQtA84BdgILI+JQ\n/tRhYGENIXyb7Itw/zdfPhV4Oz6YcKeONlkOvAV8N+/G3CXpJGpuj4g4CHwTeA04BLwD7Kb+9ug0\nVRs0+d69AfjPKuJoQ1JolKSPAd8HvhQRv+p8LrK0W+npGUkT83TurnI/PZgNnAvcGRHnkF12nnQV\namqPU8hmGlsOLAZO4sNldGPqaIPpFJlvpRdtSAo9zxVRNkknkCWE+yLi4Xz1m5IW5c8vAo5UHMaF\nwBWSXgEeIOtC3AHMlzRxF2sdbTIOjEfEznx5C1mSqLs9Pgu8HBFvRcQx4GGyNqq7PTpN1Qa1v3c7\n5lu5Lk9QpcfRhqTwFLAiH10+kWzAZFvVO1X23fR3A/si4lsdT20D1uWP15GNNVQmIjZFxNKIWEb2\nu/8oIq4DngCuqjGOw8Drkj6Zr1pN9lX9tbYHWbdhlaR5+d9oIo5a22OSqdpgG/CF/CzEKuCdjm5G\n6Wqbb6XKQaMZDKhcRjaa+nPglpr2+cdkZeCzwJ7832Vk/fkdwIvA48CCGtvhIuCR/PEf5H/YA8D3\ngDk17P9sYFfeJv8OnNJEewB/B/wMeA74V7I5RmppD+B+srGMY2TV0/qp2oBsQPg7+fv2p2RnTKqM\n4wDZ2MHE+/WfOra/JY9jP3BpkX37ikYzS7Sh+2BmLeKkYGYJJwUzSzgpmFnCScHMEk4KZpZwUjCz\nhJOCmSX+D3rHR64yNhqcAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x2e8164986a0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(out[320] > 5)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}