mirror of
https://github.com/saymrwulf/nvidia-AI.summerschool17-notebooks.git
synced 2026-11-21 09:54:43 +00:00
329 lines
211 KiB
Text
329 lines
211 KiB
Text
|
|
{
|
||
|
|
"metadata": {
|
||
|
|
"name": "hw-edge-detection"
|
||
|
|
},
|
||
|
|
"nbformat": 3,
|
||
|
|
"nbformat_minor": 0,
|
||
|
|
"worksheets": [
|
||
|
|
{
|
||
|
|
"cells": [
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"from pylab import *\n",
|
||
|
|
"from scipy.ndimage import filters,measurements\n",
|
||
|
|
"from scipy import stats"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [],
|
||
|
|
"prompt_number": 85
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Let's start with a \"trivial\" edge detector: this just finds all the pixels where local differences are above a threshold."
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"def trivedges(m,threshold=1e-3):\n",
|
||
|
|
" return abs(m-roll(m,1,0))+abs(m-roll(m,1,1))>threshold"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [],
|
||
|
|
"prompt_number": 45
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "heading",
|
||
|
|
"level": 1,
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Mondrian World"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Generate a \"Mondrian world\" image."
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"def mondrian(dims=(200,200),sr=(20,50),n=100,noise=0.5):\n",
|
||
|
|
" h,w = dims\n",
|
||
|
|
" image = zeros(dims)\n",
|
||
|
|
" for i in range(n):\n",
|
||
|
|
" y = randint(h)\n",
|
||
|
|
" x = randint(w)\n",
|
||
|
|
" ry = randint(*sr)\n",
|
||
|
|
" rx = randint(*sr)\n",
|
||
|
|
" y0 = clip(y-ry,0,h)\n",
|
||
|
|
" y1 = clip(y+ry,0,h)\n",
|
||
|
|
" x0 = clip(x-rx,0,w)\n",
|
||
|
|
" x1 = clip(x+rx,0,w)\n",
|
||
|
|
" image[y0:y1,x0:x1] = rand()\n",
|
||
|
|
" edges = trivedges(image)\n",
|
||
|
|
" return image+noise*randn(*image.shape),edges"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [],
|
||
|
|
"prompt_number": 101
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"The `mondrian` function returns both a noisy image and an edge map."
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"image,edges = mondrian(noise=0.1)\n",
|
||
|
|
"subplot(121); imshow(image,cmap=cm.gray)\n",
|
||
|
|
"subplot(122); imshow(edges,cmap=cm.gray)"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"output_type": "pyout",
|
||
|
|
"prompt_number": 110,
|
||
|
|
"text": [
|
||
|
|
"<matplotlib.image.AxesImage at 0x1499dd50>"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"output_type": "display_data",
|
||
|
|
"png": "iVBORw0KGgoAAAANSUhEUgAAAXIAAAC2CAYAAADEKgYfAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nFmpPe33oQx+Cbds/8tfncrk3
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"prompt_number": 110
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Trivial edge detection applied to the noisy image results in rather noisy output."
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"imshow(trivedges(m,threshold=0.5),cmap=cm.gray)"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"output_type": "pyout",
|
||
|
|
"prompt_number": 112,
|
||
|
|
"text": [
|
||
|
|
"<matplotlib.image.AxesImage at 0x14e97b50>"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"output_type": "display_data",
|
||
|
|
"png": "iVBORw0KGgoAAAANSUhEUgAAAQIAAAD9CAYAAABX/HkoAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzsnXdcFOf2/w/LsrCwNOlNmnQVsAIqIqEpNuwaiYjBmKDe\nWJOYGLEbr5rcBEtUosZI4jcWosGIaGK8Go0SE4OgscdG7IoVLOf3B3fmzu4+Mzttd8n9cV6vz+sF\nO888/Tk7O/PMeVsgIkKTNVmT/X9tCnNXoMmarMnMb02OoMmarMmaHEGTNVmTNTmCJmuyJoMmR9Bk\nTdZk0OQImqzJmgyM4Ah27twJ4eHhEBISAh988IHc2TdZkzWZEcxCzn0Ez58/h7CwMNi9ezf4+PhA\n+/bt4csvv4SIiAi5imiyJmsyI5isVwSHDx+GFi1aQEBAAFhZWcGQIUPgm2++kbOIJmuyJjOCyeoI\nrly5An5+fvT/vr6+cOXKFTmLaLImazIjmFLOzCwsLGRJ02RN1mTGM9LdAFmvCHx8fODSpUv0/5cu\nXQJfX1+9dI6OjgAAEBUVBcHBwaz5zZ8/H9LT0+n/XV1dIS0tDZKTk0GpVEKnTp3AxcUFioqKYMaM\nGeDr6wtt2rRhzS84OBjS0tJ4adq0aXD+/HnIysqCR48ewahRo8DFxYWzzlR95DJD/UMyjUZD9w8A\nQNu2bcHHx4f3+Yb6EADAwcGBTqNUKiE5ORns7Oz0xkuqBQcHQ2RkpCx5de/eHebOnav3udD++btY\nWVkZ5Obman0WEBDAfgLKaE+fPsWgoCA8f/481tXVYXR0NFZXV2ulAQB0dXVFADCotLQ0DAkJMZiu\nc+fOGBMTwyvPxMREnDhxosF0Pj4+2LdvX1SpVDh8+HB0dHTUS9OqVSucPXs2r3L/Durbty/m5OSI\nPp8aL6n58FVubi726dOHV7vmzZuHqamprPl89tln+Nlnn2F8fLws84eaQ0uXLkUbGxvicWPOH9La\ncXZ2RgDykpf1ikCpVEJhYSGkp6dDZGQkDB48mPjE4NGjR/TfycnJ0LdvXwAA8PDwgPfeew+sra0B\nAGDXrl1w+vRpg+Xu378ffvvtN4PpkpOTITk5GWpqagymvXLlCpSUlEB9fT188cUXcO/ePWI7Ll26\nBDY2NjB9+nRwc3MzmK9QY/aPsS0mJgYSEhJEn0+N1+3bt+HGjRsAoD+mlMnRruvXr8Pt27cNprt9\n+zYcOHAAysvLicc7deoEI0eOhJEjR0JQUBBnXvfv3weNRgOzZ88mtotp9fX1cP78eXjx4gXxODV/\njGGktXPnzh32E+S8IuBjAIAajYb2Uj179sThw4cjAKC3tzd++OGHrB6UkkqlwtzcXNrD8U3DLEtO\nOTo6Ym1tLRYUFKCvry/9eUREBPbo0UNS3saqs6nKYhtTU7aLTYMHD0ZfX1/Mzs7G5cuX4/LlyzEr\nK8vgeVFRUbhx40a8e/cu2tvby14fsefzWRcA5CVvFkfg5+cnqcNsbW1x48aNnPmQ0oSFhWHr1q1F\nlxsXF4e9evXC0NBQYlnbtm3D2NhY+vO0tDQsKChgzY+qj7u7OyYlJZl1UfzdJaYPly1bpjVeQuTv\n749fffUVqtVq2dogpT4A/NYFQCNyBOZQ8+bN8dNPP8UtW7ZgcHAw2tjYYGxsLMbGxhq8AgFo+HY/\nePAgIiLOmDFDljqNGTMGp0+fjgkJCfj111+brW905eDggOHh4VqfBQcHo4uLi0nr4eXlhc2bNyce\n062PqfvQxsYGW7dujQqFglf6iIgIWa8epPQhcV2a0gkgNjgCd3d3VKlUJu2Qjz76CK9du4bXrl3D\nr776CsPDw+n/2W5IKpVKdHNzQwsLC9y/fz/W1dXhvXv3cPLkybzKVKvV6OTkJKnednZ26O7uzpqP\ng4MD2tnZydJHdnZ2aG9vj8nJyfjjjz9qHdu4cSMOHTqUTsM3T6VSie7u7uju7o5KpZK1XRYWFujm\n5qaVZsqUKbh48WJivlR9jDFXmOPOliY8PBxPnz6NGo0GmzVrhu7u7sQvFKpdBw8eFHXVp1KpsFmz\nZqLbQupD4ro0sR9AAMC6ujpMSUkxyiByDa5KpUKVSoVKpRItLCzo/9kGPCoqCq9fv4729vZoZWWF\n5eXlmJeXh5aWlrzKzM3Nxe3bt0uq9/Tp07Guro41n5UrV+J7770nSx9Nnz4dly9fjgqFAq2srPT6\nT6FQ0Gn45hkVFYV1dXVYV1endZWh2y5HR0e8efOmVhpLS0stx0CqjzHmCnPc2dJQ8wcA8NixY1hX\nV4cjRozQS0e1q1WrVqLqm5aWhkePHhXdFlIfEtelif0AAgCePn0a09LSjDKIcikxMRG///579Pf3\npwfQ09NT0LdhXl4efvfdd5Lq4ezsjIGBgTho0CDcuXOn3nFXV1fJVx3Msgw92uWThqmWLVtibW0t\ntmzZUsu5UO3y9PREAECFQoH+/v56DsgcUqlUWuNuSL6+vhgYGIjLli3D8ePHax2T2i61Wo3dunXD\nY8eOaV35ffDBB3plsWnIkCG4du1a+n/iujSxH0AAwOrqaqM7Ajs7OywrK0N/f3/e57z99ts4cuRI\neqJKuXEDII8joCRHfcwhW1tbjIuL07qKYvYzn7FoTOrVqxcuWrSIeCwkJIT1noZYdezYEY8ePar3\nhEJIWZ6enhgVFUX/T1yXJvYDCGB8R+Dn54fr16/H4cOHo4ODA+/zOnTooNVhUiWnI2hsio6Oxn/+\n85+izuXTz3KPhVwKDg7Grl27op+fH3722We8bjRzyVA+Xl5eOGTIEOzfvz/rzyShIq5LE/sBBDC+\nI3B1dcWxY8dKviHp6emJc+bMQWtra1Hnt2rVCvv16yfq3NGjR2OHDh1EnWtjY4Nz585Fd3d3o/Vx\nQEAA8TexuepjagmdY1OnTtV77CwmHzlEskYboWjEiBHg7+8vKs3NmzehsLAQ6uvrJZelUIjvosrK\nStiyZQvxWGpqKsTFxbGeq1AoJL2gpVAoePWhWLtw4QKsW7eOV1pra2sYN26c4Pcw4uPjISUlRUz1\njG6kOcY1pmzjyWeuuri4wBtvvAEqlUp6xVms0TqC6Oho+uUkNnvrrbcgLCxMclls+fz1118wbdo0\nqKurk1yGrrVo0YL4QhZlK1asgJ9//llU3k+ePIF33nkHvLy8DPahKezp06ewdetWqK2tFXSer6+v\n4JeuzGlcY7pgwQL4448/ROVra2sL7du3B0tLS17pMzIyYNCgQRAYGMi/EBP/MkAA+X4aFBcXi758\nZspYP1U8PDwwMjISARoe43Ts2BETEhJoSd1haS4x28VHVNuZu/D8/PwwODgYnZycDL4wxicNl6iy\nABpuXnbo0IH3I2AAwMjISL2fNaR8WrRoIWpMSf2jWxY1Z/jsZCwuLsYDBw5gz549iceJ69LEfgAB\nhC88Jycno+5q2717N3bp0kX2soYMGYLr1q1DAEB7e3usqqrCCxcu0PrHP/5htDYZU8x28RH1LkZE\nRAT92YQJE3DBggXYtWtX3LVrF+f5fNJwiSoLoGFbd2VlpaBNWOvXr8dBgwZpfUbKZ+HChaLGVLd/\nXFxctN52DQsLo+dMixYtJI8fcV2a2A8gwH8dgUKh4PWs9v3338cVK1aYZJKbsqy/uywsLHjdyXZ0\ndMTbt2/rbVsmydLSknNHn7HEdy4KbTsf6fbPqlWrcPr06XQ5Qq5e+Ii4Lk3sBxDgv47gww8/xEmT\nJhmsuEajEfQYUIpMWdbfXeHh4Xjq1Cmtt0lJsrCwQA8PD14LZ9++fdi1a1eTt2Xy5Mms+wOktJ2P\ndPvH0dGRzrdbt274ww8/yNpW4ro0sR9AgP86An9/f/Ty8jL7hJZThYWF2Lt3b7PXg03x8fG4ceNG\nWfJSq9UYExMj6zdWVFSUWRyxt7e3oM1nxmg7SQ4ODgbvxwgdU+K6NLEfQAD9ewRpaWlYUlKCH3/8\nMWvl09LSWN/6GzduHJaUlGB+
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"prompt_number": 112
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "heading",
|
||
|
|
"level": 1,
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"F-Measure Evaluation"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Evaluation of an edge detector can be done by the _f-measure_.\n",
|
||
|
|
"This computes the fraction of returned edge pixesl that are correct (precision)\n",
|
||
|
|
"and the fraction of all true edge pixels that are returned (recall).\n",
|
||
|
|
"The harmonic mean of these two is called the _f-measure_.\n",
|
||
|
|
"\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"def evaluate_edges(truth,pred,b=5):\n",
|
||
|
|
" cases = array(2*(truth!=0)+(pred!=0),'i')[b:-b,b:-b]\n",
|
||
|
|
" c00,c01,c10,c11 = measurements.sum(ones(cases.shape),cases,arange(4))\n",
|
||
|
|
" precision = c11 * 1.0 / (c01 + c11)\n",
|
||
|
|
" recall = c11 * 1.0 / (c10 + c11)\n",
|
||
|
|
" fmeasure = precision * recall / (precision + recall)\n",
|
||
|
|
" return fmeasure,precision,recall\n",
|
||
|
|
"\n"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [],
|
||
|
|
"prompt_number": 113
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"evaluate_edges(edges,trivedges(m,threshold=0.2))"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"output_type": "pyout",
|
||
|
|
"prompt_number": 114,
|
||
|
|
"text": [
|
||
|
|
"(0.046415529502210263, 0.051003167898627244, 0.51602564102564108)"
|
||
|
|
]
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"prompt_number": 114
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"We can now try to find the optimal edge detection parameters.\n",
|
||
|
|
"\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"es = []\n",
|
||
|
|
"ts = linspace(0.0,1.5,100)\n",
|
||
|
|
"for threshold in ts:\n",
|
||
|
|
" e = evaluate_edges(edges,trivedges(image,threshold=threshold))\n",
|
||
|
|
" es.append(e)\n",
|
||
|
|
"es = array(es)\n",
|
||
|
|
"plot(ts,es[:,0])\n",
|
||
|
|
"plot(ts,es[:,1])\n",
|
||
|
|
"plot(ts,es[:,2])"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"output_type": "pyout",
|
||
|
|
"prompt_number": 117,
|
||
|
|
"text": [
|
||
|
|
"[<matplotlib.lines.Line2D at 0x15691dd0>]"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"output_type": "display_data",
|
||
|
|
"png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAD9CAYAAABHnDf0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nQ3in4Tv4FvfN/GAhLMg/B77j
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"prompt_number": 117
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "heading",
|
||
|
|
"level": 1,
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Blobworld"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Here is a second model image generator.\n",
|
||
|
|
"\n"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"def blobworld(dims=(200,200),sigma=10.0,frac=0.5,noise=0.5):\n",
|
||
|
|
" h,w = dims\n",
|
||
|
|
" image = randn(*dims)\n",
|
||
|
|
" image = filters.gaussian_filter(image,sigma)\n",
|
||
|
|
" threshold = stats.scoreatpercentile(image,frac*100)\n",
|
||
|
|
" image = (image>threshold)\n",
|
||
|
|
" image,n = measurements.label(image)\n",
|
||
|
|
" image = rand(n+1)[image]\n",
|
||
|
|
" edges = trivedges(image)\n",
|
||
|
|
" return image+noise*randn(*image.shape),edges"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [],
|
||
|
|
"prompt_number": 127
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [
|
||
|
|
"bw,be = blobworld()\n",
|
||
|
|
"subplot(121); imshow(bw,cmap=cm.gray)\n",
|
||
|
|
"subplot(122); imshow(be,cmap=cm.gray)"
|
||
|
|
],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"output_type": "pyout",
|
||
|
|
"prompt_number": 128,
|
||
|
|
"text": [
|
||
|
|
"<matplotlib.image.AxesImage at 0x16ccb950>"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"output_type": "display_data",
|
||
|
|
"png": "iVBORw0KGgoAAAANSUhEUgAAAXIAAAC2CAYAAADEKgYfAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzsvXeQZXWZ//+6OefbN/S9t3Punu5JMDMEQQUUcFHKgIoW\nZcS0iqHcMpW75a5CrWHX1VrKsJS1sJYoVgm6IopDchiY0JM7x3u7++ac4++PqfP5yRcDoaEHt99V\nVNHd955z+vaZ53w+z/MOslar1WIb29jGNrbxioV8qy9gG9vYxja28eKwXci3sY1tbOMVju1Cvo1t\nbGMbr3BsF/JtbGMb23iFY7uQb2Mb29jGKxzbhXwb29jGNl7h2PRC/uCDDzI0NER/fz933HHHZh9+\nG9vYMmzf29u4UCHbTB55o9FgcHCQ3/3ud/h8Pi666CJ+/OMfMzw8vFmn2MY2tgTb9/Y2LmRs6or8\n6aefpq+vj66uLlQqFW9/+9v5xS9+sZmn2MY2tgTb9/Y2LmRsaiFfW1sjEAiIr/1+P2tra5t5im1s\nY0uwfW9v40KGcjMPJpPJNuU129jGi8VmO09s39vbuBDw5+7rTS3kPp+PYDAovg4Gg/j9/me97jWv\neQ0DAwO0t7czPT1NZ2cngUCAo0eP0tPTg0KhIJ1Oo1arqdVqmM1mZmdn0Wq1BAIBlpeXMZlM5PN5\njEYjPp+PZDJJrVbDarVSKpVQKBSEw2GsVivr6+vU63VMJhNOp5NSqYRMJqNUKuHxeJDJZBQKBR5/\n/HFuuukmVlZWgPOrMOnnTqcTgGq1yurqKjabDbPZzMzMDC6XC41GQ61Ww2Qykc1mMRqNZLNZXC4X\nxWIRr9fL2bNnxfc7OjoAyOfzpFIprFYrcrkcv9/PzMwMpVIJi8WCx+PhwQcf5MCBA6TTaSqVCt3d\n3eL3yWQyRKNRvF4vfr+fSqXC6uoqHR0d5PN5qtUqMzMzOBwO6vU6CoUCl8tFo9GgXC6jUqmIRCLo\ndDrcbrf4rCqVCuVyGY1Gg0KhwG63Uy6XaTabOJ1O6vU6v/71r3nrW99KtVqlu7ubI0eOoFKpaLVa\nzM7OYjKZUKlU6HQ6duzYwaFDh3A6nbjdblQqFTKZjHQ6jdVqJRaLkUqlGB4eplgsis8uk8mgUqmo\nVqvodDoSiQTNZhOz2Ux3dzdHjx6l0Whgt9vR6XTY7XZuvfXWzbytn9e9/XKgv7+fgwcPcv3113Py\n5MktuYY/hV27dvHzn/+c7u7uTTleoVDAYDBsyrGeD3bs2IHT6eTgwYMv+7lfKDa1tbJ3717m5uZY\nXl6mWq3yk5/8hBtuuOFPvrZYLFIsFsUWNRqN0tPTw+rqKlNTU8RiMer1OoVCAblczvj4OMPDw2g0\nGvx+Pz6fD6VSSbVaRaPRYLfbMRgMWK1WisWieHJls1kCgQDhcBifz4dWqwWgvb2dWq3GE088QavV\nYn5+nmazSS6Xw2q14vF4uOSSS4hGo9TrdeLxOIlEgpmZGWKxGG1tbcjlckwmE729veTzeRqNBjab\njZWVFU6dOoVaraarq4vZ2VnOnDnD2NgY5XIZuVzO8vIyMzMz7Nixg6GhIXK5HFqtlp/97GfodDoc\nDge7d+8mGo2i1+uRyWRsbGyIz+7o0aPMzc1htVpJpVKsra0xOzvLysoKBoOB48ePUywWqVaraLVa\nHA4HAwMDjI+PE4/Hcbvd2Gw2Go0GcrmcYDDI9PQ0q6urtLW14XK5uOaaa9Dr9RiNRvHfwMAAwWCQ\nY8eOodVqqdfrHDp0iIceegiZTIbD4WBtbY3BwUH0ej1XXHEFO3fu5ODBg0xMTBAOh5mfn6derzM5\nOcni4iKTk5Ok02kikQjz8/NUq1VCoRA6nY62tjay2SxyuZxwOEwymcRisdBsNnn00Ufxer3I5XJq\ntRqJRIJDhw5t5i39gu7tlxpzc3N87Wtf4+jRo9hsti25hj+FycnJTT+myWTa9GP+Ndxzzz2vqCIO\nm1zIlUol3/nOd3jd617HyMgIN91005+c6stkMux2O0ajkZ6eHtrb27HZbFQqFfr6+ti/fz9tbW1E\no1GUSiWpVAqtVksulyMYDDI7O8vCwgIKhYLOzk7i8TgymYx6vU6tViOfz6PVamk0GhgMBqLRKMPD\nwyQSCTY2NjCZTKIIXn755cD5FVej0WB5eZm1tTUWFhbIZrPi/Y1Gg0ajQWdnJ61WC7PZzNraGjab\njfn5eWq1Gmq1mieffJJGo4Hf70elUnHw4EHxgHryySfJZDJ4vV5GR0fp7OzkzJkzNBoNVCoVsViM\nG264QazuM5kMpVKJ5eVl4vE4O3fupLu7m1arRWdnJ729veKzDAQCpNPpZzxsHA4HrVaLeDyOUqkk\nEolw7tw5Dhw4QCQSwWw2E4/HcTqdXHvttajVaoaHh1lZWaFUKrG+vk57ezuVSgVA7AB27NhBT08P\nzWaTQqFAb28vGo0Gm81GtVqlo6MDo9GIxWIhkUgwOTnJ6OgoZ8+eJRAIYLPZaDabXHzxxezevZtA\nIIBGo6Gvr0/sAHp7ewmFQsRiMYrFonjQAZw8eRKVSoXNZqNWq1GpVGhra8NisTAwMLCZt/Tzvrdf\nLnz3u9/lS1/6EsFgUNzDFwKMRiP79u3blGP993//N/fdd9+mHOu54i1vecsrcoi9qfTD53RCmYwP\nfOADuFwu+vv7WVhYIJlMks/nGRoaAs6vonU6HalUCo1Gg0ajYX19HbvdTqVSQavVotVqqVarmEwm\ncrkc5XIZt9tNMBgkm80yMDBANBpFrVZjNpsBCIfD2Gw2SqUStVoNm82G1WplcnIShUJBq9Wivb0d\nu91OOp2mUCigUqmYnp5Go9Hg9XqpVCpMT08zOjqKQqEgEongdruRyWR0dHRw7Ngx6vU6Xq+XYrGI\nXq9HrVaTSCQYGhoSLaK2tjbcbjdGo5F6vU44HCaVSmG321EoFGQyGfR6PdFoVJyjXq8zODhIMpkE\nwGKxEIvFkMvl1Ot16vU61WpVtBlUKhXBYBCj0UipVEKn04l2zYkTJ/D5fLRaLbRaLdlsFo1GQ7lc\nFivgVqtFNpvF7XaTTCYpl8sMDg6iVqs5e/YsqVSK8fFxWq0W5XKZWCwmWikKhYJYLIbFYgHA7XZz\n4sQJBgYGSCQSWCwWUqkUDocDuVxOoVCg0WjgdDrR6/UANJtN8vk8pVIJo9HIkSNHxAO5VCqxf/9+\nlpeXUSqVlEolms0mJpOJr3zlK5veI3+u9/bLjZtvvpmvfvWrHD9+nDe/+c00m82X/Rr+GB/84AcZ\nHR3lE5/4xKYc76GHHuKaa67ZlGM9FywuLtLT0/Oyne/54s/d11ui7BwcHCSTyfDkk0+SSqWQyWTc\neOONxONx0Q+VyWS43W70ej2HDh0ShTmTyZDNZkkmk2i1WqLRKHK5HKvVikajoaOjg87OTsxmM61W\ni0AgwPHjx1lcXKRcLot2gN1uF6tZh8PBzp07cbvdDA0NcezYMZ588klOnz7N1NQUGo2GN7zhDRiN\nRjweD9dffz2ZTIbOzk66u7upVCp0dXURi8WoVCoMDAyg1Wqx2+2Mj4+j0+nYu3cvkUgEm82G0WhE\noVCwsLDAI488wtLSErlcDo/HQ7lcZm5ujtXVVfR6PW63W6yss9ksMplMzBD8fj/FYpF8Pg+ATqdj\n9+7dot89NzfHxMQErVZLFMdMJkMymWT37t0UCgWy2SyNRgOfz8fs7Kw47/j4OAaDgf3799PX1yeK\ntlar5amnnkKr1bJ79270er14mIRCIbq6ugBIp9OEw2GcTidarZbl5WXa2toolUrU63VKpRJOp5N8\nPs/Kyoroh2azWRwOB8FgkIWFBQBisRjRaBSTyUQwGBTb7WPHjlGr1QiFQvT29oqH3/8l3HPPPWKX\nVqvVeOCBB7b0eh566KFNPd5VV13Fxz72sU095t8itmRF/pWvfAW5XE6z2WR9fR2LxcLKygrt7e20\nWi08Hg+1Wg2lUolMJuPMmTM4
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"prompt_number": 128
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "heading",
|
||
|
|
"level": 1,
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"Homework"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {},
|
||
|
|
"source": [
|
||
|
|
"(1) Construct an edge detector by applying a smoothing filter followed by a `trivedges`-style derivative and appropriate thresholding. For smoothing filters, try at least `uniform_filter` and `gaussian_filter`. For the given default noise levels in `mondrian` and `blobworld`, find the optimal parameters for both filters in terms of f-measure by exploring the parameter space. To do this, generate a fixed list of test cases, perform edge detection with different parameters, and evaluate the f-measure for all of them and average the f-measure (additional question: is averaging the right thing to combine f-measures?).\n",
|
||
|
|
"\n",
|
||
|
|
"(2) Add hysteresis thresholding to the edge detector and try different hysteresis thresholding parameters; can you improve the f-measure?"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"collapsed": false,
|
||
|
|
"input": [],
|
||
|
|
"language": "python",
|
||
|
|
"metadata": {},
|
||
|
|
"outputs": []
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"metadata": {}
|
||
|
|
}
|
||
|
|
]
|
||
|
|
}
|