two-rectangles-or-triangles.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:25:52.986000",
     "start_time": "2016-07-06T23:25:52.454000"
    },
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:25:58.288000",
     "start_time": "2016-07-06T23:25:54.225000"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((50000L, 16L, 16L), (50000L, 2L, 4L))"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_imgs = 50000\n",
    "\n",
    "img_size = 16\n",
    "min_rect_size = 3\n",
    "max_rect_size = 8\n",
    "num_objects = 2\n",
    "\n",
    "bboxes = np.zeros((num_imgs, num_objects, 4))\n",
    "imgs = np.zeros((num_imgs, img_size, img_size))\n",
    "shapes = np.zeros((num_imgs, num_objects, 1))\n",
    "\n",
    "for i_img in range(num_imgs):\n",
    "    for i_object in range(num_objects):\n",
    "        if np.random.choice([True, False]):\n",
    "            width, height = np.random.randint(min_rect_size, max_rect_size, size=2)\n",
    "            x = np.random.randint(0, img_size - width)\n",
    "            y = np.random.randint(0, img_size - height)\n",
    "            imgs[i_img, x:x+width, y:y+height] = 1.\n",
    "            bboxes[i_img, i_object] = [x, y, width, height]\n",
    "            shapes[i_img, i_object] = [0]\n",
    "        else:\n",
    "            size = np.random.randint(min_rect_size, max_rect_size)\n",
    "            x, y = np.random.randint(0, img_size - size, size=2)\n",
    "            mask = np.tril_indices(size)\n",
    "            imgs[i_img, x + mask[0], y + mask[1]] = 1.\n",
    "            bboxes[i_img, i_object] = [x, y, size, size]\n",
    "            shapes[i_img, i_object] = [1]\n",
    "        \n",
    "imgs.shape, bboxes.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:25:58.544000",
     "start_time": "2016-07-06T23:25:58.291000"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQQAAAEACAYAAABVmQgcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nA0FS8/8gQbUm/IBsdwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7547ac8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "i = 0\n",
    "# TODO: Why does the array have to be transposed?\n",
    "plt.imshow(imgs[i].T, cmap='Greys', interpolation='none', origin='lower', extent=[0, img_size, 0, img_size])\n",
    "for bbox, shape in zip(bboxes[i], shapes[i]):\n",
    "    plt.gca().add_patch(matplotlib.patches.Rectangle((bbox[0], bbox[1]), bbox[2], bbox[3], ec='r' if shape[0] == 0 else 'y', fc='none'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:25:59.719000",
     "start_time": "2016-07-06T23:25:59.203000"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((50000L, 256L), -4.0834557957225569e-17, 1.0000000000000016)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X = (imgs.reshape(num_imgs, -1) - np.mean(imgs)) / np.std(imgs)\n",
    "X.shape, np.mean(X), np.std(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:26:00.476000",
     "start_time": "2016-07-06T23:26:00.464000"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(50000L, 10L)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# TODO: We use binary classification here - for multiple classes, convert classes to one-hot vectors.\n",
    "y = np.concatenate([bboxes / img_size, shapes], axis=-1).reshape(num_imgs, -1)\n",
    "y.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:26:01.233000",
     "start_time": "2016-07-06T23:26:01.226000"
    },
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "i = int(0.8 * num_imgs)\n",
    "train_X = X[:i]\n",
    "test_X = X[i:]\n",
    "train_y = y[:i]\n",
    "test_y = y[i:]\n",
    "test_imgs = imgs[i:]\n",
    "test_bboxes = bboxes[i:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:26:06.005000",
     "start_time": "2016-07-06T23:26:02.162000"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using Theano backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Activation, Dropout\n",
    "from keras.optimizers import SGD\n",
    "model = Sequential([\n",
    "        Dense(256, input_dim=X.shape[-1]), \n",
    "        Activation('relu'), \n",
    "        Dropout(0.4), \n",
    "        Dense(y.shape[-1])\n",
    "    ])\n",
    "model.compile('adadelta', 'mse')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:32:54.867000",
     "start_time": "2016-07-06T23:26:06.009000"
    },
    "code_folding": [
     2,
     17
    ]
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "5s - loss: 0.1568 - val_loss: 0.0559\n",
      "Flipped 19200.0 training samples (48.0 %)\n",
      "Mean IOU: 0.242985636173\n",
      "Mean dist: 0.197557964071\n",
      "\n",
      "Epoch 1\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "5s - loss: 0.0557 - val_loss: 0.0501\n",
      "Flipped 10275.0 training samples (25.6875 %)\n",
      "Mean IOU: 0.279607567133\n",
      "Mean dist: 0.180107625927\n",
      "\n",
      "Epoch 2\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "5s - loss: 0.0492 - val_loss: 0.0484\n",
      "Flipped 7191.0 training samples (17.9775 %)\n",
      "Mean IOU: 0.319445985247\n",
      "Mean dist: 0.160972388048\n",
      "\n",
      "Epoch 3\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0442 - val_loss: 0.0493\n",
      "Flipped 4605.0 training samples (11.5125 %)\n",
      "Mean IOU: 0.362110983162\n",
      "Mean dist: 0.142787179212\n",
      "\n",
      "Epoch 4\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0397 - val_loss: 0.0514\n",
      "Flipped 2731.0 training samples (6.8275 %)\n",
      "Mean IOU: 0.398454722176\n",
      "Mean dist: 0.126876902916\n",
      "\n",
      "Epoch 5\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0365 - val_loss: 0.0521\n",
      "Flipped 2014.0 training samples (5.035 %)\n",
      "Mean IOU: 0.42149196072\n",
      "Mean dist: 0.118305101417\n",
      "\n",
      "Epoch 6\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0342 - val_loss: 0.0532\n",
      "Flipped 1499.0 training samples (3.7475 %)\n",
      "Mean IOU: 0.438339288177\n",
      "Mean dist: 0.111296744747\n",
      "\n",
      "Epoch 7\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0322 - val_loss: 0.0542\n",
      "Flipped 1367.0 training samples (3.4175 %)\n",
      "Mean IOU: 0.458791583308\n",
      "Mean dist: 0.104734359959\n",
      "\n",
      "Epoch 8\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0309 - val_loss: 0.0538\n",
      "Flipped 1145.0 training samples (2.8625 %)\n",
      "Mean IOU: 0.460957146178\n",
      "Mean dist: 0.103087726083\n",
      "\n",
      "Epoch 9\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "5s - loss: 0.0298 - val_loss: 0.0544\n",
      "Flipped 1071.0 training samples (2.6775 %)\n",
      "Mean IOU: 0.472695257841\n",
      "Mean dist: 0.0997398846543\n",
      "\n",
      "Epoch 10\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0289 - val_loss: 0.0546\n",
      "Flipped 982.0 training samples (2.455 %)\n",
      "Mean IOU: 0.475498423504\n",
      "Mean dist: 0.0987868952204\n",
      "\n",
      "Epoch 11\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0281 - val_loss: 0.0547\n",
      "Flipped 923.0 training samples (2.3075 %)\n",
      "Mean IOU: 0.488416947873\n",
      "Mean dist: 0.0947622980352\n",
      "\n",
      "Epoch 12\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0273 - val_loss: 0.0546\n",
      "Flipped 836.0 training samples (2.09 %)\n",
      "Mean IOU: 0.48874462737\n",
      "Mean dist: 0.0939458890792\n",
      "\n",
      "Epoch 13\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0266 - val_loss: 0.0552\n",
      "Flipped 854.0 training samples (2.135 %)\n",
      "Mean IOU: 0.492589727087\n",
      "Mean dist: 0.0929735561502\n",
      "\n",
      "Epoch 14\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0260 - val_loss: 0.0555\n",
      "Flipped 840.0 training samples (2.1 %)\n",
      "Mean IOU: 0.49697117486\n",
      "Mean dist: 0.0918255879526\n",
      "\n",
      "Epoch 15\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0257 - val_loss: 0.0549\n",
      "Flipped 938.0 training samples (2.345 %)\n",
      "Mean IOU: 0.498299255798\n",
      "Mean dist: 0.0906791499878\n",
      "\n",
      "Epoch 16\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0254 - val_loss: 0.0554\n",
      "Flipped 864.0 training samples (2.16 %)\n",
      "Mean IOU: 0.502686649386\n",
      "Mean dist: 0.0897506966331\n",
      "\n",
      "Epoch 17\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0250 - val_loss: 0.0558\n",
      "Flipped 919.0 training samples (2.2975 %)\n",
      "Mean IOU: 0.501087879733\n",
      "Mean dist: 0.0898817176063\n",
      "\n",
      "Epoch 18\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0245 - val_loss: 0.0553\n",
      "Flipped 789.0 training samples (1.9725 %)\n",
      "Mean IOU: 0.510024924944\n",
      "Mean dist: 0.0866388306772\n",
      "\n",
      "Epoch 19\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0243 - val_loss: 0.0554\n",
      "Flipped 816.0 training samples (2.04 %)\n",
      "Mean IOU: 0.509864361663\n",
      "Mean dist: 0.0867276995739\n",
      "\n",
      "Epoch 20\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0240 - val_loss: 0.0556\n",
      "Flipped 801.0 training samples (2.0025 %)\n",
      "Mean IOU: 0.512294895195\n",
      "Mean dist: 0.0862244741229\n",
      "\n",
      "Epoch 21\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0236 - val_loss: 0.0555\n",
      "Flipped 745.0 training samples (1.8625 %)\n",
      "Mean IOU: 0.515253308823\n",
      "Mean dist: 0.0849313586484\n",
      "\n",
      "Epoch 22\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0235 - val_loss: 0.0552\n",
      "Flipped 847.0 training samples (2.1175 %)\n",
      "Mean IOU: 0.509704473702\n",
      "Mean dist: 0.0860384929421\n",
      "\n",
      "Epoch 23\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0231 - val_loss: 0.0554\n",
      "Flipped 805.0 training samples (2.0125 %)\n",
      "Mean IOU: 0.522583586843\n",
      "Mean dist: 0.082750892523\n",
      "\n",
      "Epoch 24\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0231 - val_loss: 0.0559\n",
      "Flipped 795.0 training samples (1.9875 %)\n",
      "Mean IOU: 0.5220165657\n",
      "Mean dist: 0.0833447329007\n",
      "\n",
      "Epoch 25\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "5s - loss: 0.0228 - val_loss: 0.0554\n",
      "Flipped 692.0 training samples (1.73 %)\n",
      "Mean IOU: 0.518472561734\n",
      "Mean dist: 0.0832177979367\n",
      "\n",
      "Epoch 26\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0227 - val_loss: 0.0560\n",
      "Flipped 715.0 training samples (1.7875 %)\n",
      "Mean IOU: 0.531976024594\n",
      "Mean dist: 0.0807062015938\n",
      "\n",
      "Epoch 27\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0225 - val_loss: 0.0558\n",
      "Flipped 755.0 training samples (1.8875 %)\n",
      "Mean IOU: 0.527110370286\n",
      "Mean dist: 0.0818592475274\n",
      "\n",
      "Epoch 28\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0223 - val_loss: 0.0558\n",
      "Flipped 664.0 training samples (1.66 %)\n",
      "Mean IOU: 0.524141956699\n",
      "Mean dist: 0.0824798966059\n",
      "\n",
      "Epoch 29\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0220 - val_loss: 0.0563\n",
      "Flipped 833.0 training samples (2.0825 %)\n",
      "Mean IOU: 0.531040570366\n",
      "Mean dist: 0.0800875488942\n",
      "\n",
      "Epoch 30\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0219 - val_loss: 0.0564\n",
      "Flipped 856.0 training samples (2.14 %)\n",
      "Mean IOU: 0.532365466129\n",
      "Mean dist: 0.0798171989575\n",
      "\n",
      "Epoch 31\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0218 - val_loss: 0.0560\n",
      "Flipped 666.0 training samples (1.665 %)\n",
      "Mean IOU: 0.532150241524\n",
      "Mean dist: 0.0805836183106\n",
      "\n",
      "Epoch 32\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0216 - val_loss: 0.0562\n",
      "Flipped 750.0 training samples (1.875 %)\n",
      "Mean IOU: 0.53260218024\n",
      "Mean dist: 0.0799956398766\n",
      "\n",
      "Epoch 33\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0216 - val_loss: 0.0563\n",
      "Flipped 751.0 training samples (1.8775 %)\n",
      "Mean IOU: 0.537927557793\n",
      "Mean dist: 0.0789594690559\n",
      "\n",
      "Epoch 34\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0215 - val_loss: 0.0558\n",
      "Flipped 774.0 training samples (1.935 %)\n",
      "Mean IOU: 0.525038210403\n",
      "Mean dist: 0.0818134408655\n",
      "\n",
      "Epoch 35\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0213 - val_loss: 0.0561\n",
      "Flipped 777.0 training samples (1.9425 %)\n",
      "Mean IOU: 0.539060265778\n",
      "Mean dist: 0.0774311544654\n",
      "\n",
      "Epoch 36\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0213 - val_loss: 0.0563\n",
      "Flipped 742.0 training samples (1.855 %)\n",
      "Mean IOU: 0.540322599552\n",
      "Mean dist: 0.0776645600821\n",
      "\n",
      "Epoch 37\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0211 - val_loss: 0.0564\n",
      "Flipped 771.0 training samples (1.9275 %)\n",
      "Mean IOU: 0.533745474506\n",
      "Mean dist: 0.0788337795559\n",
      "\n",
      "Epoch 38\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0210 - val_loss: 0.0559\n",
      "Flipped 710.0 training samples (1.775 %)\n",
      "Mean IOU: 0.537828049212\n",
      "Mean dist: 0.0781276750452\n",
      "\n",
      "Epoch 39\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0210 - val_loss: 0.0565\n",
      "Flipped 666.0 training samples (1.665 %)\n",
      "Mean IOU: 0.537717195805\n",
      "Mean dist: 0.0780453172558\n",
      "\n",
      "Epoch 40\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0208 - val_loss: 0.0562\n",
      "Flipped 765.0 training samples (1.9125 %)\n",
      "Mean IOU: 0.536087355366\n",
      "Mean dist: 0.0787861236678\n",
      "\n",
      "Epoch 41\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0205 - val_loss: 0.0567\n",
      "Flipped 674.0 training samples (1.685 %)\n",
      "Mean IOU: 0.544342451278\n",
      "Mean dist: 0.0764891585959\n",
      "\n",
      "Epoch 42\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0206 - val_loss: 0.0561\n",
      "Flipped 696.0 training samples (1.74 %)\n",
      "Mean IOU: 0.538981326683\n",
      "Mean dist: 0.0777417385232\n",
      "\n",
      "Epoch 43\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0205 - val_loss: 0.0564\n",
      "Flipped 648.0 training samples (1.62 %)\n",
      "Mean IOU: 0.539015426899\n",
      "Mean dist: 0.0771807135874\n",
      "\n",
      "Epoch 44\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0204 - val_loss: 0.0566\n",
      "Flipped 627.0 training samples (1.5675 %)\n",
      "Mean IOU: 0.544229123534\n",
      "Mean dist: 0.0759150881751\n",
      "\n",
      "Epoch 45\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0204 - val_loss: 0.0564\n",
      "Flipped 596.0 training samples (1.49 %)\n",
      "Mean IOU: 0.545409298404\n",
      "Mean dist: 0.0765501037944\n",
      "\n",
      "Epoch 46\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0204 - val_loss: 0.0560\n",
      "Flipped 631.0 training samples (1.5775 %)\n",
      "Mean IOU: 0.542919962167\n",
      "Mean dist: 0.0766910073689\n",
      "\n",
      "Epoch 47\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0201 - val_loss: 0.0567\n",
      "Flipped 759.0 training samples (1.8975 %)\n",
      "Mean IOU: 0.543450512949\n",
      "Mean dist: 0.0763397818658\n",
      "\n",
      "Epoch 48\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0202 - val_loss: 0.0570\n",
      "Flipped 701.0 training samples (1.7525 %)\n",
      "Mean IOU: 0.545041588618\n",
      "Mean dist: 0.0763403655117\n",
      "\n",
      "Epoch 49\n",
      "Train on 40000 samples, validate on 10000 samples\n",
      "Epoch 1/1\n",
      "4s - loss: 0.0201 - val_loss: 0.0566\n",
      "Flipped 566.0 training samples (1.415 %)\n",
      "Mean IOU: 0.548117733988\n",
      "Mean dist: 0.0750292489711\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Flip bboxes during training.\n",
    "# Note: The validation loss is always quite big here because we don't flip the bounding boxes for the validation data. \n",
    "def IOU(bbox1, bbox2):\n",
    "    '''Calculate overlap between two bounding boxes [x, y, w, h] as the area of intersection over the area of unity'''\n",
    "    x1, y1, w1, h1 = bbox1[0], bbox1[1], bbox1[2], bbox1[3]  # TODO: Check if its more performant if tensor elements are accessed directly below.\n",
    "    x2, y2, w2, h2 = bbox2[0], bbox2[1], bbox2[2], bbox2[3]\n",
    "\n",
    "    w_I = min(x1 + w1, x2 + w2) - max(x1, x2)\n",
    "    h_I = min(y1 + h1, y2 + h2) - max(y1, y2)\n",
    "    w_I = max(w_I, 0)  # set w_I and h_I zero if there is no intersection\n",
    "    h_I = max(h_I, 0)\n",
    "    I = w_I * h_I\n",
    "\n",
    "    U = w1 * h1 + w2 * h2 - I\n",
    "\n",
    "    return I / U\n",
    "\n",
    "def distance(bbox1, bbox2):\n",
    "    return np.sqrt(np.sum(np.square(bbox1[:2] - bbox2[:2])))\n",
    "\n",
    "num_epochs = 50\n",
    "flipped_train_y = np.array(train_y)\n",
    "flipped = np.zeros((len(flipped_train_y), num_epochs))\n",
    "ious = np.zeros((len(flipped_train_y), num_epochs))\n",
    "dists = np.zeros((len(flipped_train_y), num_epochs))\n",
    "\n",
    "for epoch in range(num_epochs):\n",
    "    print 'Epoch', epoch\n",
    "    model.fit(train_X, flipped_train_y, nb_epoch=1, validation_data=(test_X, test_y), verbose=2)\n",
    "    pred_y = model.predict(train_X)\n",
    "\n",
    "    for i, (pred_bboxes, exp_bboxes) in enumerate(zip(pred_y, flipped_train_y)):\n",
    "        \n",
    "        flipped_exp_bboxes = np.concatenate([exp_bboxes[4:], exp_bboxes[:4]])\n",
    "        \n",
    "        mse = np.mean(np.square(pred_bboxes - exp_bboxes))\n",
    "        mse_flipped = np.mean(np.square(pred_bboxes - flipped_exp_bboxes))\n",
    "        \n",
    "        iou = IOU(pred_bboxes[:4], exp_bboxes[:4]) + IOU(pred_bboxes[4:], exp_bboxes[4:])\n",
    "        iou_flipped = IOU(pred_bboxes[:4], flipped_exp_bboxes[:4]) + IOU(pred_bboxes[4:], flipped_exp_bboxes[4:])\n",
    "        \n",
    "        dist = distance(pred_bboxes[:4], exp_bboxes[:4]) + IOU(pred_bboxes[4:], exp_bboxes[4:])\n",
    "        dist_flipped = distance(pred_bboxes[:4], flipped_exp_bboxes[:4]) + IOU(pred_bboxes[4:], flipped_exp_bboxes[4:])\n",
    "        \n",
    "        if mse_flipped < mse:  # using iou or dist here leads to similar results\n",
    "            flipped_train_y[i] = flipped_exp_bboxes\n",
    "            flipped[i, epoch] = 1\n",
    "            ious[i, epoch] = iou_flipped / 2.\n",
    "            dists[i, epoch] = dist_flipped / 2.\n",
    "        else:\n",
    "            ious[i, epoch] = iou / 2.\n",
    "            dists[i, epoch] = dist / 2.\n",
    "            \n",
    "    print 'Flipped {} training samples ({} %)'.format(np.sum(flipped[:, epoch]), np.mean(flipped[:, epoch]) * 100.)\n",
    "    print 'Mean IOU: {}'.format(np.mean(ious[:, epoch]))\n",
    "    print 'Mean dist: {}'.format(np.mean(dists[:, epoch]))\n",
    "    print"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:55:00.367000",
     "start_time": "2016-07-06T23:54:57.699000"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.text.Text at 0x1ee69c18>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZAAAAEPCAYAAABsj5JaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1edc4208>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.pcolor(flipped[:1000], cmap='Greys')\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Training sample')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:55:01.541000",
     "start_time": "2016-07-06T23:55:01.245000"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0, 1)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEACAYAAABI5zaHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1ede4390>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(np.mean(ious, axis=0), label='Mean IOU')  # between predicted and assigned true bboxes\n",
    "plt.plot(np.mean(dists, axis=0), label='Mean distance')  # relative to image size\n",
    "plt.legend()\n",
    "plt.ylim(0, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:55:02.905000",
     "start_time": "2016-07-06T23:55:02.569000"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((10000L, 2L, 4L), (10000L, 2L, 1L))"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_y = model.predict(test_X)\n",
    "pred_y = pred_y.reshape(len(pred_y), num_objects, -1)\n",
    "pred_bboxes = pred_y[..., :4] * img_size\n",
    "pred_shapes = pred_y[..., 4:5]\n",
    "pred_bboxes.shape, pred_shapes.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2016-07-06T23:57:24.479000",
     "start_time": "2016-07-06T23:57:22.606000"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA6IAAADjCAYAAACfFKnTAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x257bab00>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(16, 8))\n",
    "for i_subplot in range(1, 5):\n",
    "    plt.subplot(1, 4, i_subplot)\n",
    "    i = np.random.randint(len(test_X))\n",
    "    plt.imshow(test_imgs[i].T, cmap='Greys', interpolation='none', origin='lower', extent=[0, img_size, 0, img_size])\n",
    "    for pred_bbox, exp_bbox, pred_shape in zip(pred_bboxes[i], test_bboxes[i], pred_shapes[i]):\n",
    "        plt.gca().add_patch(matplotlib.patches.Rectangle((pred_bbox[0], pred_bbox[1]), pred_bbox[2], pred_bbox[3], ec='r' if pred_shape[0] <= 0.5 else 'y', fc='none'))\n",
    "        # TODO: Calculate max IOU with all expected bounding boxes.\n",
    "#         plt.annotate('IOU: {:.2f}'.format(IOU(pred_bbox, exp_bbox)), (pred_bbox[0], pred_bbox[1]+pred_bbox[3]+0.4), color='r')\n",
    "\n",
    "# plt.savefig('plots/bw-two-rectangles-or-triangles4.png', dpi=300)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.14"
  },
  "toc": {
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}