Single_Rectangle_detection.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "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": 84,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(70000, 8, 8) (70000, 1, 4)\n"
     ]
    }
   ],
   "source": [
    "num_imgs = 70000\n",
    "\n",
    "img_size = 8\n",
    "min_object_size = 1\n",
    "max_object_size = 4\n",
    "num_objects = 1\n",
    "\n",
    "bboxes = np.zeros((num_imgs, num_objects, 4))\n",
    "imgs = np.zeros((num_imgs, img_size, img_size))  # set background to 0\n",
    "for i_img in range(num_imgs):\n",
    "    for i_object in range(num_objects):\n",
    "        w, h = np.random.randint(min_object_size, max_object_size, size=2)\n",
    "        x = np.random.randint(0, img_size - w)\n",
    "        y = np.random.randint(0, img_size - h)\n",
    "        imgs[i_img, x:x+w, y:y+h] = 1.  # set rectangle to 1\n",
    "        bboxes[i_img, i_object] = [x, y, w, h]\n",
    "        \n",
    "print imgs.shape, bboxes.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPgAAAD8CAYAAABaQGkdAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fcbdd035790>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "i = 2\n",
    "plt.imshow(imgs[i].T, cmap='Greys', interpolation='none', origin='lower', extent=[0, img_size, 0, img_size])\n",
    "for bbox in bboxes[i]:\n",
    "    plt.gca().add_patch(matplotlib.patches.Rectangle((bbox[0], bbox[1]), bbox[2], bbox[3], ec='r', fc='none'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((70000, 64), -5.0594449265060702e-18, 0.99999999999999989)"
      ]
     },
     "execution_count": 86,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Reshape and normalize the image data to mean 0 and std 1. \n",
    "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": 87,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((70000, 4), 0.28119241071428569, 0.17528762087406075)"
      ]
     },
     "execution_count": 87,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Normalize x, y, w, h by img_size, so that all values are between 0 and 1.\n",
    "# Important: Do not shift to negative values (e.g. by setting to mean 0), because the IOU calculation needs positive w and h.\n",
    "y = bboxes.reshape(num_imgs, -1) / img_size\n",
    "y.shape, np.mean(y), np.std(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Split training and test.\n",
    "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": 89,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "from torch.autograd import Variable\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [],
   "source": [
    "class LinearRegressionModel(nn.Module):\n",
    "    def __init__(self,insize,ousize):\n",
    "        super(LinearRegressionModel, self).__init__()\n",
    "        self.linear1 = nn.Linear(inp_dim,200)\n",
    "        #self.dropout = nn.Dropout(0.2)\n",
    "        self.linear2 = nn.Linear(200,out_dim)\n",
    "    def forward(self, x):\n",
    "        out = F.relu(self.linear1(x))\n",
    "        #out = self.dropout(out)\n",
    "        out = self.linear2(out)\n",
    "        return out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [],
   "source": [
    "inp_dim=64\n",
    "out_dim=4\n",
    "\n",
    "model=LinearRegressionModel(inp_dim,out_dim)\n",
    "\n",
    "criterion = nn.MSELoss()\n",
    "learning_rate=0.01\n",
    "optimizer= torch.optim.Adam(model.parameters(),lr=learning_rate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 100, loss 8.0657977378e-05\n",
      "epoch 200, loss 6.94841310178e-06\n",
      "epoch 300, loss 9.91367755887e-07\n",
      "epoch 400, loss 1.68999989114e-07\n",
      "epoch 500, loss 3.10694332484e-08\n",
      "epoch 600, loss 5.8100342315e-09\n",
      "epoch 700, loss 1.05185671284e-09\n",
      "epoch 800, loss 1.85037901668e-10\n",
      "epoch 900, loss 2.98142691091e-11\n",
      "epoch 1000, loss 4.25626461159e-12\n"
     ]
    }
   ],
   "source": [
    "loss_list=[]\n",
    "epochs = 1000\n",
    "for epoch in range(epochs):\n",
    "    epoch+=1\n",
    "    \n",
    "    inputs=Variable(torch.FloatTensor(train_X))\n",
    "    labels=Variable(torch.FloatTensor(train_y))\n",
    "    \n",
    "    optimizer.zero_grad()\n",
    "    \n",
    "    outputs = model(inputs)\n",
    "    #print outputs.data.numpy()\n",
    "    #print inputs.data.numpy()\n",
    "    loss=criterion(outputs,labels)\n",
    "    \n",
    "    loss.backward()\n",
    "    \n",
    "    optimizer.step()\n",
    "    if epoch%100==0:\n",
    "        print ('epoch {}, loss {}'.format(epoch,loss.data[0])) \n",
    "    loss_list.append(loss.data[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYQAAAD8CAYAAAB3u9PLAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fcbdd9a4390>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(loss_list)\n",
    "plt.text(900, 0.175, 'Loss', fontdict={'size': 10, 'color':  'green'})\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(14000, 1, 4)"
      ]
     },
     "execution_count": 93,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "inputs=Variable(torch.FloatTensor(test_X))\n",
    "labels=Variable(torch.FloatTensor(test_y))\n",
    "outputs = model(inputs)\n",
    "pred_y = outputs.data.numpy()\n",
    "            \n",
    "pred_bboxes = pred_y * img_size\n",
    "pred_bboxes = pred_bboxes.reshape(len(pred_bboxes), num_objects, -1)\n",
    "pred_bboxes.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "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]\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",
    "    if w_I <= 0 or h_I <= 0:  # no overlap\n",
    "        return 0.\n",
    "    I = w_I * h_I\n",
    "\n",
    "    U = w1 * h1 + w2 * h2 - I\n",
    "\n",
    "    return I / U"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAArwAAAC0CAYAAAB2dv8HAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nDNgqiwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fcbdd6b7c10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Show a few images and predicted bounding boxes from the test dataset. \n",
    "plt.figure(figsize=(12, 3))\n",
    "for i_subplot in range(1, 5):\n",
    "    plt.subplot(1, 4, i_subplot)\n",
    "    i = np.random.randint(len(test_imgs))\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 in zip(pred_bboxes[i], test_bboxes[i]):\n",
    "        plt.gca().add_patch(matplotlib.patches.Rectangle((pred_bbox[0], pred_bbox[1]), pred_bbox[2], pred_bbox[3], ec='r', fc='none'))\n",
    "        plt.gca().add_patch(matplotlib.patches.Rectangle((exp_bbox[0], exp_bbox[1]), exp_bbox[2], exp_bbox[3], color='lightblue', fc='none'))\n",
    "        plt.annotate('IOU: {:.2f}'.format(IOU(pred_bbox, exp_bbox)), (pred_bbox[0], pred_bbox[1]+pred_bbox[3]+0.2), color='r')\n",
    "        \n",
    "# plt.savefig('plots/bw-single-rectangle_prediction.png', dpi=300)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Variable containing:\n",
      " 0.3750\n",
      " 0.6250\n",
      " 0.1250\n",
      " 0.1250\n",
      "[torch.FloatTensor of size 4]\n",
      " Variable containing:\n",
      " 0.3750\n",
      " 0.6250\n",
      " 0.1250\n",
      " 0.1250\n",
      "[torch.FloatTensor of size 4]\n",
      "\n"
     ]
    }
   ],
   "source": [
    "input1=Variable(torch.FloatTensor(test_X[]))\n",
    "labels1=Variable(torch.FloatTensor(test_y[1]))\n",
    "outputs1 = model(input1)\n",
    "print outputs1, labels1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.99998360718840895"
      ]
     },
     "execution_count": 99,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Calculate the mean IOU (overlap) between the predicted and expected bounding boxes on the test dataset. \n",
    "summed_IOU = 0.\n",
    "for pred_bbox, test_bbox in zip(pred_bboxes.reshape(-1, 4), test_bboxes.reshape(-1, 4)):\n",
    "    summed_IOU += IOU(pred_bbox, test_bbox)\n",
    "mean_IOU = summed_IOU / len(pred_bboxes)\n",
    "mean_IOU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "torch.save(model.state_dict(), 'singlebox_model_chk.pt')"
   ]
  }
 ],
 "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": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}