notebooks/6.2-6.3 CNN_1.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "c07ae157",
   "metadata": {},
   "source": [
    "# [6.2 Convolutions for Images](https://d2l.ai/chapter_convolutional-neural-networks/conv-layer.html)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5b5115e8",
   "metadata": {},
   "source": [
    "<img src='./images/conv_operation.png' width='800'/>\n",
    "\n",
    "&emsp; <sub>  * Image Credit : https://www.kaggle.com/manmohan291/57-cnn-basics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ddd5f165",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "from torchvision import datasets, transforms\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "plt.rc('font', size=16)\n",
    "\n",
    "%matplotlib inline\n",
    "%config InlineBackend.figure_format = 'retina'\n",
    "\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "device\n",
    "\n",
    "# load FashionMNIST training data\n",
    "fmnist_train = datasets.FashionMNIST(\n",
    "    root=\"../../../data\", train=True, transform=transforms.ToTensor(), download=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "df2ab7e9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 936x288 with 5 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 174,
       "width": 739
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# display FashionMNIST images\n",
    "labels_map = {\n",
    "    0: \"T-Shirt\",\n",
    "    1: \"Trouser\",\n",
    "    2: \"Pullover\",\n",
    "    3: \"Dress\",\n",
    "    4: \"Coat\",\n",
    "    5: \"Sandal\",\n",
    "    6: \"Shirt\",\n",
    "    7: \"Sneaker\",\n",
    "    8: \"Bag\",\n",
    "    9: \"Ankle Boot\",\n",
    "}\n",
    "\n",
    "fig, axes = plt.subplots(1, 5, figsize=(13, 4))\n",
    "plt.subplots_adjust(hspace=0.001, wspace=0.001)\n",
    "\n",
    "for i in range(5):\n",
    "    image, label = fmnist_train.__getitem__(i)\n",
    "    axes[i].imshow(image.squeeze().numpy())\n",
    "    axes[i].set_title(f'{label} : {labels_map[label]}')\n",
    "    axes[i].axis('off')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "313fbc57",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "id": "18bcffe0",
   "metadata": {},
   "source": [
    "# Convolution of Images\n",
    "\n",
    "- **Feature Map** : The feature map on CNN is the output of one filter applied to the previous layer.\n",
    "\n",
    "- **Receptive Field** : All the elements (from all the previous layers) that may affect the calculation of a pixel.\n",
    "\n",
    "<img src='./images/receptive_field.png' width='600'/>\n",
    "\n",
    "&emsp; <sub>  * Image Credit : https://www.baeldung.com/cs/cnn-receptive-field-size\n",
    "\n",
    "    \n",
    "- Convolution layer in PyTorch: \n",
    "```\n",
    "nn.Conv2d(\n",
    "    in_channels: int,\n",
    "    out_channels: int,\n",
    "    kernel_size: Union[int, Tuple[int, int]],\n",
    "    stride: Union[int, Tuple[int, int]] = 1,\n",
    "    padding: Union[int, Tuple[int, int]] = 0,\n",
    "    dilation: Union[int, Tuple[int, int]] = 1,\n",
    "    groups: int = 1,\n",
    "    bias: bool = True,\n",
    "    padding_mode: str = 'zeros',\n",
    ")\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1882ab8d",
   "metadata": {},
   "source": [
    "### Vertical Edge detection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "d78d2e77",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      ">> randomly init conv2D weight:\n",
      " tensor([[[[ 0.0267,  0.0846, -0.2162],\n",
      "          [ 0.0993, -0.2918,  0.2586],\n",
      "          [-0.0733, -0.0782, -0.2924]]]])\n",
      ">> filled conv2D weight:\n",
      " tensor([[[[-1.,  2., -1.],\n",
      "          [-1.,  2., -1.],\n",
      "          [-1.,  2., -1.]]]])\n"
     ]
    }
   ],
   "source": [
    "conv2d = nn.Conv2d(1, 1, kernel_size=(3, 3), bias=False)\n",
    "print('>> randomly init conv2D weight:\\n', conv2d.weight.data)\n",
    "\n",
    "# filled_weight = torch.tensor([[1., 0., -1.], \n",
    "#                               [1., 0., -1.], \n",
    "#                               [1., 0., -1.]]).view(1, 1, 3,3)\n",
    "\n",
    "filled_weight = torch.tensor([[-1., 2., -1.], \n",
    "                              [-1., 2., -1.], \n",
    "                              [-1., 2., -1.]]).view(1, 1, 3,3)\n",
    "\n",
    "conv2d.weight.data = filled_weight\n",
    "\n",
    "print('>> filled conv2D weight:\\n', conv2d.weight.data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "967d1901",
   "metadata": {},
   "outputs": [],
   "source": [
    "imageID = 1\n",
    "X, _ = fmnist_train.__getitem__(imageID)\n",
    "outX = conv2d(X.unsqueeze(dim=0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "d1dd23e1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 247,
       "width": 455
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axes = plt.subplots(1, 2, figsize=(8, 4))\n",
    "plt.subplots_adjust(hspace=0.001, wspace=0.001)\n",
    "axes[0].imshow(X.squeeze().numpy())\n",
    "axes[1].imshow(outX.squeeze().detach().numpy())\n",
    "\n",
    "axes[0].set_title('original')\n",
    "axes[1].set_title('feature map')\n",
    "\n",
    "for i in range(2):\n",
    "    axes[i].axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "28f3ed91",
   "metadata": {},
   "source": [
    "### Horizontal Edge detection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "c8a6416a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      ">> filled conv2D weight:\n",
      " tensor([[[[-1., -1., -1.],\n",
      "          [ 2.,  2.,  2.],\n",
      "          [-1., -1., -1.]]]])\n"
     ]
    }
   ],
   "source": [
    "# filled_weight = torch.tensor([[ 1.,  1.,  1.], \n",
    "#                               [ 0.,  0.,  0.], \n",
    "#                               [-1., -1., -1.]]).view(1, 1, 3,3)\n",
    "\n",
    "filled_weight = torch.tensor([[-1., -1., -1.], \n",
    "                              [ 2.,  2.,  2.], \n",
    "                              [-1., -1., -1.]]).view(1, 1, 3,3)\n",
    "\n",
    "conv2d.weight.data = filled_weight\n",
    "\n",
    "print('>> filled conv2D weight:\\n', conv2d.weight.data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "8e19ede4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 247,
       "width": 455
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "imageID = 1\n",
    "X, _ = fmnist_train.__getitem__(imageID)\n",
    "outX = conv2d(X.unsqueeze(dim=0))\n",
    "\n",
    "### --- plot --- ###\n",
    "fig, axes = plt.subplots(1, 2, figsize=(8, 4))\n",
    "plt.subplots_adjust(hspace=0.001, wspace=0.001)\n",
    "axes[0].imshow(X.squeeze().numpy())\n",
    "axes[1].imshow(outX.squeeze().detach().numpy())\n",
    "\n",
    "axes[0].set_title('original')\n",
    "axes[1].set_title('feature map')\n",
    "\n",
    "for i in range(2):\n",
    "    axes[i].axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "224ec367",
   "metadata": {},
   "source": [
    "### Edge detection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "a94dc896",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      ">> filled conv2D weight:\n",
      " tensor([[[[-1., -1., -1.],\n",
      "          [-1.,  8., -1.],\n",
      "          [-1., -1., -1.]]]])\n"
     ]
    }
   ],
   "source": [
    "filled_weight = torch.tensor([[-1.0, -1.0, -1.0], \n",
    "                              [-1.0,  8.0, -1.0], \n",
    "                              [-1.0, -1.0, -1.0]]).view(1, 1, 3,3)\n",
    "\n",
    "conv2d.weight.data = filled_weight\n",
    "\n",
    "print('>> filled conv2D weight:\\n', conv2d.weight.data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "eab23ce0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 247,
       "width": 455
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "imageID = 1\n",
    "X, _ = fmnist_train.__getitem__(imageID)\n",
    "outX = conv2d(X.unsqueeze(dim=0))\n",
    "\n",
    "### --- plot --- ###\n",
    "fig, axes = plt.subplots(1, 2, figsize=(8, 4))\n",
    "plt.subplots_adjust(hspace=0.001, wspace=0.001)\n",
    "axes[0].imshow(X.squeeze().numpy())\n",
    "axes[1].imshow(outX.squeeze().detach().numpy())\n",
    "\n",
    "axes[0].set_title('original')\n",
    "axes[1].set_title('feature map')\n",
    "\n",
    "for i in range(2):\n",
    "    axes[i].axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5f629da6",
   "metadata": {},
   "source": [
    "# [6.3 Padding and Stride](https://d2l.ai/chapter_convolutional-neural-networks/padding-and-strides.html)\n",
    "\n",
    "#### Padding and stride can be used to adjust the dimensionality of the data effectively.\n",
    "\n",
    "## Padding\n",
    "\n",
    "- **n x n** images convolve with a **k x k** kernel → Output dimension = **(n-k+1) x (n-k+1)**\n",
    "- **Padding** can increase the height and width of the output. This is often used to give the output the same height and width as the input.\n",
    "- Padding helps to keep more of the information at the border of an image.\n",
    "- **n x n** image + padding **p**, convolve with a **k x k** kernel → Output dimension = **(n+2p-k+1) x (n+2p-k+1)**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "22141ed1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "input  X.shape   :  torch.Size([1, 28, 28])\n",
      "output outX.shape:  torch.Size([1, 1, 28, 28])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 504x216 with 2 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 192,
       "width": 372
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Check the output dimension with a kernel_size = 3x3, padding = 1 on a 28x28 image\n",
    "\n",
    "conv2d = nn.Conv2d(1, 1, kernel_size=(3, 3), bias=False, padding=1)\n",
    "\n",
    "filled_weight = torch.tensor([[-1.0, -1.0, -1.0], \n",
    "                              [-1.0,  8.0, -1.0], \n",
    "                              [-1.0, -1.0, -1.0]]).view(1, 1, 3, 3)\n",
    "\n",
    "conv2d.weight.data = filled_weight\n",
    "\n",
    "\n",
    "imageID = 1\n",
    "X, _ = fmnist_train.__getitem__(imageID)\n",
    "outX = conv2d(X.unsqueeze(dim=0))\n",
    "\n",
    "print('input  X.shape   : ', X.shape)\n",
    "print('output outX.shape: ', outX.shape)\n",
    "\n",
    "### --- plot --- ###\n",
    "fig, axes = plt.subplots(1, 2, figsize=(7, 3))\n",
    "plt.subplots_adjust(hspace=0.001, wspace=0.001)\n",
    "axes[0].imshow(X.squeeze().numpy())\n",
    "axes[1].imshow(outX.squeeze().detach().numpy())\n",
    "\n",
    "axes[0].set_title('original')\n",
    "axes[1].set_title('feature map')\n",
    "\n",
    "for i in range(2):\n",
    "    axes[i].axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "39dc052c",
   "metadata": {},
   "source": [
    "## Stride\n",
    "\n",
    "- For computational efficiency or because we wish to downsample, we move our window more than one element at a time, skipping the intermediate locations.\n",
    "\n",
    "- **n x n** image + padding **p**, convolve with a **k x k** kernel, stride **s** → Output dimension = **(n+2p-k)/s + 1**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "c75c059b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "input  X.shape   :  torch.Size([1, 28, 28])\n",
      "output outX.shape:  torch.Size([1, 1, 14, 14])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 504x216 with 2 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 192,
       "width": 372
      },
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "conv2d = nn.Conv2d(1, 1, kernel_size=(3, 3), bias=False, padding=1, stride=2)\n",
    "\n",
    "filled_weight = torch.tensor([[-1.0, -1.0, -1.0], \n",
    "                              [-1.0,  8.0, -1.0], \n",
    "                              [-1.0, -1.0, -1.0]]).view(1, 1, 3,3)\n",
    "\n",
    "conv2d.weight.data = filled_weight\n",
    "\n",
    "\n",
    "imageID = 1\n",
    "X, _ = fmnist_train.__getitem__(imageID)\n",
    "outX = conv2d(X.unsqueeze(dim=0))\n",
    "\n",
    "print('input  X.shape   : ', X.shape)\n",
    "print('output outX.shape: ', outX.shape)\n",
    "\n",
    "### --- plot --- ###\n",
    "fig, axes = plt.subplots(1, 2, figsize=(7, 3))\n",
    "plt.subplots_adjust(hspace=0.001, wspace=0.001)\n",
    "axes[0].imshow(X.squeeze().numpy())\n",
    "axes[1].imshow(outX.squeeze().detach().numpy())\n",
    "\n",
    "axes[0].set_title('original')\n",
    "axes[1].set_title('feature map')\n",
    "\n",
    "for i in range(2):\n",
    "    axes[i].axis('off')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2aedfbe6",
   "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}