SRGAN.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "196b528c",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np \n",
    "import torch\n",
    "import matplotlib.pyplot as plt \n",
    "import torch.nn as nn \n",
    "import torchvision.transforms as transforms\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "import os\n",
    "from PIL import Image\n",
    "\n",
    "import glob"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "87cfc872",
   "metadata": {},
   "outputs": [],
   "source": [
    "device = torch.device(\"cuda\") if torch.cuda.is_available() else torch.device(\"cpu\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "c2769dac",
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_DIR = os.path.join(os.getcwd(),\"data/img\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bb8c4fe5",
   "metadata": {},
   "source": [
    "### Hyper parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "90da425e",
   "metadata": {},
   "outputs": [],
   "source": [
    "hr_size = 96\n",
    "r = 4\n",
    "beta1 = 0.9\n",
    "lr = 10e-4\n",
    "batch_size = 32\n",
    "epochs = 20"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "943bcbd8",
   "metadata": {},
   "source": [
    "### Custom dataset \n",
    "\n",
    "#### At each batch iteration, we should be able to retrieve the HR image and the LR image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "b4a4e46b",
   "metadata": {},
   "outputs": [],
   "source": [
    "class SRGANDataset(Dataset): \n",
    "    \n",
    "    def __init__(self, folder, hr_size, stride=4): \n",
    "        self.folder = folder \n",
    "        self.hr_size = folder\n",
    "        self.stride = stride\n",
    "        \n",
    "        self.hr_transform = transforms.Compose([\n",
    "            transforms.Resize((hr_size, hr_size)), \n",
    "            transforms.ToTensor(), \n",
    "            transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n",
    "        ])\n",
    "        \n",
    "        self.lr_transform = transforms.Compose([\n",
    "            transforms.Resize((hr_size//stride, hr_size//stride)), \n",
    "            transforms.ToTensor(), \n",
    "            transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n",
    "        ])\n",
    "        \n",
    "        self.files = sorted(glob.glob(folder + \"/*.*\"))\n",
    "        \n",
    "    def __getitem__(self, idx): \n",
    "        \n",
    "        img_name = self.files[idx] \n",
    "        image_obj = Image.open(img_name)\n",
    "        \n",
    "        return {\n",
    "            \"hr\" : self.hr_transform(image_obj), \n",
    "            \"lr\" : self.lr_transform(image_obj)\n",
    "        }\n",
    "    \n",
    "    def __len__(self): \n",
    "        return len(self.files)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "16c9f1ae",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = SRGANDataset(DATA_DIR, hr_size)\n",
    "data_loader = torch.utils.data.DataLoader(data, batch_size=batch_size, shuffle=True, num_workers=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "200db235",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_both(hr_img, lr_img): \n",
    "    hr_img = np.transpose(hr_img, (1, 2, 0)) \n",
    "    lr_img = np.transpose(lr_img, (1, 2, 0))\n",
    "    \n",
    "    \n",
    "    f, ax = plt.subplots(1,2)\n",
    "    \n",
    "    ax[0].imshow(hr_img)\n",
    "    ax[1].imshow(lr_img)\n",
    "    \n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "5b2c3b43",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n",
      "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sample = data[0]\n",
    "plot_both(sample[\"hr\"], sample[\"lr\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "be01aa11",
   "metadata": {},
   "source": [
    "### Implementing the generator model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "1bf13a65",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResidualBlock(nn.Module): \n",
    "    \n",
    "    def __init__(self, in_c, kernel_size=3): \n",
    "        super(ResidualBlock, self).__init__()\n",
    "        self.in_c = in_c\n",
    "        self.kernel_size = kernel_size\n",
    "        self.conv1 = nn.Conv2d(in_c, in_c, kernel_size, padding=1)\n",
    "        self.bn1 = nn.BatchNorm2d(in_c)\n",
    "        self.prelu = nn.PReLU() \n",
    "        self.conv2 = nn.Conv2d(in_c, in_c, kernel_size, padding=1) \n",
    "        self.bn2 = nn.BatchNorm2d(in_c)\n",
    "    \n",
    "    def forward(self, x): \n",
    "        identity = x \n",
    "        x = self.conv1(x) \n",
    "        x = self.bn1(x) \n",
    "        x = self.prelu(x) \n",
    "        x = self.conv2(x)\n",
    "        x = self.bn2(x)\n",
    "        return x + identity\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "dd8d8ca6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ResidualBlock(\n",
       "  (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (prelu): PReLU(num_parameters=1)\n",
       "  (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "  (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       ")"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from torchsummary import summary\n",
    "\n",
    "resblock = ResidualBlock(64)\n",
    "resblock.to(device)\n",
    "#summary(resblock, (64, 224, 224), device=\"cuda\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "0f7f9093",
   "metadata": {},
   "outputs": [],
   "source": [
    "class UpSamplingBlock(nn.Module): \n",
    "    \n",
    "    def __init__(self, in_c, out_c, kernel_size=3): \n",
    "        super(UpSamplingBlock, self).__init__()\n",
    "        self.in_c = in_c\n",
    "        self.out_c = out_c\n",
    "        \n",
    "        self.conv1 = nn.Conv2d(in_c, out_c, kernel_size) \n",
    "        # important : pixelshuffle also divide the nb for channels \n",
    "        # by 4 !!\n",
    "        self.upsample = nn.PixelShuffle(2) # upsampling factor\n",
    "        self.prelu = nn.PReLU()\n",
    "        \n",
    "    def forward(self, x): \n",
    "        \n",
    "        x = self.conv1(x) \n",
    "        x = self.upsample(x) \n",
    "        x = self.prelu(x) \n",
    "        \n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "2ff5abcb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "UpSamplingBlock(\n",
       "  (conv1): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1))\n",
       "  (upsample): PixelShuffle(upscale_factor=2)\n",
       "  (prelu): PReLU(num_parameters=1)\n",
       ")"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "upsamplingblock = UpSamplingBlock(64, 256) \n",
    "upsamplingblock.to(device)\n",
    "#summary(upsamplingblock, (64, 32, 32), device=\"cuda\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "8dcbdb94",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Generator(nn.Module): \n",
    "    \n",
    "    def __init__(self, in_c, out_c): \n",
    "        super(Generator, self).__init__()\n",
    "        self.in_c = in_c \n",
    "        self.out_c = out_c\n",
    "        \n",
    "        self.conv1 = nn.Conv2d(3, 64, 9, padding=4)\n",
    "        self.prelu1 = torch.nn.PReLU() \n",
    "        \n",
    "        residuals = [] \n",
    "        \n",
    "        for i in range(5): \n",
    "            residuals.append(ResidualBlock(64))\n",
    "            \n",
    "        self.residuals = nn.Sequential(*residuals)\n",
    "        \n",
    "        self.conv2 = nn.Conv2d(64, 64, 3, padding=1)\n",
    "        self.bn2 = nn.BatchNorm2d(64)\n",
    "        \n",
    "        upsamplings = []\n",
    "        \n",
    "        for i in range(2): \n",
    "            upsamplings.append(UpSamplingBlock(64, 256))\n",
    "        \n",
    "        self.upsampling = nn.Sequential(*upsamplings)\n",
    "        \n",
    "        self.conv3 = nn.Conv2d(64, 3, kernel_size=3, padding=7)\n",
    "        \n",
    "        \n",
    "    def forward(self, x): \n",
    "        \n",
    "        x = self.conv1(x) \n",
    "        identity = self.prelu1(x) \n",
    "        x = self.residuals(identity) \n",
    "        x = self.conv2(x) \n",
    "        x = self.bn2(x) \n",
    "        # residual \n",
    "        x = x + identity\n",
    "        x = self.upsampling(x)\n",
    "        \n",
    "        x = torch.tanh(self.conv3(x))\n",
    "        \n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "29c95bfc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Generator(\n",
       "  (conv1): Conv2d(3, 64, kernel_size=(9, 9), stride=(1, 1), padding=(4, 4))\n",
       "  (prelu1): PReLU(num_parameters=1)\n",
       "  (residuals): Sequential(\n",
       "    (0): ResidualBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (prelu): PReLU(num_parameters=1)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (1): ResidualBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (prelu): PReLU(num_parameters=1)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (2): ResidualBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (prelu): PReLU(num_parameters=1)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (3): ResidualBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (prelu): PReLU(num_parameters=1)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "    (4): ResidualBlock(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (prelu): PReLU(num_parameters=1)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    )\n",
       "  )\n",
       "  (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "  (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (upsampling): Sequential(\n",
       "    (0): UpSamplingBlock(\n",
       "      (conv1): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1))\n",
       "      (upsample): PixelShuffle(upscale_factor=2)\n",
       "      (prelu): PReLU(num_parameters=1)\n",
       "    )\n",
       "    (1): UpSamplingBlock(\n",
       "      (conv1): Conv2d(64, 256, kernel_size=(3, 3), stride=(1, 1))\n",
       "      (upsample): PixelShuffle(upscale_factor=2)\n",
       "      (prelu): PReLU(num_parameters=1)\n",
       "    )\n",
       "  )\n",
       "  (conv3): Conv2d(64, 3, kernel_size=(3, 3), stride=(1, 1), padding=(7, 7))\n",
       ")"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gnet = Generator(64, 256) \n",
    "gnet.to(device)\n",
    "#summary(gnet, (3, hr_size//4, hr_size//4), device=\"cuda\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "155561f7",
   "metadata": {},
   "outputs": [],
   "source": [
    "class DiscConvBlock(nn.Module): \n",
    "    \n",
    "    def __init__(self, in_c, out_c, stride=1): \n",
    "        super(DiscConvBlock, self).__init__()\n",
    "        self.in_c = in_c\n",
    "        self.stride=stride\n",
    "        self.out_c = out_c\n",
    "        self.conv1 = nn.Conv2d(in_c, out_c, 3, stride=stride)\n",
    "        self.bn = nn.BatchNorm2d(out_c)\n",
    "        self.lrelu = nn.LeakyReLU(0.2)\n",
    "        \n",
    "    def forward(self, x): \n",
    "        x = self.conv1(x) \n",
    "        x = self.bn(x) \n",
    "        x = self.lrelu(x)\n",
    "        \n",
    "        return x\n",
    "\n",
    "class Discriminator(nn.Module): \n",
    "    \n",
    "    def __init__(self, in_c, out_c): \n",
    "        super(Discriminator, self).__init__()\n",
    "        self.conv1 = nn.Conv2d(in_c, 64, 3) \n",
    "        self.lrelu = nn.LeakyReLU(0.2)\n",
    "                \n",
    "        self.convblocks = nn.Sequential(*[\n",
    "            DiscConvBlock(64, 128, 2), \n",
    "            DiscConvBlock(128, 128, 1), \n",
    "            DiscConvBlock(128, 256, 2), \n",
    "            DiscConvBlock(256, 256, 1), \n",
    "            DiscConvBlock(256, 512, 2),\n",
    "            DiscConvBlock(512, 512, 1), \n",
    "        ])\n",
    "        self.fc = nn.Linear(512 * 7 * 7, 1024) \n",
    "        self.lrelu = nn.LeakyReLU(0.2) \n",
    "        self.final = nn.Linear(1024, 1) \n",
    "        self.sigmoid = torch.nn.functional.sigmoid\n",
    "        \n",
    "    def forward(self, x): \n",
    "        x = self.conv1(x) \n",
    "        x = self.lrelu(x)\n",
    "        x = self.convblocks(x) \n",
    "        x = torch.flatten(x, 1)\n",
    "        x = self.fc(x) \n",
    "        x = self.lrelu(x) \n",
    "        x = self.final(x) \n",
    "        x = self.sigmoid(x)\n",
    "        \n",
    "        return x\n",
    "        \n",
    "            \n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "019eb1bc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Discriminator(\n",
       "  (conv1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1))\n",
       "  (lrelu): LeakyReLU(negative_slope=0.2)\n",
       "  (convblocks): Sequential(\n",
       "    (0): DiscConvBlock(\n",
       "      (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2))\n",
       "      (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (lrelu): LeakyReLU(negative_slope=0.2)\n",
       "    )\n",
       "    (1): DiscConvBlock(\n",
       "      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1))\n",
       "      (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (lrelu): LeakyReLU(negative_slope=0.2)\n",
       "    )\n",
       "    (2): DiscConvBlock(\n",
       "      (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2))\n",
       "      (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (lrelu): LeakyReLU(negative_slope=0.2)\n",
       "    )\n",
       "    (3): DiscConvBlock(\n",
       "      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))\n",
       "      (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (lrelu): LeakyReLU(negative_slope=0.2)\n",
       "    )\n",
       "    (4): DiscConvBlock(\n",
       "      (conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2))\n",
       "      (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (lrelu): LeakyReLU(negative_slope=0.2)\n",
       "    )\n",
       "    (5): DiscConvBlock(\n",
       "      (conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1))\n",
       "      (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (lrelu): LeakyReLU(negative_slope=0.2)\n",
       "    )\n",
       "  )\n",
       "  (fc): Linear(in_features=25088, out_features=1024, bias=True)\n",
       "  (final): Linear(in_features=1024, out_features=1, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dnet = Discriminator(3, 512) \n",
    "dnet.to(device)\n",
    "#summary(dnet, (3, hr_size, hr_size), batch_size=batch_size, device=\"cuda\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d57a92c5",
   "metadata": {},
   "source": [
    "### Important the pre-trained vgg16"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "2f0b7803",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torchvision.models as models\n",
    "\n",
    "vgg16 = models.vgg16(pretrained=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "2e562822",
   "metadata": {},
   "outputs": [],
   "source": [
    "class VGGExtractor(nn.Module): \n",
    "    \n",
    "    def __init__(self, vgg):\n",
    "        super(VGGExtractor, self).__init__() \n",
    "        # extract layers\n",
    "        self.vgg = vgg\n",
    "        self.layer19 = list(vgg.children())[0][:36]    \n",
    "        for param in self.layer19.parameters():\n",
    "            param.requires_grad = False\n",
    "        \n",
    "    def forward(self, x):\n",
    "        return self.layer19(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "52437d50",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "VGGExtractor(\n",
       "  (vgg): VGG(\n",
       "    (features): Sequential(\n",
       "      (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (1): ReLU(inplace=True)\n",
       "      (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (3): ReLU(inplace=True)\n",
       "      (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "      (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (6): ReLU(inplace=True)\n",
       "      (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (8): ReLU(inplace=True)\n",
       "      (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "      (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (11): ReLU(inplace=True)\n",
       "      (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (13): ReLU(inplace=True)\n",
       "      (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (15): ReLU(inplace=True)\n",
       "      (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "      (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (18): ReLU(inplace=True)\n",
       "      (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (20): ReLU(inplace=True)\n",
       "      (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (22): ReLU(inplace=True)\n",
       "      (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "      (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (25): ReLU(inplace=True)\n",
       "      (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (27): ReLU(inplace=True)\n",
       "      (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "      (29): ReLU(inplace=True)\n",
       "      (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "    )\n",
       "    (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))\n",
       "    (classifier): Sequential(\n",
       "      (0): Linear(in_features=25088, out_features=4096, bias=True)\n",
       "      (1): ReLU(inplace=True)\n",
       "      (2): Dropout(p=0.5, inplace=False)\n",
       "      (3): Linear(in_features=4096, out_features=4096, bias=True)\n",
       "      (4): ReLU(inplace=True)\n",
       "      (5): Dropout(p=0.5, inplace=False)\n",
       "      (6): Linear(in_features=4096, out_features=1000, bias=True)\n",
       "    )\n",
       "  )\n",
       "  (layer19): Sequential(\n",
       "    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (1): ReLU(inplace=True)\n",
       "    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (3): ReLU(inplace=True)\n",
       "    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (6): ReLU(inplace=True)\n",
       "    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (8): ReLU(inplace=True)\n",
       "    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (11): ReLU(inplace=True)\n",
       "    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (13): ReLU(inplace=True)\n",
       "    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (15): ReLU(inplace=True)\n",
       "    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (18): ReLU(inplace=True)\n",
       "    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (20): ReLU(inplace=True)\n",
       "    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (22): ReLU(inplace=True)\n",
       "    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (25): ReLU(inplace=True)\n",
       "    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (27): ReLU(inplace=True)\n",
       "    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (29): ReLU(inplace=True)\n",
       "    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "feature_extractor = VGGExtractor(vgg16)\n",
    "feature_extractor.to(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "068597bb",
   "metadata": {},
   "outputs": [],
   "source": [
    "class GContentLoss(nn.Module): \n",
    "    \n",
    "    def __init__(self, loss=\"MSE\"): \n",
    "        # extractor \n",
    "        super(GContentLoss, self).__init__()\n",
    "        self.extractor = VGGExtractor(vgg16)\n",
    "        if loss == \"MSE\": \n",
    "            self.loss = torch.nn.MSELoss()\n",
    "        elif loss == \"l1\": \n",
    "            self.loss = torch.nn.L1Loss()\n",
    "        \n",
    "    def forward(self, hr, lr): \n",
    "                \n",
    "        hr_feat = self.extractor(hr)\n",
    "        lr_feat = self.extractor(lr)\n",
    "        \n",
    "        return self.loss(hr_feat, lr_feat)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "a061480d",
   "metadata": {},
   "outputs": [],
   "source": [
    "#loss = GContentLoss()\n",
    "#loss.to(device)\n",
    "\n",
    "#x = torch.randn((1, 3,hr_size, hr_size), device=device)\n",
    "#y = torch.randn((1, 3,hr_size, hr_size), device=device)\n",
    "#loss(x,y)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e4036a8a",
   "metadata": {},
   "source": [
    "### Setting up training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "ec3d74d3",
   "metadata": {},
   "outputs": [],
   "source": [
    "g_optim = torch.optim.Adam(gnet.parameters(), lr=lr, betas=(beta1, 0.999))\n",
    "d_optim = torch.optim.Adam(dnet.parameters(), lr=lr, betas=(beta1, 0.999))\n",
    "\n",
    "\n",
    "bce_criterion = torch.nn.BCELoss()\n",
    "content_criterion = GContentLoss()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "734ad662",
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.tensorboard import SummaryWriter\n",
    "\n",
    "\n",
    "writer = SummaryWriter()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "14534468",
   "metadata": {},
   "outputs": [],
   "source": [
    "def show(img): \n",
    "    \"\"\"\n",
    "        from torch to np\n",
    "    \"\"\"\n",
    "    img = img.cpu().detach().numpy()\n",
    "    img = np.transpose(img, (1,2,0))\n",
    "    plt.imshow(img)\n",
    "    return plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fd1ec250",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# training loop \n",
    "torch.autograd.set_detect_anomaly(True)\n",
    "dnet.train()\n",
    "gnet.train()\n",
    "\n",
    "for epoch in range(epochs): \n",
    "    \n",
    "    for i, sample in enumerate(data_loader): \n",
    "        hr, lr = sample[\"hr\"], sample[\"lr\"] \n",
    "        hr = hr.to(device)\n",
    "        lr = lr.to(device)\n",
    "        # generate synthetic images\n",
    "        generated_sr = gnet(lr)\n",
    "        true_preds = dnet(hr)\n",
    "        # labels | we will replace the value inside after\n",
    "        true_labels = torch.full(size=(batch_size,1), fill_value=0, device=device, dtype=torch.float32)\n",
    "        \n",
    "        \n",
    "        ################# TRAIN GENERATOR ####################\n",
    "        for p in dnet.parameters():\n",
    "            p.requires_grad = False\n",
    "            \n",
    "        # zero grad \n",
    "        # tricking the discriminator\n",
    "        dg_out = dnet(generated_sr)\n",
    "        # compute total loss regarding the generator\n",
    "        g_bce_loss = bce_criterion(dg_out, true_preds)\n",
    "        # content loss\n",
    "        content_loss = content_criterion(generated_sr, hr)\n",
    "        \n",
    "        # update parameters \n",
    "        g_loss = content_loss + g_bce_loss \n",
    "        g_optim.zero_grad()\n",
    "        g_loss.backward()\n",
    "        g_optim.step()\n",
    "        \n",
    "        \n",
    "        ################# TRAIN DISCRIMINATOR ###############\n",
    "        for p in dnet.parameters():\n",
    "            p.requires_grad = True\n",
    "        # zero grad \n",
    "        true_loss = bce_criterion(true_preds, true_labels)        \n",
    "        fake_labels = true_labels.fill_(0) \n",
    "        fake_preds = dnet(generated_sr.detach())\n",
    "        fake_loss = bce_criterion(fake_preds, fake_labels)\n",
    "        \n",
    "        # update parameters \n",
    "        d_loss = fake_loss + true_loss \n",
    "        d_optim.zero_grad()\n",
    "        d_loss = fake_loss + true_loss\n",
    "        d_optim.step()\n",
    "        \n",
    "        \n",
    "        if i % 20 == 0: \n",
    "            writer.add_scalar(\"Loss/discriminator\", d_loss, i)\n",
    "            writer.add_scalar(\"Loss/generator\", g_loss, i)\n",
    "            print(f\"EPOCH={epoch} | BATCH={i} | GLOSS={g_loss} | DLOSS={d_loss}\")\n",
    "            \n",
    "        if i % 100 == 0: \n",
    "            with torch.no_grad(): \n",
    "                downsampled = lr[0]\n",
    "                generated = generated_sr[0]\n",
    "                gt = hr[0]\n",
    "                \n",
    "                show(downsampled)\n",
    "                show(generated) \n",
    "                show(gt)\n",
    "        \n",
    "        \n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ebd4e093",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b4eb2915",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "058c0e14",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c76f02e3",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}