Zero_DCE.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "view-in-github"
   },
   "source": [
    "<a href=\"https://colab.research.google.com/github/talha1503/Zero-DCE/blob/main/Zero_DCE.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 55
    },
    "id": "mXJnm1p973T_",
    "outputId": "79aabdca-895a-4e38-813e-24109121e01b"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
     ]
    }
   ],
   "source": [
    "from google.colab import drive\n",
    "drive.mount('/content/drive')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "id": "7LXtV_kiOGmq"
   },
   "outputs": [],
   "source": [
    "import numpy as np \n",
    "import pandas as pd\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torchvision\n",
    "import torchvision.transforms as transforms\n",
    "import matplotlib.pyplot as plt\n",
    "import os\n",
    "from PIL import Image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {
    "id": "z15vsnKnO5Hh"
   },
   "outputs": [],
   "source": [
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else 'cpu')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {
    "id": "eOrxEjP8XBEB"
   },
   "outputs": [],
   "source": [
    "IMG_SHAPE = (256,256)\n",
    "TRAIN_BATCH_SIZE = 8\n",
    "VALID_BATCH_SIZE = 4\n",
    "EPOCHS = 35\n",
    "LEARNING_RATE = 1e-3  \n",
    "COLOR_LOSS_WEIGHT = 5\n",
    "SMOOTHNESS_LOSS_WEIGHT = 20  \n",
    "TRAIN_VAL_SPLIT = 0.95  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 71
    },
    "id": "aeUsBeQ5SnGO",
    "outputId": "c48d92cb-d192-45a0-932c-bb4f145fa36c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total images:  2003\n",
      "Validation Data:  101\n",
      "Training Data:  1902\n"
     ]
    }
   ],
   "source": [
    "train_data_transform = transforms.Compose([transforms.Resize(IMG_SHAPE),transforms.ToTensor()])\n",
    "valid_data_transform = transforms.Compose([transforms.Resize(IMG_SHAPE),transforms.ToTensor()])\n",
    "\n",
    "data_link = '/content/drive/My Drive/train_data_Zero-DCE/train_data'\n",
    "total_image_list = os.listdir(data_link)\n",
    "\n",
    "shuffle = np.random.permutation(len(total_image_list))\n",
    "split_values = int(len(total_image_list) * TRAIN_VAL_SPLIT)\n",
    "\n",
    "print(\"Total images: \",len(total_image_list))\n",
    "print(\"Validation Data: \",len(shuffle[split_values:]))\n",
    "print(\"Training Data: \",len(shuffle[:split_values]))\n",
    "\n",
    "class TrainDataset(torch.utils.data.Dataset):\n",
    "  def __init__(self,image_shape,transforms):\n",
    "    self.image_list = np.array([total_image_list[i] for i in shuffle[:split_values]])\n",
    "    self.image_shape = image_shape \n",
    "    self.transform = transforms\n",
    "\n",
    "  def __len__(self):\n",
    "    return len(self.image_list) \n",
    "\n",
    "  def __getitem__(self,index):\n",
    "    image = self.image_list[index]\n",
    "    image = Image.open(data_link + '/' + image)\n",
    "    transformed_image = self.transform(image)\n",
    "    normalized_image = transformed_image/255\n",
    "    return normalized_image\n",
    "\n",
    "class ValidDataset(torch.utils.data.Dataset):\n",
    "  def __init__(self,image_shape,transforms):\n",
    "    self.image_list = np.array([total_image_list[i] for i in shuffle[split_values:]])\n",
    "    self.image_shape = image_shape \n",
    "    self.transform = transforms\n",
    "\n",
    "  def __len__(self):\n",
    "    return len(self.image_list) \n",
    "\n",
    "  def __getitem__(self,index):\n",
    "    image = self.image_list[index]\n",
    "    image = Image.open(data_link + '/' + image)\n",
    "    transformed_image = self.transform(image)\n",
    "    normalized_image = transformed_image/255\n",
    "    return normalized_image\n",
    "  \n",
    "train_dataset = TrainDataset(IMG_SHAPE,train_data_transform)\n",
    "train_loader = torch.utils.data.DataLoader(train_dataset,batch_size = TRAIN_BATCH_SIZE,shuffle=True)\n",
    "valid_dataset = ValidDataset(IMG_SHAPE,valid_data_transform)\n",
    "valid_loader = torch.utils.data.DataLoader(valid_dataset,batch_size = VALID_BATCH_SIZE,shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {
    "id": "i8eCFCfUUYLm"
   },
   "outputs": [],
   "source": [
    "class DCENet(nn.Module):\n",
    "  def __init__(self):  # Input ->  256x256x3\n",
    "    super(DCENet, self).__init__()\n",
    "    self.conv1 = nn.Conv2d(in_channels = 3 , out_channels = 32, kernel_size = 3, stride = 1, padding = 1,bias = True)\n",
    "    self.conv2 = nn.Conv2d(in_channels = 32 , out_channels = 32, kernel_size = 3, stride = 1, padding = 1,bias = True)\n",
    "    self.conv3 = nn.Conv2d(in_channels = 32 , out_channels = 32, kernel_size = 3, stride = 1, padding = 1,bias = True)\n",
    "    self.conv4 = nn.Conv2d(in_channels = 32 , out_channels = 32, kernel_size = 3, stride = 1, padding = 1,bias = True) \n",
    "    self.conv5 = nn.Conv2d(in_channels = 64 , out_channels = 32, kernel_size = 3, stride = 1, padding = 1,bias = True)\n",
    "    self.conv6 = nn.Conv2d(in_channels = 64 , out_channels = 32, kernel_size = 3, stride = 1, padding = 1,bias = True)\n",
    "    self.conv7 = nn.Conv2d(in_channels = 64 , out_channels = 24, kernel_size = 3, stride = 1, padding = 1,bias = True)\n",
    "    \n",
    "\n",
    "  def forward(self,input):\n",
    "    output1 = F.relu(self.conv1(input))\n",
    "    output2 = F.relu(self.conv2(output1))  \n",
    "    output3 = F.relu(self.conv3(output2))\n",
    "    output4 = F.relu(self.conv4(output3))\n",
    "    output5 = F.relu(self.conv5(torch.cat([output3,output4],1)))   \n",
    "    output6 = F.relu(self.conv6(torch.cat([output2,output5],1)))   \n",
    "    output7 = F.tanh(self.conv7(torch.cat([output1,output6],1)))  # Output here is 256x256x24.   \n",
    "\n",
    "    LE_curves = list(torch.split(output7,3,dim=1))  \n",
    "    for i in range(8):\n",
    "        input = input + LE_curves[i]*(torch.pow(input,2)-input)\n",
    "\n",
    "    LE_curves = torch.cat(LE_curves,1)\n",
    "\n",
    "    return input,LE_curves"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {
    "id": "X6NMXNjnUfMP"
   },
   "outputs": [],
   "source": [
    "class SpatialConsistencyLoss(nn.Module):\n",
    "  def __init__(self):\n",
    "    super(SpatialConsistencyLoss, self).__init__()\n",
    "    self.left_kernel = torch.FloatTensor([[0,0,0],[-1,1,0],[0,0,0]]).unsqueeze(0).unsqueeze(0)\n",
    "    self.top_kernel = torch.FloatTensor([[0,-1,0],[0,1,0],[0,0,0]]).unsqueeze(0).unsqueeze(0)\n",
    "    self.right_kernel = torch.FloatTensor([[0,0,0],[0,1,-1],[0,0,0]]).unsqueeze(0).unsqueeze(0)\n",
    "    self.bottom_kernel = torch.FloatTensor([[0,0,0],[0,1,0],[0,-1,0]]).unsqueeze(0).unsqueeze(0) \n",
    "    self.left_kernel = self.left_kernel.to(device)\n",
    "    self.top_kernel = self.top_kernel.to(device)\n",
    "    self.right_kernel = self.right_kernel.to(device)\n",
    "    self.bottom_kernel = self.bottom_kernel.to(device)\n",
    "\n",
    "  def forward(self,input_image,enhanced_image):\n",
    "    mean_input = torch.mean(input_image,dim=1,keepdims=True)\n",
    "    mean_enhanced = torch.mean(enhanced_image,dim=1,keepdims=True)    \n",
    "    pool_input = F.avg_pool2d(input = mean_input,kernel_size = 4)\n",
    "    pool_enhanced = F.avg_pool2d(input = mean_enhanced, kernel_size = 4)\n",
    "    left_input = F.conv2d(pool_input,self.left_kernel,padding = 1)\n",
    "    right_input = F.conv2d(pool_input,self.right_kernel,padding = 1)\n",
    "    top_input = F.conv2d(pool_input,self.top_kernel,padding = 1)\n",
    "    bottom_input = F.conv2d(pool_input,self.bottom_kernel,padding = 1)\n",
    "    left_enhanced = F.conv2d(pool_enhanced,self.left_kernel,padding = 1)\n",
    "    right_enhanced = F.conv2d(pool_enhanced,self.right_kernel,padding = 1)\n",
    "    top_enhanced = F.conv2d(pool_enhanced,self.top_kernel,padding = 1)\n",
    "    bottom_enhanced = F.conv2d(pool_enhanced,self.bottom_kernel,padding = 1)\n",
    "    return torch.mean(torch.pow(left_enhanced-left_input,2) + torch.pow(right_enhanced-right_input,2) + torch.pow(top_enhanced-top_input,2) + torch.pow(bottom_enhanced-bottom_input,2))\n",
    "\n",
    "class ExposureControlLoss(nn.Module):\n",
    "  def __init__(self,E=0.6):\n",
    "    super(ExposureControlLoss,self).__init__() \n",
    "    self.E = E\n",
    "\n",
    "  def forward(self,enhanced_image):\n",
    "    enhanced_image = torch.mean(enhanced_image,1,keepdims=True)\n",
    "    val = F.avg_pool2d(enhanced_image,kernel_size = 16,stride = 16)  \n",
    "    # E_matrix  = self.E*torch.ones(val.shape,device = device) # My equation\n",
    "    return torch.mean(torch.pow(val - torch.FloatTensor([self.E]).to(device),2))   # Author's equation.\n",
    "\n",
    "class ColorConstancyLoss(nn.Module):\n",
    "  def __init__(self,colour_weight = COLOR_LOSS_WEIGHT):\n",
    "    super(ColorConstancyLoss,self).__init__()\n",
    "    self.colour_weight = colour_weight\n",
    "\n",
    "  def forward(self,enhanced_image): \n",
    "    mean_enhanced = torch.mean(enhanced_image,[2,3],keepdims=True)\n",
    "    splits = torch.split(mean_enhanced,1,dim=1)\n",
    "    R = splits[0]\n",
    "    G = splits[1]\n",
    "    B = splits[2]\n",
    "    # total = torch.pow(R-G,2) + torch.pow(B-G,2) + torch.pow(B-R,2) # My equation\n",
    "    total = torch.pow(torch.pow(torch.pow(R-G,2),2) + torch.pow(torch.pow(R-B,2),2) + torch.pow(torch.pow(B-G,2),2),0.5) # Author's implementation varies from the paper.\n",
    "    return self.colour_weight*torch.mean(total) \n",
    "\n",
    "\n",
    "class IlluminationSmoothnessLoss(nn.Module):\n",
    "  def __init__(self,smoothness = SMOOTHNESS_LOSS_WEIGHT,):\n",
    "    super(IlluminationSmoothnessLoss,self).__init__()\n",
    "#     self.smoothness = smoothness\n",
    "\n",
    "  def forward(self,curves):\n",
    "    horizontal_gradients = torch.pow(curves[:,:,1:,:] - curves[:,:,:curves.shape[2]-1,:],2).sum() \n",
    "    vertical_gradients = torch.pow(curves[:,:,:,1:] - curves[:,:,:,:curves.shape[3]-1],2).sum()\n",
    "    # total = (vertical_gradients + horizontal_gradients)/8   # My equation\n",
    "    total = 2*(vertical_gradients/ (curves.shape[2] * (curves.shape[3] - 1)) + (horizontal_gradients/(curves.shape[2]-1) * curves.shape[3]))/curves.shape[0]\n",
    "    return total\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {
    "id": "jJuR7DAu8alU"
   },
   "outputs": [],
   "source": [
    "model = DCENet()\n",
    "model = model.to(device)\n",
    "optimizer = torch.optim.SGD(model.parameters(),lr=LEARNING_RATE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {
    "id": "zWa5435aVHGk"
   },
   "outputs": [],
   "source": [
    "def train(train_iterator,valid_iterator,epochs,model,optimizer): \n",
    "  spatial_loss = SpatialConsistencyLoss()\n",
    "  exposure_loss = ExposureControlLoss()\n",
    "  colour_loss = ColorConstancyLoss()\n",
    "  illumination_loss = IlluminationSmoothnessLoss()\n",
    "\n",
    "  train_loss_list = []\n",
    "  valid_loss_list = []\n",
    "\n",
    "  for epoch in range(epochs):\n",
    "    train_epoch_loss = 0\n",
    "    val_epoch_loss = 0\n",
    "    for train_iteration,image in enumerate(train_iterator):\n",
    "      model.train()\n",
    "      optimizer.zero_grad()\n",
    "      image = image.to(device)\n",
    "      output,curves = model(image)\n",
    "      \n",
    "      loss = spatial_loss(image,output)  + 10*exposure_loss(output)  + colour_loss(output) + illumination_loss(curves)\n",
    "      loss.backward()\n",
    "      optimizer.step()\n",
    "      train_epoch_loss += loss.item()\n",
    "\n",
    "    for val_iteration,image in enumerate(valid_iterator):\n",
    "      model.eval()\n",
    "      image = image.to(device)\n",
    "      output,curves = model(image)\n",
    "      val_loss = spatial_loss(image,output) + 10*exposure_loss(output) + colour_loss(output) + 200*illumination_loss(curves)\n",
    "      val_epoch_loss += val_loss.item() \n",
    "\n",
    "    train_loss_list.append(train_epoch_loss)\n",
    "    valid_loss_list.append(val_epoch_loss)\n",
    "    print(\"-\"*50)\n",
    "    print(\"Train Loss at epoch:{} is {} \".format(epoch,train_epoch_loss/len(train_iterator)))\n",
    "    print(\"Valid Loss at epoch:{} is {} \".format(epoch,val_epoch_loss/len(valid_iterator)))\n",
    "    print(\"-\"*50)\n",
    "  \n",
    "\n",
    "  return model,train_loss_list,valid_loss_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "id": "Lg5JXfBr-J4F",
    "outputId": "4236c411-af64-4f96-b987-ef0fb17dacc7"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/torch/nn/functional.py:1614: UserWarning: nn.functional.tanh is deprecated. Use torch.tanh instead.\n",
      "  warnings.warn(\"nn.functional.tanh is deprecated. Use torch.tanh instead.\")\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--------------------------------------------------\n",
      "Train Loss at epoch:0 is 3.643341083486541 \n",
      "Valid Loss at epoch:0 is 3.585906505584717 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:1 is 3.5913184049750577 \n",
      "Valid Loss at epoch:1 is 3.5923867225646973 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:2 is 3.5882208617795417 \n",
      "Valid Loss at epoch:2 is 3.5917270183563232 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:3 is 3.5862167352387884 \n",
      "Valid Loss at epoch:3 is 3.5856595039367676 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:4 is 3.5847244673416396 \n",
      "Valid Loss at epoch:4 is 3.586408853530884 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:5 is 3.5835198915305257 \n",
      "Valid Loss at epoch:5 is 3.5892088413238525 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:6 is 3.5824742287146947 \n",
      "Valid Loss at epoch:6 is 3.5742509365081787 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:7 is 3.5815564804718276 \n",
      "Valid Loss at epoch:7 is 3.5658719539642334 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:8 is 3.580728652096596 \n",
      "Valid Loss at epoch:8 is 3.5792651176452637 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:9 is 3.5799618939391706 \n",
      "Valid Loss at epoch:9 is 3.581206798553467 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:10 is 3.5792221441990186 \n",
      "Valid Loss at epoch:10 is 3.570855140686035 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:11 is 3.578534729340497 \n",
      "Valid Loss at epoch:11 is 3.5800418853759766 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:12 is 3.5778425721561207 \n",
      "Valid Loss at epoch:12 is 3.5728397369384766 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:13 is 3.5771590912041544 \n",
      "Valid Loss at epoch:13 is 3.563950777053833 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:14 is 3.576487766594446 \n",
      "Valid Loss at epoch:14 is 3.5648210048675537 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:15 is 3.575825332593517 \n",
      "Valid Loss at epoch:15 is 3.5746617317199707 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:16 is 3.575157577250184 \n",
      "Valid Loss at epoch:16 is 3.5844080448150635 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:17 is 3.5744630248606706 \n",
      "Valid Loss at epoch:17 is 3.5742228031158447 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:18 is 3.573759083988286 \n",
      "Valid Loss at epoch:18 is 3.5709328651428223 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:19 is 3.573048055672846 \n",
      "Valid Loss at epoch:19 is 3.57619047164917 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:20 is 3.572320715719912 \n",
      "Valid Loss at epoch:20 is 3.5826542377471924 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:21 is 3.571555655543544 \n",
      "Valid Loss at epoch:21 is 3.5751428604125977 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:22 is 3.570762008178134 \n",
      "Valid Loss at epoch:22 is 3.5649609565734863 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:23 is 3.5699570128897657 \n",
      "Valid Loss at epoch:23 is 3.5696053504943848 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:24 is 3.5690976511530517 \n",
      "Valid Loss at epoch:24 is 3.552560567855835 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:25 is 3.5682493197817764 \n",
      "Valid Loss at epoch:25 is 3.582019329071045 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:26 is 3.5673166012563624 \n",
      "Valid Loss at epoch:26 is 3.565725803375244 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:27 is 3.566350995993414 \n",
      "Valid Loss at epoch:27 is 3.5623974800109863 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:29 is 3.564273924386802 \n",
      "Valid Loss at epoch:29 is 3.5544660091400146 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:30 is 3.5631581805333368 \n",
      "Valid Loss at epoch:30 is 3.558523654937744 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:31 is 3.5619968416310157 \n",
      "Valid Loss at epoch:31 is 3.5787439346313477 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:32 is 3.5607361122339714 \n",
      "Valid Loss at epoch:32 is 3.5755369663238525 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:33 is 3.5593819427890936 \n",
      "Valid Loss at epoch:33 is 3.5604095458984375 \n",
      "--------------------------------------------------\n",
      "--------------------------------------------------\n",
      "Train Loss at epoch:34 is 3.5579476035943554 \n",
      "Valid Loss at epoch:34 is 3.557718276977539 \n",
      "--------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "model,train_loss_list,valid_loss_list = train(train_loader,valid_loader,EPOCHS,model,optimizer)\n",
    "torch.save(model,'/content/drive/My Drive/Zero-DCEModels/model_15.pth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {
    "id": "bxAHECyZVIPe"
   },
   "outputs": [],
   "source": [
    "def plot_list(loss_list):\n",
    "  plt.plot(loss_list)\n",
    "  plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 265
    },
    "id": "MrDLisIsFYMB",
    "outputId": "9d91bdc7-1305-4c13-be77-81e4d6d5ddd2"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_list(train_loss_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 265
    },
    "id": "7tUtCCPMFbxF",
    "outputId": "6c3f26e7-2e81-47c3-94a8-e2f97cb00ca7"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_list(valid_loss_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 73
    },
    "id": "dx7rWOzoiIam",
    "outputId": "978527e7-3b32-4298-d2dc-c69d52f33da5"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/dist-packages/torch/nn/functional.py:1614: UserWarning: nn.functional.tanh is deprecated. Use torch.tanh instead.\n",
      "  warnings.warn(\"nn.functional.tanh is deprecated. Use torch.tanh instead.\")\n"
     ]
    }
   ],
   "source": [
    "test_data_transform = transforms.Compose([transforms.Resize(IMG_SHAPE),transforms.ToTensor()])\n",
    "save_directory = '/content/drive/My Drive/ZeroDCEResult/'\n",
    "\n",
    "class TestDataset(torch.utils.data.Dataset):\n",
    "  def __init__(self,image_shape,transforms,image_path, image_directory):\n",
    "    self.image_list = [image_path]\n",
    "    self.image_shape = image_shape \n",
    "    self.transform = transforms\n",
    "    self.image_directory = image_directory\n",
    "\n",
    "  def __len__(self):\n",
    "    return len(self.image_list) \n",
    "\n",
    "  def __getitem__(self,index):\n",
    "    image = self.image_list[index]\n",
    "    image = Image.open(self.image_directory + '/' + image)\n",
    "    transformed_image = self.transform(image)\n",
    "    normalized_image = transformed_image/255\n",
    "    return normalized_image\n",
    "\n",
    "def eval(image_name,image_directory,model_path):\n",
    "  test_dataset = TestDataset(IMG_SHAPE,test_data_transform,image_name, image_directory)\n",
    "  test_loader = torch.utils.data.DataLoader(test_dataset,batch_size = 1,shuffle=True)\n",
    "  image = next(iter(test_loader))\n",
    "  image = image.to(device)\n",
    "  model = torch.load(model_path)\n",
    "  model.to(device)\n",
    "  output,curves = model(image)\n",
    "  output = output*255\n",
    "  output = output[0]\n",
    "\n",
    "  save_transform = transforms.Compose([\n",
    "    transforms.ToPILImage(),\n",
    "    transforms.Resize(size=256),\n",
    "    transforms.ToTensor()\n",
    "  ])\n",
    "\n",
    "  torchvision.utils.save_image(output,save_directory+image_name)\n",
    "\n",
    "for i in range(1,11):\n",
    "  eval('LIME/{}.bmp'.format(i),'/content/drive/My Drive/ZeroDCEtest/','/content/drive/My Drive/Zero-DCEModels/model_15.pth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "id": "ArWcK-r0KqWh"
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "include_colab_link": true,
   "name": "Zero-DCE",
   "provenance": []
  },
  "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": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}