EuroSat_Pytorch.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "EuroSat_Pytorch.ipynb",
      "provenance": [],
      "authorship_tag": "ABX9TyPpNNUeuVZpoIb0zZADLjZ/",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/mugheesahmad/EuroSat-PyTorch/blob/master/EuroSat_Pytorch.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "471-zI0ffkbD",
        "colab_type": "text"
      },
      "source": [
        "**Importing** **Libraries**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2v-oPZk7lwdW",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Importing Libraries\n",
        "from skimage import io\n",
        "import torch\n",
        "from torchvision import models\n",
        "import torchvision\n",
        "import torch.nn as nn\n",
        "import numpy as np\n",
        "import random\n",
        "from IPython.display import clear_output\n",
        "import time\n",
        "import os.path\n",
        "from sklearn.metrics import f1_score\n",
        "import matplotlib.pyplot as plt\n",
        "import torch.optim as optim\n",
        "import torch.optim.lr_scheduler\n",
        "import torch.nn.init\n",
        "from torch.autograd import Variable\n",
        "from sklearn.metrics import accuracy_score\n",
        "from skimage.transform import resize\n",
        "from sklearn.model_selection import train_test_split"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "04wVNnWbeoEm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "EuroSat_Type = 'RGB'    # use 'RGB' or 'ALL' for type of Eurosat Dataset. Just change in this line. Rest of the code is managed for both type\n",
        "lr = 0.01               # learn_rate\n",
        "milestones = [50,75,90] # multistep scheduler\n",
        "epochs = 100            # no of epochs"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nIeCE6YXrynl",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def iloader(path):\n",
        "    image = np.asarray((io.imread(path))/32000,dtype='float32')\n",
        "    return image.transpose(2,0,1)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6O3Qper7obov",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def Load_data():\n",
        "    if EuroSat_Type == 'RGB':\n",
        "      data = torchvision.datasets.DatasetFolder(root=root,loader = iloader, transform=None, extensions = 'jpg')\n",
        "    elif EuroSat_Type == 'ALL':\n",
        "      data = torchvision.datasets.DatasetFolder(root=root,loader = iloader, transform=None, extensions = 'tif')\n",
        "    train_set, val_set = train_test_split(data, test_size=0.2, stratify=data.targets)\n",
        "    #print(np.unique(train_set, return_counts=True))  #uncomment for class IDs\n",
        "    #print(np.unique(val_set, return_counts=True))    #uncomment for class IDs\n",
        "      \n",
        "    train_loader = torch.utils.data.DataLoader(train_set, batch_size=16, shuffle=True, num_workers=3, drop_last = True)\n",
        "    val_loader = torch.utils.data.DataLoader(val_set, batch_size=16, shuffle=True, num_workers=0, drop_last = True)\n",
        "    #test_loader = torch.utils.data.DataLoader(test_set, batch_size=128, shuffle=True, num_workers=0, drop_last = True)\n",
        "    return train_loader, val_loader #,test_loader"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nfg9TSOZgbvD",
        "colab_type": "text"
      },
      "source": [
        "**Functions for Augmentation**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MUn6ClNbsDCU",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def get_random_pos(img, window_shape = [55,55] ):\n",
        "    \"\"\" Extract of 2D random patch of shape window_shape in the image \"\"\"\n",
        "    w, h = window_shape\n",
        "    W, H = img.shape[-2:]\n",
        "    x1 = random.randint(0, W - w - 1)\n",
        "    #x2 = x1 + w\n",
        "    y1 = random.randint(0, H - h - 1)\n",
        "    #y2 = y1 + h\n",
        "    return x1, x1 + w, y1, y1 + h #x1, x2, y1, y2\n",
        "\n",
        "def random_crop_area(img):\n",
        "    x1,x2,y1,y2 = get_random_pos(img)\n",
        "    Sen_Im = img[:, x1:x2,y1:y2]\n",
        "    return resize(Sen_Im,img.shape,anti_aliasing=True)\n",
        "\n",
        "def sigmoid(z):\n",
        "    return 1/(1+np.exp(-z))\n",
        "\n",
        "def cus_aug(data):\n",
        "    data = torch.rot90(data,random.randint(-3,3), dims=random.choice([[3,2],[2,3]]))\n",
        "    if random.random()>0.75:\n",
        "        data = torch.flip(data, dims = random.choice([[2,],[3,],[2,3]]))\n",
        "    pixmis = torch.empty_like(data).random_(data.shape[-1])\n",
        "    pixmis = torch.where(pixmis>(data.shape[-1]/8),torch.ones_like(data),torch.zeros_like(data))\n",
        "    return data* pixmis"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OMA3QKW_qe-x",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def Load_model():\n",
        "    model_ft = models.resnet50()#pretrained=True)\n",
        "    num_ftrs = model_ft.fc.in_features\n",
        "    model_ft.fc = nn.Linear(num_ftrs, 10)\n",
        "    if EuroSat_Type == 'RGB':\n",
        "      inBands = 3\n",
        "    elif EuroSat_Type == 'ALL':\n",
        "      inBands = 13\n",
        "    model_ft.conv1 = nn.Conv2d(inBands, 64, kernel_size=7, stride=2, padding = 3, bias = False)\n",
        "    print('Model Loaded')\n",
        "    return model_ft"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t0DHlZDMrCjs",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def accuracy(gt_S,pred_S):       \n",
        "    _, alp = torch.max(torch.from_numpy(pred_S), 1)\n",
        "    return accuracy_score(gt_S,np.asarray(alp))#np.mean(F1score)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "O1b_BTGsrNQG",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def validation(model, test_,):\n",
        "    model.eval()\n",
        "    #tot_acc=[]\n",
        "    test_iter=0\n",
        "    with torch.no_grad():\n",
        "        for batch_idx, (data, target) in enumerate(test_):\n",
        "            data, target = cus_aug(Variable(data.cuda())), Variable(target.cuda())\n",
        "            output = model(data)\n",
        "            _, pred = torch.max(output, 1)\n",
        "            pred = output.data.cpu().numpy()\n",
        "            gt = target.data.cpu().numpy()\n",
        "            if test_iter==0:\n",
        "                all_pred=pred\n",
        "                all_gt=gt\n",
        "            else:\n",
        "                all_pred=np.vstack((all_pred,pred))\n",
        "                all_gt  =np.vstack((all_gt,gt))\n",
        "\n",
        "            test_iter=test_iter+1\n",
        "        acc=accuracy(all_gt.reshape(all_gt.shape[0] * all_gt.shape[1]),all_pred)\n",
        "        model.train()\n",
        "        return acc#,cm"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "klA5FBADrdQ2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def train(net, train_, val_, criterion, optimizer, epochs=None, scheduler=None, weights=None, save_epoch = 10):\n",
        "    losses=[]; acc=[]; mean_losses=[]; val_acc=[]\n",
        "    iter_ = t0 =0\n",
        "    for e in range(1, epochs + 1):\n",
        "        net.train()\n",
        "        for batch_idx, (data, target) in enumerate(train_):\n",
        "            data, target =  cus_aug(Variable(data.cuda())), Variable(target.cuda()),\n",
        "            optimizer.zero_grad()\n",
        "            output = net(data)\n",
        "            loss = criterion(output, target)\n",
        "            loss.backward()\n",
        "            optimizer.step()\n",
        "            losses = np.append(losses,loss.item())\n",
        "            mean_losses = np.append(mean_losses, np.mean(losses[max(0,iter_-100):iter_]))\n",
        "            if iter_ % 600 == 0: #printing after 600 epochs\n",
        "                clear_output()\n",
        "                print('Iteration Number',iter_,'{} seconds'.format(time.time() - t0))\n",
        "                t0 = time.time()\n",
        "                pred = output.data.cpu().numpy()#[0]\n",
        "                pred=sigmoid(pred)\n",
        "                gt = target.data.cpu().numpy()#[0]\n",
        "                acc = np.append(acc,accuracy(gt,pred))\n",
        "                print('Train (epoch {}/{}) [{}/{} ({:.0f}%)]\\tLoss: {:.6f}\\tAccuracy: {}\\tLearning Rate:{}'.format(\n",
        "                    e, epochs, batch_idx, len(train_),\n",
        "                    100. * batch_idx / len(train_), loss.item(), acc[-1],optimizer.param_groups[0]['lr']))\n",
        "                plt.plot(mean_losses) and plt.show()\n",
        "                val_acc = np.append(val_acc,validation(net, val_))\n",
        "                print('validation accuracy : {}'.format(val_acc[-1]))\n",
        "                plt.plot( range(len(acc)) ,acc,'b',label = 'training')\n",
        "                plt.plot( range(len(val_acc)), val_acc,'r--',label = 'validation')\n",
        "                plt.legend() and plt.show()\n",
        "                #print(mylabels[np.where(gt[1,:])[0]])\n",
        "            iter_ += 1\n",
        "            \n",
        "            del(data, target, loss)\n",
        "        if scheduler is not None:\n",
        "           scheduler.step()\n",
        "        if e % save_epoch == 0:\n",
        "            \n",
        "            torch.save(net.state_dict(), '.\\Eurosat{}'.format(e))\n",
        "    return net"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-dsot4MafMwP",
        "colab_type": "text"
      },
      "source": [
        "# Use 'RGB' or 'ALL' for type of Eurosat Dataset. Read the details about both [here ](https://https://github.com/phelber/EuroSAT)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aKuM_xkXnVMl",
        "colab_type": "text"
      },
      "source": [
        "Downloading and unzipping EuroSAT database : https://arxiv.org/abs/1709.00029\n",
        "(2500-3000) x 10 x 64 x 64 px RGB Sentinel 2 images.\n",
        "\n",
        "This below mentioned code block for downloading the dataset is taken from the code by [RomainHugues](https://https://github.com/RomainHugues/EuroSat/blob/master/EuroSAT.ipynb)\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xRBhwf-aaavg",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "if EuroSat_Type == 'RGB':\n",
        "  data_folder = '/content/sample_data/'\n",
        "  #root = os.path.join(data_folder, '2750/')\n",
        "  root = '2750/'\n",
        "  download_ON = os.path.exists(root)\n",
        "\n",
        "  if not download_ON:\n",
        "    # This can be long...\n",
        "    #os.chdir(data_folder)\n",
        "    os.system('wget http://madm.dfki.de/files/sentinel/EuroSAT.zip') #Just RGB Bands\n",
        "    !unzip EuroSAT.zip\n",
        "    download_ON = True\n",
        "elif EuroSat_Type == 'ALL':\n",
        "    root = 'ds/images/remote_sensing/otherDatasets/sentinel_2/tif/'\n",
        "    download_ON = os.path.exists(root)\n",
        "    if not download_ON:\n",
        "      os.system('wget http://madm.dfki.de/files/sentinel/EuroSATallBands.zip') #All bands\n",
        "      !unzip EuroSATallBands.zip\n",
        "      download_ON = True"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W4VZLxZerkFg",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 568
        },
        "outputId": "4c9a8ba7-f455-44ee-ae34-4c9b28f9a50e"
      },
      "source": [
        "if __name__ == '__main__': #https://discuss.pytorch.org/t/brokenpipeerror-errno-32-broken-pipe-when-i-run-cifar10-tutorial-py/6224/4   \n",
        "    torch.manual_seed(0)\n",
        "    np.random.seed(0)\n",
        "    torch.cuda.manual_seed(0)\n",
        "    random.seed(0)\n",
        "    print(\"pytorch version\", torch.__version__)\n",
        "    criteria = torch.nn.CrossEntropyLoss()\n",
        "    net = Load_model()\n",
        "    train_loader, val_loader =Load_data()\n",
        "    net=net.cuda()\n",
        "    optimizer = optim.Adam(net.parameters(), lr=lr, betas=(0.9, 0.999), eps=1e-08, weight_decay=0, amsgrad=False)\n",
        "    scheduler = optim.lr_scheduler.MultiStepLR(optimizer, milestones=milestones, gamma=0.1)\n",
        "    net = train(net, train_loader, val_loader, criteria, optimizer, epochs, scheduler)\n",
        "    #test_loader=torch.utils.data.DataLoader(test_loader,batch_size=128)\n",
        "    #test_accuracy = test(net, test_loader)\n",
        "    #print('Test Accuracy: {}'.format(test_accuracy))       \n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Iteration Number 1800 32.25961780548096 seconds\n",
            "Train (epoch 2/100) [450/1350 (33%)]\tLoss: 1.376774\tAccuracy: 0.5625\tLearning Rate:1.0000000000000003e-05\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXAAAAD7CAYAAABzGc+QAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "validation accuracy : 0.49369436201780414\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAD4CAYAAAD8Zh1EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2zATjcnTdOxm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}