SiameseNN_RAML.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "language": "python",
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3.7.10",
      "mimetype": "text/x-python",
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "pygments_lexer": "ipython3",
      "nbconvert_exporter": "python",
      "file_extension": ".py"
    },
    "colab": {
      "name": "SiameseNN_RAML.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gxiEcM4EdvGT"
      },
      "source": [
        "Dataset in Google Drive"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "T_oSczoF2seI",
        "outputId": "2345cd8e-c77a-4c48-fd3e-6abf10ce5096"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mounted at /content/drive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PDzfM5tPd8P8"
      },
      "source": [
        "Unzipping the data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7v_Wd1iz2jnx"
      },
      "source": [
        "!unzip \"/content/drive/MyDrive/archive.zip\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.076884Z",
          "iopub.execute_input": "2021-06-25T13:31:20.077328Z",
          "iopub.status.idle": "2021-06-25T13:31:20.092654Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.077294Z",
          "shell.execute_reply": "2021-06-25T13:31:20.091623Z"
        },
        "trusted": true,
        "id": "B-GcEa0A2fYj"
      },
      "source": [
        "import sys\n",
        "import numpy as np\n",
        "import pickle\n",
        "import os\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "\n",
        "# Load the TensorBoard notebook extension\n",
        "%load_ext tensorboard\n",
        "\n",
        "\n",
        "import cv2\n",
        "import time\n",
        "import itertools\n",
        "import random\n",
        "\n",
        "from sklearn.utils import shuffle\n",
        "import seaborn as sn\n",
        "import pandas as pd\n",
        "\n",
        "import tensorflow as tf\n",
        "from keras.models import Sequential\n",
        "from keras.optimizers import Adam, RMSprop\n",
        "from keras.layers import Conv2D, ZeroPadding2D, Activation, Input, concatenate, Dropout\n",
        "from keras.models import Model\n",
        "\n",
        "from keras.layers.normalization import BatchNormalization\n",
        "from keras.layers.pooling import MaxPooling2D, AveragePooling2D\n",
        "from keras.layers.merge import Concatenate\n",
        "from keras.layers.core import Lambda, Flatten, Dense\n",
        "from keras.initializers import glorot_uniform\n",
        "from keras.utils.vis_utils import plot_model\n",
        "\n",
        "from keras.engine.topology import Layer\n",
        "from keras.regularizers import l2\n",
        "from keras import backend as K\n",
        "from keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.576182Z",
          "iopub.execute_input": "2021-06-25T13:31:20.576537Z",
          "iopub.status.idle": "2021-06-25T13:31:20.584096Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.576511Z",
          "shell.execute_reply": "2021-06-25T13:31:20.583105Z"
        },
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lEXAwbdk2fYz",
        "outputId": "4c746487-c721-4952-d521-bbc3db9f43a2"
      },
      "source": [
        "# Checking Keras version\n",
        "import keras\n",
        "print(keras.__version__)"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2.5.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.094375Z",
          "iopub.execute_input": "2021-06-25T13:31:20.094889Z",
          "iopub.status.idle": "2021-06-25T13:31:20.112832Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.094838Z",
          "shell.execute_reply": "2021-06-25T13:31:20.111784Z"
        },
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gxewxLgR2fYo",
        "outputId": "fd8504a2-4fd2-4bb9-bb63-3eb6d15fe345"
      },
      "source": [
        "# Mapping dev to GPU\n",
        "sess = tf.compat.v1.Session(config=tf.compat.v1.ConfigProto(log_device_placement=True))"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Device mapping:\n",
            "/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: Tesla T4, pci bus id: 0000:00:04.0, compute capability: 7.5\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.115317Z",
          "iopub.execute_input": "2021-06-25T13:31:20.115973Z",
          "iopub.status.idle": "2021-06-25T13:31:20.122481Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.115861Z",
          "shell.execute_reply": "2021-06-25T13:31:20.121550Z"
        },
        "trusted": true,
        "id": "pBQYO_sf2fYq"
      },
      "source": [
        "# Path to Hindi and Bengali datasets (Comment the one you don't need)\n",
        "path = \"/content/BHSig260/Hindi/\"\n",
        "#path = \"/content/BHSig260/Bengali/\""
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.125671Z",
          "iopub.execute_input": "2021-06-25T13:31:20.125913Z",
          "iopub.status.idle": "2021-06-25T13:31:20.137429Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.125890Z",
          "shell.execute_reply": "2021-06-25T13:31:20.136643Z"
        },
        "trusted": true,
        "id": "dxtg1FUR2fYr"
      },
      "source": [
        "# Get the list of all directories and sort them\n",
        "dir_list = next(os.walk(path))[1]  # next(os.walk(path)) returns a tuple ('path', [dir list], [.txt and otr files])\n",
        "dir_list.sort()"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.139033Z",
          "iopub.execute_input": "2021-06-25T13:31:20.139418Z",
          "iopub.status.idle": "2021-06-25T13:31:20.218050Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.139383Z",
          "shell.execute_reply": "2021-06-25T13:31:20.217134Z"
        },
        "trusted": true,
        "id": "Oqx_kavd2fYr"
      },
      "source": [
        "# For each person segregate the genuine signatures from the forged signatures\n",
        "# Genuine signatures are stored in the list \"orig_groups\"\n",
        "# Forged signatures are stored i the list \"forged_groups\"\n",
        "orig_groups, forg_groups = [], []\n",
        "for directory in dir_list:\n",
        "    images = os.listdir(path+directory)\n",
        "    images.sort()\n",
        "    images = [path+directory+'/'+x for x in images]\n",
        "    forg_groups.append(images[:30]) # First 30 signatures in each folder are forged\n",
        "    orig_groups.append(images[30:]) # Next 24 signatures are genuine"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.220865Z",
          "iopub.execute_input": "2021-06-25T13:31:20.221166Z",
          "iopub.status.idle": "2021-06-25T13:31:20.229984Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.221137Z",
          "shell.execute_reply": "2021-06-25T13:31:20.229090Z"
        },
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "KsAFnZN-2fYs",
        "outputId": "b9b9321f-08ab-4b8a-b4a9-354f5b1b03e5"
      },
      "source": [
        "# Quick check to confirm we have data of all the 160 individuals\n",
        "len(orig_groups), len(forg_groups)"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(160, 160)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.233177Z",
          "iopub.execute_input": "2021-06-25T13:31:20.233448Z",
          "iopub.status.idle": "2021-06-25T13:31:20.239273Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.233415Z",
          "shell.execute_reply": "2021-06-25T13:31:20.238383Z"
        },
        "trusted": true,
        "id": "j8ecW_5p2fYs"
      },
      "source": [
        "orig_lengths = [len(x) for x in orig_groups]\n",
        "forg_lengths = [len(x) for x in forg_groups]"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.242376Z",
          "iopub.execute_input": "2021-06-25T13:31:20.242752Z",
          "iopub.status.idle": "2021-06-25T13:31:20.251085Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.242717Z",
          "shell.execute_reply": "2021-06-25T13:31:20.249953Z"
        },
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "NYhphvap2fYt",
        "outputId": "1a4caadd-6ec8-455b-a108-768cc55c9e2b"
      },
      "source": [
        "# Quick check to confirm that there are 24 Genuine signatures for each individual\n",
        "print(orig_lengths)"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.253244Z",
          "iopub.execute_input": "2021-06-25T13:31:20.253673Z",
          "iopub.status.idle": "2021-06-25T13:31:20.264923Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.253624Z",
          "shell.execute_reply": "2021-06-25T13:31:20.263767Z"
        },
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BqVokQ8-2fYt",
        "outputId": "88d830ac-5c6b-49dd-a960-806df5a7c95d"
      },
      "source": [
        "# Quick check to confirm that there are 30 Forged signatures for each individual\n",
        "print(forg_lengths)"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.266353Z",
          "iopub.execute_input": "2021-06-25T13:31:20.266809Z",
          "iopub.status.idle": "2021-06-25T13:31:20.274355Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.266774Z",
          "shell.execute_reply": "2021-06-25T13:31:20.273456Z"
        },
        "trusted": true,
        "id": "bZ5HrrnS2fYu"
      },
      "source": [
        "# Data split for both Hindi and Bengali dataset (Comment the one you don't need)\n",
        "\n",
        "# For Hindi\n",
        "orig_train, orig_val, orig_test = orig_groups[:120], orig_groups[120:140], orig_groups[140:]\n",
        "forg_train, forg_val, forg_test = forg_groups[:120], forg_groups[120:140], forg_groups[140:]\n",
        "\n",
        "# For Bengali\n",
        "#orig_train, orig_val, orig_test = orig_groups[:70], orig_groups[70:85], orig_groups[85:]\n",
        "#forg_train, forg_val, forg_test = forg_groups[:70], forg_groups[70:85], forg_groups[85:]"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.275158Z",
          "iopub.execute_input": "2021-06-25T13:31:20.275416Z",
          "iopub.status.idle": "2021-06-25T13:31:20.284572Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.275386Z",
          "shell.execute_reply": "2021-06-25T13:31:20.283623Z"
        },
        "trusted": true,
        "id": "2LzgUt7V2fYu"
      },
      "source": [
        "# All the images will be converted to the same size before processing\n",
        "img_h, img_w = 155, 220"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.285727Z",
          "iopub.execute_input": "2021-06-25T13:31:20.286021Z",
          "iopub.status.idle": "2021-06-25T13:31:20.297808Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.285995Z",
          "shell.execute_reply": "2021-06-25T13:31:20.297033Z"
        },
        "trusted": true,
        "id": "lOvXi4Mp2fYv"
      },
      "source": [
        "def visualize_sample_signature():\n",
        "    '''Function to randomly select a signature from train set and\n",
        "    print two genuine copies and one forged copy'''\n",
        "    fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize = (10, 10))\n",
        "    k = np.random.randint(len(orig_train))\n",
        "    orig_img_names = random.sample(orig_train[k], 2)\n",
        "    forg_img_name = random.sample(forg_train[k], 1)\n",
        "    orig_img1 = cv2.imread(orig_img_names[0], 0)\n",
        "    orig_img2 = cv2.imread(orig_img_names[1], 0)\n",
        "    forg_img = plt.imread(forg_img_name[0], 0)\n",
        "    orig_img1 = cv2.resize(orig_img1, (img_w, img_h))\n",
        "    orig_img2 = cv2.resize(orig_img2, (img_w, img_h))\n",
        "    forg_img = cv2.resize(forg_img, (img_w, img_h))\n",
        "\n",
        "    ax1.imshow(orig_img1, cmap = 'gray')\n",
        "    ax2.imshow(orig_img2, cmap = 'gray')\n",
        "    ax3.imshow(forg_img, cmap = 'gray')\n",
        "\n",
        "    ax1.set_title('Genuine Copy')\n",
        "    ax1.axis('off')\n",
        "    ax2.set_title('Genuine Copy')\n",
        "    ax2.axis('off')\n",
        "    ax3.set_title('Forged Copy')\n",
        "    ax3.axis('off')"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.299168Z",
          "iopub.execute_input": "2021-06-25T13:31:20.299548Z",
          "iopub.status.idle": "2021-06-25T13:31:20.494331Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.299513Z",
          "shell.execute_reply": "2021-06-25T13:31:20.493410Z"
        },
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 163
        },
        "id": "yq9cV0Ab2fYv",
        "outputId": "a1fbebfe-b0b6-4dbb-9890-8235893a5e72"
      },
      "source": [
        "visualize_sample_signature()"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x720 with 3 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.495805Z",
          "iopub.execute_input": "2021-06-25T13:31:20.496202Z",
          "iopub.status.idle": "2021-06-25T13:31:20.513255Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.496163Z",
          "shell.execute_reply": "2021-06-25T13:31:20.512303Z"
        },
        "trusted": true,
        "id": "PNWQWFE-2fYw"
      },
      "source": [
        "def generate_batch(orig_groups, forg_groups, batch_size = 32):\n",
        "    '''Function to generate a batch of data with batch_size number of data points\n",
        "    Half of the data points will be Genuine-Genuine pairs and half will be Genuine-Forged pairs'''\n",
        "    while True:\n",
        "        orig_pairs = []\n",
        "        forg_pairs = []\n",
        "        gen_gen_labels = []\n",
        "        gen_for_labels = []\n",
        "        all_pairs = []\n",
        "        all_labels = []\n",
        "        \n",
        "        # Here we create pairs of Genuine-Genuine image names and Genuine-Forged image names\n",
        "        # For every person we have 24 genuine signatures, hence we have \n",
        "        # 24 choose 2 = 276 Genuine-Genuine image pairs for one person.\n",
        "        # To make Genuine-Forged pairs, we pair every Genuine signature of a person\n",
        "        # with 12 randomly sampled Forged signatures of the same person.\n",
        "        # Thus we make 24 * 12 = 288 Genuine-Forged image pairs for one person.\n",
        "        # In all we have 120 person's data in the training data.\n",
        "\n",
        "        # Total no. of Genuine-Genuine pairs = 120 * 276 = 33120 [HINDI]\n",
        "        # Total number of Genuine-Forged pairs = 120 * 288 = 34560 [HINDI]\n",
        "        # Total no. of data points = 33120 + 34560 = 67680 [HINDI]\n",
        "\n",
        "        # Total no. of Genuine-Genuine pairs = 70 * 276 = 19320 [BENGALI]\n",
        "        # Total number of Genuine-Forged pairs = 70 * 288 = 20160 [BENGALI]\n",
        "        # Total no. of data points = 19320 + 20160 = 39480 [BENGALI]\n",
        "        \n",
        "        for orig, forg in zip(orig_groups, forg_groups):\n",
        "            orig_pairs.extend(list(itertools.combinations(orig, 2)))\n",
        "            for i in range(len(forg)):\n",
        "                forg_pairs.extend(list(itertools.product(orig[i:i+1], random.sample(forg, 12))))\n",
        "        \n",
        "        # Label for Genuine-Genuine pairs is 1\n",
        "        # Label for Genuine-Forged pairs is 0\n",
        "        gen_gen_labels = [1]*len(orig_pairs)\n",
        "        gen_for_labels = [0]*len(forg_pairs)\n",
        "        \n",
        "        # Concatenate all the pairs together along with their labels and shuffle them\n",
        "        all_pairs = orig_pairs + forg_pairs\n",
        "        all_labels = gen_gen_labels + gen_for_labels\n",
        "        del orig_pairs, forg_pairs, gen_gen_labels, gen_for_labels\n",
        "        all_pairs, all_labels = shuffle(all_pairs, all_labels)\n",
        "        \n",
        "        # Note the lists above contain only the image names and\n",
        "        # actual images are loaded and yielded below in batches\n",
        "        # Below we prepare a batch of data points and yield the batch\n",
        "        # In each batch we load \"batch_size\" number of image pairs\n",
        "        # These images are then removed from the original set so that\n",
        "        # they are not added again in the next batch.\n",
        "        k = 0\n",
        "        pairs=[np.zeros((batch_size, img_h, img_w, 1)) for i in range(2)]\n",
        "        targets=np.zeros((batch_size,))\n",
        "        for ix, pair in enumerate(all_pairs):\n",
        "            img1 = cv2.imread(pair[0], 0)\n",
        "            img2 = cv2.imread(pair[1], 0)\n",
        "            img1 = cv2.resize(img1, (img_w, img_h))\n",
        "            img2 = cv2.resize(img2, (img_w, img_h))\n",
        "            img1 = np.array(img1, dtype = np.float64)\n",
        "            img2 = np.array(img2, dtype = np.float64)\n",
        "            img1 /= 255\n",
        "            img2 /= 255\n",
        "            img1 = img1[..., np.newaxis]\n",
        "            img2 = img2[..., np.newaxis]\n",
        "            pairs[0][k, :, :, :] = img1\n",
        "            pairs[1][k, :, :, :] = img2\n",
        "            targets[k] = all_labels[ix]\n",
        "            k += 1\n",
        "            if k == batch_size:\n",
        "                yield pairs, targets\n",
        "                k = 0\n",
        "                pairs=[np.zeros((batch_size, img_h, img_w, 1)) for i in range(2)]\n",
        "                targets=np.zeros((batch_size,))"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.514759Z",
          "iopub.execute_input": "2021-06-25T13:31:20.515246Z",
          "iopub.status.idle": "2021-06-25T13:31:20.524429Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.515207Z",
          "shell.execute_reply": "2021-06-25T13:31:20.523372Z"
        },
        "trusted": true,
        "id": "dRgFl_Yz2fYx"
      },
      "source": [
        "def euclidean_distance(vects):\n",
        "    '''Compute Euclidean Distance between two vectors'''\n",
        "    x, y = vects\n",
        "    return K.sqrt(K.sum(K.square(x - y), axis=1, keepdims=True))"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.526884Z",
          "iopub.execute_input": "2021-06-25T13:31:20.527363Z",
          "iopub.status.idle": "2021-06-25T13:31:20.534034Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.527331Z",
          "shell.execute_reply": "2021-06-25T13:31:20.532928Z"
        },
        "trusted": true,
        "id": "Fw_l_hMc2fYx"
      },
      "source": [
        "def eucl_dist_output_shape(shapes):\n",
        "    shape1, shape2 = shapes\n",
        "    return (shape1[0], 1)"
      ],
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.535025Z",
          "iopub.execute_input": "2021-06-25T13:31:20.535335Z",
          "iopub.status.idle": "2021-06-25T13:31:20.544306Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.535306Z",
          "shell.execute_reply": "2021-06-25T13:31:20.543283Z"
        },
        "trusted": true,
        "id": "2jdT_ed82fYy"
      },
      "source": [
        "def contrastive_loss(y_true, y_pred):\n",
        "    '''Contrastive loss from Hadsell-et-al.'06\n",
        "    http://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf\n",
        "    '''\n",
        "    margin = 1\n",
        "    return K.mean(y_true * K.square(y_pred) + (1 - y_true) * K.square(K.maximum(margin - y_pred, 0)))"
      ],
      "execution_count": 20,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HTyh588Z0n7X"
      },
      "source": [
        "**SigNet**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.546693Z",
          "iopub.execute_input": "2021-06-25T13:31:20.546982Z",
          "iopub.status.idle": "2021-06-25T13:31:20.561344Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.546949Z",
          "shell.execute_reply": "2021-06-25T13:31:20.560287Z"
        },
        "trusted": true,
        "id": "Tg9PCKv72fYy"
      },
      "source": [
        "def Signet(input_shape):\n",
        "\n",
        "    seq = Sequential()\n",
        "    seq.add(Conv2D(96, kernel_size=(11, 11), activation='relu', name='conv1_1', strides=4, input_shape= input_shape, kernel_initializer='glorot_uniform'))\n",
        "    seq.add(BatchNormalization(epsilon=1e-06, axis=1, momentum=0.9))\n",
        "    seq.add(MaxPooling2D((3,3), strides=(2, 2)))    \n",
        "    seq.add(ZeroPadding2D((2, 2)))\n",
        "    \n",
        "    seq.add(Conv2D(256, kernel_size=(5, 5), activation='relu', name='conv2_1', strides=1, kernel_initializer='glorot_uniform'))\n",
        "    seq.add(BatchNormalization(epsilon=1e-06, axis=1, momentum=0.9))\n",
        "    seq.add(MaxPooling2D((3,3), strides=(2, 2)))\n",
        "    seq.add(Dropout(0.3))\n",
        "    seq.add(ZeroPadding2D((1, 1)))\n",
        "    \n",
        "    seq.add(Conv2D(384, kernel_size=(3, 3), activation='relu', name='conv3_1', strides=1, kernel_initializer='glorot_uniform'))\n",
        "    seq.add(ZeroPadding2D((1, 1)))\n",
        "    \n",
        "    seq.add(Conv2D(256, kernel_size=(3, 3), activation='relu', name='conv3_2', strides=1, kernel_initializer='glorot_uniform'))    \n",
        "    seq.add(MaxPooling2D((3,3), strides=(2, 2)))\n",
        "    seq.add(Dropout(0.3))\n",
        "    seq.add(Flatten(name='flatten'))\n",
        "    seq.add(Dense(1024, kernel_regularizer=l2(0.0005), activation='relu', kernel_initializer='glorot_uniform'))\n",
        "    seq.add(Dropout(0.5))\n",
        "    \n",
        "    seq.add(Dense(128, kernel_regularizer=l2(0.0005), activation='relu', kernel_initializer='glorot_uniform'))\n",
        "    \n",
        "    return seq"
      ],
      "execution_count": 71,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yexqGnk8Wqn1"
      },
      "source": [
        "**SmallCNN**\n",
        "\n",
        "*Avani Rateria and Suneeta Agarwal. Off-line signature verification through machine learning. In2018 5th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics andComputer Engineering (UPCON), pages 1–7. IEEE, 2018.*"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w_DDvlegWk9l"
      },
      "source": [
        "def SmallCNN(input_shape):\n",
        "\n",
        "    seq = Sequential()\n",
        "\n",
        "    seq.add(Conv2D(40, kernel_size=(5, 5), activation='relu', name='conv_1', strides=1, input_shape= input_shape))\n",
        "    seq.add(MaxPooling2D((2,2), strides=(2, 2)))    \n",
        "    seq.add(Conv2D(30, kernel_size=(3, 3), activation='relu', name='conv_2', strides=1))\n",
        "    seq.add(Conv2D(20, kernel_size=(3, 3), activation='relu', name='conv_3', strides=1))\n",
        "\n",
        "    seq.add(Flatten(name='flatten'))\n",
        "    \n",
        "    seq.add(Dense(20, activation='relu'))\n",
        "    return seq"
      ],
      "execution_count": 22,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lrUJ2zIPWmC7"
      },
      "source": [
        "**CustomNET** (Self-made model)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "awuTbhtOmfcs"
      },
      "source": [
        "def customNET(input_shape):\n",
        "\n",
        "    seq = Sequential()\n",
        "\n",
        "    seq.add(Conv2D(69, kernel_size=(10, 10), activation='relu', name='conv1_1', strides=2, input_shape= input_shape, kernel_initializer='glorot_uniform'))\n",
        "    seq.add(BatchNormalization(epsilon=1e-06, axis=1, momentum=0.9)) \n",
        "\n",
        "    seq.add(Conv2D(49, kernel_size=(5, 5), activation='relu', name='conv1_2', strides=2, input_shape= input_shape, kernel_initializer='glorot_uniform'))  \n",
        "    seq.add(BatchNormalization(epsilon=1e-06, axis=1, momentum=0.9))\n",
        "\n",
        "    seq.add(Conv2D(29, kernel_size=(3, 3), activation='relu', name='conv1_3', strides=2, input_shape= input_shape, kernel_initializer='glorot_uniform'))  \n",
        "    seq.add(BatchNormalization(epsilon=1e-06, axis=1, momentum=0.9))\n",
        "\n",
        "    seq.add(Flatten(name='flatten'))\n",
        "\n",
        "    seq.add(Dense(12, kernel_regularizer=l2(0.0005), activation='relu', kernel_initializer='glorot_uniform'))\n",
        "    return seq"
      ],
      "execution_count": 23,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.562909Z",
          "iopub.execute_input": "2021-06-25T13:31:20.563361Z",
          "iopub.status.idle": "2021-06-25T13:31:20.572877Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.563322Z",
          "shell.execute_reply": "2021-06-25T13:31:20.571957Z"
        },
        "trusted": true,
        "id": "XNBieOlq2fYy"
      },
      "source": [
        "# input image shape\n",
        "input_shape=(img_h, img_w, 1)"
      ],
      "execution_count": 72,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.585593Z",
          "iopub.execute_input": "2021-06-25T13:31:20.585948Z",
          "iopub.status.idle": "2021-06-25T13:31:20.766247Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.585911Z",
          "shell.execute_reply": "2021-06-25T13:31:20.765412Z"
        },
        "trusted": true,
        "id": "leX1zh7_2fYz"
      },
      "source": [
        "# Creating the base network\n",
        "# Comment out the network to be used\n",
        "\n",
        "base_network = Signet(input_shape)\n",
        "#base_network = SmallCNN(input_shape)\n",
        "#base_network = CustomNET(input_shape)\n",
        "\n",
        "input_a = Input(shape=(input_shape))\n",
        "input_b = Input(shape=(input_shape))\n",
        "\n",
        "# Because we re-use the same instance `base_network`,\n",
        "# the weights of the network\n",
        "# will be shared across the two branches\n",
        "processed_a = base_network(input_a)\n",
        "processed_b = base_network(input_b)\n",
        "\n",
        "# Compute the Euclidean distance between the two vectors in the latent space\n",
        "distance = Lambda(euclidean_distance, output_shape=eucl_dist_output_shape)([processed_a, processed_b])\n",
        "\n",
        "model = Model(inputs=[input_a, input_b], outputs=distance)"
      ],
      "execution_count": 73,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.767442Z",
          "iopub.execute_input": "2021-06-25T13:31:20.767920Z",
          "iopub.status.idle": "2021-06-25T13:31:20.774510Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.767882Z",
          "shell.execute_reply": "2021-06-25T13:31:20.773269Z"
        },
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "btpRNaU82fYz",
        "outputId": "e30343b7-95f2-40db-a19c-6616abcfa53a"
      },
      "source": [
        "# Calculating the sample sizes - Comment the dataset which is not being used\n",
        "\n",
        "# Hindi\n",
        "batch_sz = 128\n",
        "num_train_samples = 276*120 + 288*120\n",
        "num_val_samples = num_test_samples = 276*20 + 288*20\n",
        "num_train_samples, num_val_samples, num_test_samples\n",
        "\n",
        "# Bengali\n",
        "#batch_sz = 128\n",
        "#num_train_samples = 276*70 + 288*70\n",
        "#num_val_samples = num_test_samples = 276*15 + 288*15\n",
        "#num_train_samples, num_val_samples, num_test_samples"
      ],
      "execution_count": 74,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(67680, 11280, 11280)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 74
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.778817Z",
          "iopub.execute_input": "2021-06-25T13:31:20.779227Z",
          "iopub.status.idle": "2021-06-25T13:31:20.792489Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.779188Z",
          "shell.execute_reply": "2021-06-25T13:31:20.791640Z"
        },
        "trusted": true,
        "id": "_65nMHNp2fY0"
      },
      "source": [
        "# Compile model using RMSProp Optimizer and Contrastive loss function for SigNet and CustomCNN \n",
        "rms = RMSprop(learning_rate=1e-4, rho=0.9, epsilon=1e-08)\n",
        "model.compile(loss=contrastive_loss, optimizer=rms)"
      ],
      "execution_count": 75,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b1CJnBKhtENs"
      },
      "source": [
        "# Compile model using Adam Optimizer and Contrastive loss function for SmallCNN\n",
        "adam = keras.optimizers.Adam(learning_rate=1e-4, epsilon=1e-08)\n",
        "model.compile(loss=contrastive_loss, optimizer=adam)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.795414Z",
          "iopub.execute_input": "2021-06-25T13:31:20.795694Z",
          "iopub.status.idle": "2021-06-25T13:31:20.800957Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.795669Z",
          "shell.execute_reply": "2021-06-25T13:31:20.800002Z"
        },
        "trusted": true,
        "id": "i9SNSjxC2fY0"
      },
      "source": [
        "# Using Keras Callbacks, save the model after every epoch\n",
        "# Reduce the learning rate by a factor of 0.1 if the validation loss does not improve for 5 epochs\n",
        "# Stop the training using early stopping if the validation loss does not improve for 10 epochs\n",
        "callbacks = [\n",
        "    EarlyStopping(monitor='val_loss',patience=10, verbose=1),\n",
        "    ReduceLROnPlateau(factor=0.1, patience=5, min_lr=0.000001, verbose=1),\n",
        "    tf.keras.callbacks.TensorBoard(log_dir='logs', write_graph=True, update_freq='epoch'),\n",
        "    ModelCheckpoint('./Weights.h5', monitor='val_loss', verbose=1, save_weights_only=True, save_best_only=True)\n",
        "]"
      ],
      "execution_count": 76,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:20.802586Z",
          "iopub.execute_input": "2021-06-25T13:31:20.803079Z",
          "iopub.status.idle": "2021-06-25T13:31:21.178748Z",
          "shell.execute_reply.started": "2021-06-25T13:31:20.803001Z",
          "shell.execute_reply": "2021-06-25T13:31:21.177722Z"
        },
        "collapsed": true,
        "jupyter": {
          "outputs_hidden": true
        },
        "trusted": true,
        "id": "KrWAK65C2fY0"
      },
      "source": [
        "# Network plot\n",
        "plot_model(model, to_file='model.png', show_shapes=True, show_layer_names=True, rankdir='TB', expand_nested=True, dpi=96)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dykImkqGGRWh",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "b01c49f1-1e91-461d-ae90-08ddb6b896f8"
      },
      "source": [
        "# Installing Visualkeras (Pictorial representation of the network)\n",
        "%pip install visualkeras"
      ],
      "execution_count": 78,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Requirement already satisfied: visualkeras in /usr/local/lib/python3.7/dist-packages (0.0.2)\n",
            "Requirement already satisfied: aggdraw>=1.3.11 in /usr/local/lib/python3.7/dist-packages (from visualkeras) (1.3.12)\n",
            "Requirement already satisfied: pillow>=6.2.0 in /usr/local/lib/python3.7/dist-packages (from visualkeras) (7.1.2)\n",
            "Requirement already satisfied: numpy>=1.18.1 in /usr/local/lib/python3.7/dist-packages (from visualkeras) (1.19.5)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "aPxXfmzqFV5m",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 315
        },
        "outputId": "6b8237ea-9fd0-443d-982e-5e8f6ce22bb3"
      },
      "source": [
        "import visualkeras\n",
        "visualkeras.layered_view(base_network, legend=True, scale_xy=4, scale_z=0.05, to_file='network_struct.png')"
      ],
      "execution_count": 79,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<PIL.Image.Image image mode=RGBA size=813x298 at 0x7FA985014450>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 79
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "p1UrJ2EG8zj2"
      },
      "source": [
        "# Calling the Tensorboard\n",
        "%tensorboard --logdir logs"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gcVLdlpjh8mC"
      },
      "source": [
        "**Training and Validation**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T13:31:21.180474Z",
          "iopub.execute_input": "2021-06-25T13:31:21.180999Z",
          "iopub.status.idle": "2021-06-25T15:39:13.090295Z",
          "shell.execute_reply.started": "2021-06-25T13:31:21.180951Z",
          "shell.execute_reply": "2021-06-25T15:39:13.089364Z"
        },
        "trusted": true,
        "id": "XsBdlC1z2fY1"
      },
      "source": [
        "results = model.fit(generate_batch(orig_train, forg_train, batch_sz),\n",
        "                              steps_per_epoch = num_train_samples//batch_sz,\n",
        "                              epochs = 100,\n",
        "                              validation_data = generate_batch(orig_val, forg_val, batch_sz),\n",
        "                              validation_steps = num_val_samples//batch_sz,\n",
        "                              callbacks = callbacks)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5Zi84ZUaiCUa"
      },
      "source": [
        "**Testing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T16:28:21.211772Z",
          "iopub.execute_input": "2021-06-25T16:28:21.212138Z",
          "iopub.status.idle": "2021-06-25T16:28:21.219128Z",
          "shell.execute_reply.started": "2021-06-25T16:28:21.212105Z",
          "shell.execute_reply": "2021-06-25T16:28:21.218097Z"
        },
        "trusted": true,
        "id": "CWZT1DSx2fY2"
      },
      "source": [
        "# Function to compute accuracy\n",
        "def compute_accuracy_roc(predictions, labels):\n",
        "    '''Compute ROC accuracy with a range of thresholds on distances.\n",
        "    '''\n",
        "    dmax = np.max(predictions)\n",
        "    dmin = np.min(predictions)\n",
        "    nsame = np.sum(labels == 1)\n",
        "    ndiff = np.sum(labels == 0)\n",
        "   \n",
        "    step = 0.01\n",
        "    max_acc = 0\n",
        "    best_thresh = -1\n",
        "   \n",
        "    for d in np.arange(dmin, dmax+step, step):\n",
        "        idx1 = predictions.ravel() <= d\n",
        "        idx2 = predictions.ravel() > d\n",
        "     \n",
        "        tpr = float(np.sum(labels[idx1] == 1)) / nsame       \n",
        "        tnr = float(np.sum(labels[idx2] == 0)) / ndiff\n",
        "        acc = 0.5 * (tpr + tnr)       \n",
        "       \n",
        "        if (acc > max_acc):\n",
        "            max_acc, best_thresh, best_tpr, best_tnr = acc, d, tpr, tnr\n",
        "           \n",
        "    return max_acc, best_thresh, best_tpr, best_tnr"
      ],
      "execution_count": 81,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T15:39:13.114923Z",
          "iopub.execute_input": "2021-06-25T15:39:13.115393Z",
          "iopub.status.idle": "2021-06-25T15:39:13.156275Z",
          "shell.execute_reply.started": "2021-06-25T15:39:13.115350Z",
          "shell.execute_reply": "2021-06-25T15:39:13.155521Z"
        },
        "trusted": true,
        "id": "VB1wZH8R2fY2"
      },
      "source": [
        "# Loading the best weights to the model\n",
        "model.load_weights('./Weights.h5')"
      ],
      "execution_count": 82,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T15:39:13.157358Z",
          "iopub.execute_input": "2021-06-25T15:39:13.157701Z",
          "iopub.status.idle": "2021-06-25T15:49:07.441516Z",
          "shell.execute_reply.started": "2021-06-25T15:39:13.157666Z",
          "shell.execute_reply": "2021-06-25T15:49:07.440636Z"
        },
        "trusted": true,
        "id": "3yUEAnlB2fY2"
      },
      "source": [
        "# Testing\n",
        "test_gen = generate_batch(orig_test, forg_test, 1)\n",
        "pred, tr_y = [], []\n",
        "for i, j in enumerate(range(num_test_samples)):\n",
        "    (img1, img2), label = next(test_gen)\n",
        "    tr_y.append(label)\n",
        "    pred.append(model.predict([img1, img2])[0][0])"
      ],
      "execution_count": 83,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T16:28:23.711769Z",
          "iopub.execute_input": "2021-06-25T16:28:23.712110Z",
          "iopub.status.idle": "2021-06-25T16:28:23.780306Z",
          "shell.execute_reply.started": "2021-06-25T16:28:23.712078Z",
          "shell.execute_reply": "2021-06-25T16:28:23.779344Z"
        },
        "trusted": true,
        "id": "q3ZY0t3q2fY2"
      },
      "source": [
        "# Accuracy, Threshold, TPR & TNR (Best) \n",
        "tr_acc, threshold, best_tpr, best_tnr = compute_accuracy_roc(np.array(pred), np.array(tr_y))\n",
        "print(\"Accuracy: \", tr_acc, \"\\nThreshold: \", threshold, \"\\nTPR: \", best_tpr,\"\\nTNR: \", best_tnr)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T15:49:07.512297Z",
          "iopub.execute_input": "2021-06-25T15:49:07.512778Z",
          "iopub.status.idle": "2021-06-25T15:49:07.520237Z",
          "shell.execute_reply.started": "2021-06-25T15:49:07.512740Z",
          "shell.execute_reply": "2021-06-25T15:49:07.519213Z"
        },
        "trusted": true,
        "id": "s8trH1wh2fY3"
      },
      "source": [
        "# Predict distance score and classify test images as Genuine or Forged\n",
        "def predict_score():\n",
        "    test_point, test_label = next(test_gen)\n",
        "    img1, img2 = test_point[0], test_point[1]\n",
        "    \n",
        "    fig, (ax1, ax2) = plt.subplots(1, 2, figsize = (10, 10))\n",
        "    ax1.imshow(np.squeeze(img1), cmap='gray')\n",
        "    ax2.imshow(np.squeeze(img2), cmap='gray')\n",
        "    ax1.set_title('Genuine')\n",
        "    if test_label == 1:\n",
        "        ax2.set_title('Genuine')\n",
        "    else:\n",
        "        ax2.set_title('Forged')\n",
        "    ax1.axis('off')\n",
        "    ax2.axis('off')\n",
        "    plt.show()\n",
        "    result = model.predict([img1, img2])\n",
        "    diff = result[0][0]\n",
        "    print(\"Difference Score = \", diff)\n",
        "    if diff > threshold:\n",
        "        print(\"Its a Forged Signature\")\n",
        "    else:\n",
        "        print(\"Its a Genuine Signature\")"
      ],
      "execution_count": 87,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T15:49:07.521314Z",
          "iopub.execute_input": "2021-06-25T15:49:07.521861Z",
          "iopub.status.idle": "2021-06-25T15:49:07.703408Z",
          "shell.execute_reply.started": "2021-06-25T15:49:07.521820Z",
          "shell.execute_reply": "2021-06-25T15:49:07.702528Z"
        },
        "trusted": true,
        "id": "fSu6hABS2fY4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 260
        },
        "outputId": "68bcc04a-5169-4c26-ad05-4c4ecfdd56b4"
      },
      "source": [
        "predict_score()"
      ],
      "execution_count": 88,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x720 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Difference Score =  0.029860467\n",
            "Its a Genuine Signature\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AqTqG1shAE3c"
      },
      "source": [
        "**Plotting**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "execution": {
          "iopub.status.busy": "2021-06-25T16:11:48.453198Z",
          "iopub.execute_input": "2021-06-25T16:11:48.453522Z",
          "iopub.status.idle": "2021-06-25T16:11:48.614592Z",
          "shell.execute_reply.started": "2021-06-25T16:11:48.453493Z",
          "shell.execute_reply": "2021-06-25T16:11:48.613859Z"
        },
        "trusted": true,
        "id": "h7tDKCXu2fY1"
      },
      "source": [
        "# Loss plot\n",
        "plt.plot(range(len(results.history['loss'])), results.history['loss'], label=\"tr_loss\")\n",
        "plt.plot(range(len(results.history['val_loss'])), results.history['val_loss'], label=\"val_loss\")\n",
        "plt.plot(range(len(results.history[ 'lr'])), results.history['lr'], label=\"learning rate\")\n",
        "\n",
        "plt.xlabel(\"Epochs\")\n",
        "plt.legend()\n",
        "plt.grid(True)\n",
        "plt.savefig(\"lossplot.png\",dpi=300)\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UpNNytN6Ne0w"
      },
      "source": [
        "# Confusion Matrix\n",
        "# Gen-Forged pair (0) \n",
        "# Gen-Gen pair (1) \n",
        "\n",
        "orig_labels = np.array(tr_y)\n",
        "pred_labels = np.array(pred)\n",
        "TP, TN, FP, FN = 0, 0, 0, 0\n",
        "\n",
        "for i, j in zip(orig_labels, pred_labels):\n",
        "  if i[0]==0 and j>=threshold:\n",
        "    TN+=1\n",
        "  if i[0]==0 and j<threshold:\n",
        "    FP+=1\n",
        "  if i[0]==1 and j<threshold:\n",
        "    TP+=1\n",
        "  if i[0]==1 and j>=threshold:\n",
        "    FN+=1\n",
        "\n",
        "array = [[TP,FP],\n",
        "         [FN,TN]]\n",
        "\n",
        "df_cm = pd.DataFrame(array, range(2), range(2))\n",
        "plt.figure(dpi=150)\n",
        "sn.set(font_scale=1)\n",
        "sn.heatmap(df_cm, annot=True, annot_kws={\"size\": 15}, fmt='g', xticklabels=['1 (actual)','0 (actual)'], yticklabels=['1 (pred)','0 (pred)'])\n",
        "plt.savefig(\"confmat.png\",dpi=300)\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}