waveletCNN_Keras_0_6_[Exp_on_10K_data]_grayscale.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "waveletCNN_Keras_0.6_[Exp on 10K_data]_grayscale.ipynb",
      "provenance": [],
      "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/menon92/WaveletCNN/blob/master/waveletCNN_Keras_0_6_%5BExp_on_10K_data%5D_grayscale.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zQyX8ddJk32y"
      },
      "source": [
        "import numpy as np\n",
        "from matplotlib import pyplot as plt\n",
        "\n",
        "from keras import backend as K\n",
        "from keras.models import Model\n",
        "\n",
        "from keras.layers.merge import add, concatenate\n",
        "from keras.layers import Input, Dense, Conv2D, Lambda\n",
        "from keras.layers import Flatten, Reshape, Dropout, Activation\n",
        "from keras.layers import AveragePooling2D, BatchNormalization\n",
        "\n",
        "from keras.utils.vis_utils import plot_model\n",
        "from keras.preprocessing.image import ImageDataGenerator"
      ],
      "execution_count": 49,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EMcF2AfPk_SX"
      },
      "source": [
        "# batch operation usng tensor slice\n",
        "def WaveletTransformAxisY(batch_img):\n",
        "    odd_img  = batch_img[:,0::2]\n",
        "    even_img = batch_img[:,1::2]\n",
        "    L = (odd_img + even_img) / 2.0\n",
        "    H = K.abs(odd_img - even_img)\n",
        "    return L, H\n",
        "\n",
        "def WaveletTransformAxisX(batch_img):\n",
        "    # transpose + fliplr\n",
        "    tmp_batch = K.permute_dimensions(batch_img, [0, 2, 1])[:,:,::-1]\n",
        "    _dst_L, _dst_H = WaveletTransformAxisY(tmp_batch)\n",
        "    # transpose + flipud\n",
        "    dst_L = K.permute_dimensions(_dst_L, [0, 2, 1])[:,::-1,...]\n",
        "    dst_H = K.permute_dimensions(_dst_H, [0, 2, 1])[:,::-1,...]\n",
        "    return dst_L, dst_H"
      ],
      "execution_count": 50,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WLXXxjIhlCvn"
      },
      "source": [
        "def Wavelet(batch_image):\n",
        "    # make channel first image\n",
        "    batch_image = K.permute_dimensions(batch_image, [0, 3, 1, 2])\n",
        "    r = batch_image[:,0]\n",
        "  \n",
        "    # level 1 decomposition\n",
        "    wavelet_L, wavelet_H = WaveletTransformAxisY(r)\n",
        "    r_wavelet_LL, r_wavelet_LH = WaveletTransformAxisX(wavelet_L)\n",
        "    r_wavelet_HL, r_wavelet_HH = WaveletTransformAxisX(wavelet_H)\n",
        "\n",
        "    wavelet_data = [r_wavelet_LL, r_wavelet_LH, r_wavelet_HL, r_wavelet_HH]\n",
        "    transform_batch = K.stack(wavelet_data, axis=1)\n",
        " \n",
        "    # level 2 decomposition\n",
        "    wavelet_L2, wavelet_H2 = WaveletTransformAxisY(r_wavelet_LL)\n",
        "    r_wavelet_LL2, r_wavelet_LH2 = WaveletTransformAxisX(wavelet_L2)\n",
        "    r_wavelet_HL2, r_wavelet_HH2 = WaveletTransformAxisX(wavelet_H2)\n",
        "\n",
        "    wavelet_data_l2 = [r_wavelet_LL2, r_wavelet_LH2, r_wavelet_HL2, r_wavelet_HH2] \n",
        "    transform_batch_l2 = K.stack(wavelet_data_l2, axis=1)\n",
        "\n",
        "    # level 3 decomposition\n",
        "    wavelet_L3, wavelet_H3 = WaveletTransformAxisY(r_wavelet_LL2)\n",
        "    r_wavelet_LL3, r_wavelet_LH3 = WaveletTransformAxisX(wavelet_L3)\n",
        "    r_wavelet_HL3, r_wavelet_HH3 = WaveletTransformAxisX(wavelet_H3)\n",
        "\n",
        "    wavelet_data_l3 = [r_wavelet_LL3, r_wavelet_LH3, r_wavelet_HL3, r_wavelet_HH3]\n",
        "    transform_batch_l3 = K.stack(wavelet_data_l3, axis=1)\n",
        "\n",
        "    # level 4 decomposition\n",
        "    wavelet_L4, wavelet_H4 = WaveletTransformAxisY(r_wavelet_LL3)\n",
        "    r_wavelet_LL4, r_wavelet_LH4 = WaveletTransformAxisX(wavelet_L4)\n",
        "    r_wavelet_HL4, r_wavelet_HH4 = WaveletTransformAxisX(wavelet_H4)\n",
        "\n",
        "    wavelet_data_l4 = [r_wavelet_LL4, r_wavelet_LH4, r_wavelet_HL4, r_wavelet_HH4]\n",
        "    transform_batch_l4 = K.stack(wavelet_data_l4, axis=1)\n",
        "\n",
        "    # print('shape before')\n",
        "    # print(transform_batch.shape)\n",
        "    # print(transform_batch_l2.shape)\n",
        "    # print(transform_batch_l3.shape)\n",
        "    # print(transform_batch_l4.shape)\n",
        "\n",
        "    decom_level_1 = K.permute_dimensions(transform_batch, [0, 2, 3, 1])\n",
        "    decom_level_2 = K.permute_dimensions(transform_batch_l2, [0, 2, 3, 1])\n",
        "    decom_level_3 = K.permute_dimensions(transform_batch_l3, [0, 2, 3, 1])\n",
        "    decom_level_4 = K.permute_dimensions(transform_batch_l4, [0, 2, 3, 1])\n",
        "    \n",
        "    # print('shape after')\n",
        "    # print(decom_level_1.shape)\n",
        "    # print(decom_level_2.shape)\n",
        "    # print(decom_level_3.shape)\n",
        "    # print(decom_level_4.shape)\n",
        "    return [decom_level_1, decom_level_2, decom_level_3, decom_level_4]\n",
        "\n",
        "\n",
        "def Wavelet_out_shape(input_shapes):\n",
        "    return [tuple([None, 112, 112, 4]), tuple([None, 56, 56, 4]), \n",
        "            tuple([None, 28, 28, 4]), tuple([None, 14, 14, 4])]"
      ],
      "execution_count": 51,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "h6dmgkdTlDOk",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d87d3762-8533-4346-ea2e-438f4ce8b074"
      },
      "source": [
        "img_batch = K.zeros(shape=(2, 224, 224, 1), dtype='float32')\n",
        "decom_level_1, decom_level_2, decom_level_3, decom_level_4 = Wavelet(img_batch)\n",
        "\n",
        "print(decom_level_1.shape, decom_level_2.shape, decom_level_3.shape, decom_level_4.shape)"
      ],
      "execution_count": 52,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(2, 112, 112, 4) (2, 56, 56, 4) (2, 28, 28, 4) (2, 14, 14, 4)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2MpxaRCZlF1J"
      },
      "source": [
        "def get_wavelet_cnn_model():\n",
        "\n",
        "    input_shape = 224, 224, 1\n",
        "\n",
        "    input_ = Input(input_shape, name='the_input')\n",
        "    # wavelet = Lambda(Wavelet, name='wavelet')\n",
        "    wavelet = Lambda(Wavelet, Wavelet_out_shape, name='wavelet')\n",
        "    input_l1, input_l2, input_l3, input_l4 = wavelet(input_)\n",
        "    # print(input_l1)\n",
        "    # print(input_l2)\n",
        "    # print(input_l3)\n",
        "    # print(input_l4)\n",
        "    # level one decomposition starts\n",
        "    conv_1 = Conv2D(64, kernel_size=(3, 3), padding='same', name='conv_1')(input_l1)\n",
        "    norm_1 = BatchNormalization(name='norm_1')(conv_1)\n",
        "    relu_1 = Activation('relu', name='relu_1')(norm_1)\n",
        "\n",
        "    conv_1_2 = Conv2D(64, kernel_size=(3, 3), strides=(2, 2), padding='same', name='conv_1_2')(relu_1)\n",
        "    norm_1_2 = BatchNormalization(name='norm_1_2')(conv_1_2)\n",
        "    relu_1_2 = Activation('relu', name='relu_1_2')(norm_1_2)\n",
        "\n",
        "    # level two decomposition starts\n",
        "    conv_a = Conv2D(filters=64, kernel_size=(3, 3), padding='same', name='conv_a')(input_l2)\n",
        "    norm_a = BatchNormalization(name='norm_a')(conv_a)\n",
        "    relu_a = Activation('relu', name='relu_a')(norm_a)\n",
        "\n",
        "    # concate level one and level two decomposition\n",
        "    concate_level_2 = concatenate([relu_1_2, relu_a])\n",
        "    conv_2 = Conv2D(128, kernel_size=(3, 3), padding='same', name='conv_2')(concate_level_2)\n",
        "    norm_2 = BatchNormalization(name='norm_2')(conv_2)\n",
        "    relu_2 = Activation('relu', name='relu_2')(norm_2)\n",
        "\n",
        "    conv_2_2 = Conv2D(128, kernel_size=(3, 3), strides=(2, 2), padding='same', name='conv_2_2')(relu_2)\n",
        "    norm_2_2 = BatchNormalization(name='norm_2_2')(conv_2_2)\n",
        "    relu_2_2 = Activation('relu', name='relu_2_2')(norm_2_2)\n",
        "\n",
        "    # level three decomposition starts \n",
        "    conv_b = Conv2D(filters=64, kernel_size=(3, 3), padding='same', name='conv_b')(input_l3)\n",
        "    norm_b = BatchNormalization(name='norm_b')(conv_b)\n",
        "    relu_b = Activation('relu', name='relu_b')(norm_b)\n",
        "\n",
        "    conv_b_2 = Conv2D(128, kernel_size=(3, 3), padding='same', name='conv_b_2')(relu_b)\n",
        "    norm_b_2 = BatchNormalization(name='norm_b_2')(conv_b_2)\n",
        "    relu_b_2 = Activation('relu', name='relu_b_2')(norm_b_2)\n",
        "\n",
        "    # concate level two and level three decomposition \n",
        "    concate_level_3 = concatenate([relu_2_2, relu_b_2])\n",
        "    conv_3 = Conv2D(256, kernel_size=(3, 3), padding='same', name='conv_3')(concate_level_3)\n",
        "    norm_3 = BatchNormalization(name='nomr_3')(conv_3)\n",
        "    relu_3 = Activation('relu', name='relu_3')(norm_3)\n",
        "\n",
        "    conv_3_2 = Conv2D(256, kernel_size=(3, 3), strides=(2, 2), padding='same', name='conv_3_2')(relu_3)\n",
        "    norm_3_2 = BatchNormalization(name='norm_3_2')(conv_3_2)\n",
        "    relu_3_2 = Activation('relu', name='relu_3_2')(norm_3_2)\n",
        "\n",
        "    # level four decomposition start\n",
        "    conv_c = Conv2D(64, kernel_size=(3, 3), padding='same', name='conv_c')(input_l4)\n",
        "    norm_c = BatchNormalization(name='norm_c')(conv_c)\n",
        "    relu_c = Activation('relu', name='relu_c')(norm_c)\n",
        "\n",
        "    conv_c_2 = Conv2D(256, kernel_size=(3, 3), padding='same', name='conv_c_2')(relu_c)\n",
        "    norm_c_2 = BatchNormalization(name='norm_c_2')(conv_c_2)\n",
        "    relu_c_2 = Activation('relu', name='relu_c_2')(norm_c_2)\n",
        "\n",
        "    conv_c_3 = Conv2D(256, kernel_size=(3, 3), padding='same', name='conv_c_3')(relu_c_2)\n",
        "    norm_c_3 = BatchNormalization(name='norm_c_3')(conv_c_3)\n",
        "    relu_c_3 = Activation('relu', name='relu_c_3')(norm_c_3)\n",
        "\n",
        "    # concate level level three and level four decomposition\n",
        "    concate_level_4 = concatenate([relu_3_2, relu_c_3])\n",
        "    conv_4 = Conv2D(256, kernel_size=(3, 3), padding='same', name='conv_4')(concate_level_4)\n",
        "    norm_4 = BatchNormalization(name='norm_4')(conv_4)\n",
        "    relu_4 = Activation('relu', name='relu_4')(norm_4)\n",
        "\n",
        "    conv_4_2 = Conv2D(256, kernel_size=(3, 3), strides=(2, 2), padding='same', name='conv_4_2')(relu_4)\n",
        "    norm_4_2 = BatchNormalization(name='norm_4_2')(conv_4_2)\n",
        "    relu_4_2 = Activation('relu', name='relu_4_2')(norm_4_2)\n",
        "\n",
        "    conv_5_1 = Conv2D(128, kernel_size=(3, 3), padding='same', name='conv_5_1')(relu_4_2)\n",
        "    norm_5_1 = BatchNormalization(name='norm_5_1')(conv_5_1)\n",
        "    relu_5_1 = Activation('relu', name='relu_5_1')(norm_5_1)\n",
        "\n",
        "    pool_5_1 = AveragePooling2D(pool_size=(7, 7), strides=1, padding='same', name='avg_pool_5_1')(relu_5_1)\n",
        "    flat_5_1 = Flatten(name='flat_5_1')(pool_5_1) \n",
        "\n",
        "    fc_5 = Dense(2048, name='fc_5')(flat_5_1)\n",
        "    norm_5 = BatchNormalization(name='norm_5')(fc_5)\n",
        "    relu_5 = Activation('relu', name='relu_5')(norm_5)\n",
        "    drop_5 = Dropout(0.5, name='drop_5')(relu_5)\n",
        "\n",
        "    fc_6 = Dense(2048, name='fc_6')(drop_5)\n",
        "    norm_6 = BatchNormalization(name='norm_6')(fc_6)\n",
        "    relu_6 = Activation('relu', name='relu_6')(norm_6)\n",
        "    drop_6 = Dropout(0.5, name='drop_6')(relu_6)\n",
        "\n",
        "    output = Dense(2, activation='softmax', name='fc_7')(drop_6)\n",
        "\n",
        "    model = Model(inputs=input_, outputs=output)\n",
        "    model.summary()\n",
        "    plot_model(model, to_file='wavelet_cnn_0.5.png')\n",
        "\n",
        "    return model"
      ],
      "execution_count": 53,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XlNN-MeWlQHX",
        "outputId": "5e470544-1d53-41bd-8fef-df4bfcea5058",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "model = get_wavelet_cnn_model()"
      ],
      "execution_count": 54,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"model_2\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " the_input (InputLayer)         [(None, 224, 224, 1  0           []                               \n",
            "                                )]                                                                \n",
            "                                                                                                  \n",
            " wavelet (Lambda)               [(None, 112, 112, 4  0           ['the_input[0][0]']              \n",
            "                                ),                                                                \n",
            "                                 (None, 56, 56, 4),                                               \n",
            "                                 (None, 28, 28, 4),                                               \n",
            "                                 (None, 14, 14, 4)]                                               \n",
            "                                                                                                  \n",
            " conv_1 (Conv2D)                (None, 112, 112, 64  2368        ['wavelet[0][0]']                \n",
            "                                )                                                                 \n",
            "                                                                                                  \n",
            " norm_1 (BatchNormalization)    (None, 112, 112, 64  256         ['conv_1[0][0]']                 \n",
            "                                )                                                                 \n",
            "                                                                                                  \n",
            " relu_1 (Activation)            (None, 112, 112, 64  0           ['norm_1[0][0]']                 \n",
            "                                )                                                                 \n",
            "                                                                                                  \n",
            " conv_1_2 (Conv2D)              (None, 56, 56, 64)   36928       ['relu_1[0][0]']                 \n",
            "                                                                                                  \n",
            " conv_a (Conv2D)                (None, 56, 56, 64)   2368        ['wavelet[0][1]']                \n",
            "                                                                                                  \n",
            " norm_1_2 (BatchNormalization)  (None, 56, 56, 64)   256         ['conv_1_2[0][0]']               \n",
            "                                                                                                  \n",
            " norm_a (BatchNormalization)    (None, 56, 56, 64)   256         ['conv_a[0][0]']                 \n",
            "                                                                                                  \n",
            " relu_1_2 (Activation)          (None, 56, 56, 64)   0           ['norm_1_2[0][0]']               \n",
            "                                                                                                  \n",
            " relu_a (Activation)            (None, 56, 56, 64)   0           ['norm_a[0][0]']                 \n",
            "                                                                                                  \n",
            " concatenate_6 (Concatenate)    (None, 56, 56, 128)  0           ['relu_1_2[0][0]',               \n",
            "                                                                  'relu_a[0][0]']                 \n",
            "                                                                                                  \n",
            " conv_2 (Conv2D)                (None, 56, 56, 128)  147584      ['concatenate_6[0][0]']          \n",
            "                                                                                                  \n",
            " conv_b (Conv2D)                (None, 28, 28, 64)   2368        ['wavelet[0][2]']                \n",
            "                                                                                                  \n",
            " norm_2 (BatchNormalization)    (None, 56, 56, 128)  512         ['conv_2[0][0]']                 \n",
            "                                                                                                  \n",
            " norm_b (BatchNormalization)    (None, 28, 28, 64)   256         ['conv_b[0][0]']                 \n",
            "                                                                                                  \n",
            " relu_2 (Activation)            (None, 56, 56, 128)  0           ['norm_2[0][0]']                 \n",
            "                                                                                                  \n",
            " relu_b (Activation)            (None, 28, 28, 64)   0           ['norm_b[0][0]']                 \n",
            "                                                                                                  \n",
            " conv_2_2 (Conv2D)              (None, 28, 28, 128)  147584      ['relu_2[0][0]']                 \n",
            "                                                                                                  \n",
            " conv_b_2 (Conv2D)              (None, 28, 28, 128)  73856       ['relu_b[0][0]']                 \n",
            "                                                                                                  \n",
            " norm_2_2 (BatchNormalization)  (None, 28, 28, 128)  512         ['conv_2_2[0][0]']               \n",
            "                                                                                                  \n",
            " norm_b_2 (BatchNormalization)  (None, 28, 28, 128)  512         ['conv_b_2[0][0]']               \n",
            "                                                                                                  \n",
            " conv_c (Conv2D)                (None, 14, 14, 64)   2368        ['wavelet[0][3]']                \n",
            "                                                                                                  \n",
            " relu_2_2 (Activation)          (None, 28, 28, 128)  0           ['norm_2_2[0][0]']               \n",
            "                                                                                                  \n",
            " relu_b_2 (Activation)          (None, 28, 28, 128)  0           ['norm_b_2[0][0]']               \n",
            "                                                                                                  \n",
            " norm_c (BatchNormalization)    (None, 14, 14, 64)   256         ['conv_c[0][0]']                 \n",
            "                                                                                                  \n",
            " concatenate_7 (Concatenate)    (None, 28, 28, 256)  0           ['relu_2_2[0][0]',               \n",
            "                                                                  'relu_b_2[0][0]']               \n",
            "                                                                                                  \n",
            " relu_c (Activation)            (None, 14, 14, 64)   0           ['norm_c[0][0]']                 \n",
            "                                                                                                  \n",
            " conv_3 (Conv2D)                (None, 28, 28, 256)  590080      ['concatenate_7[0][0]']          \n",
            "                                                                                                  \n",
            " conv_c_2 (Conv2D)              (None, 14, 14, 256)  147712      ['relu_c[0][0]']                 \n",
            "                                                                                                  \n",
            " nomr_3 (BatchNormalization)    (None, 28, 28, 256)  1024        ['conv_3[0][0]']                 \n",
            "                                                                                                  \n",
            " norm_c_2 (BatchNormalization)  (None, 14, 14, 256)  1024        ['conv_c_2[0][0]']               \n",
            "                                                                                                  \n",
            " relu_3 (Activation)            (None, 28, 28, 256)  0           ['nomr_3[0][0]']                 \n",
            "                                                                                                  \n",
            " relu_c_2 (Activation)          (None, 14, 14, 256)  0           ['norm_c_2[0][0]']               \n",
            "                                                                                                  \n",
            " conv_3_2 (Conv2D)              (None, 14, 14, 256)  590080      ['relu_3[0][0]']                 \n",
            "                                                                                                  \n",
            " conv_c_3 (Conv2D)              (None, 14, 14, 256)  590080      ['relu_c_2[0][0]']               \n",
            "                                                                                                  \n",
            " norm_3_2 (BatchNormalization)  (None, 14, 14, 256)  1024        ['conv_3_2[0][0]']               \n",
            "                                                                                                  \n",
            " norm_c_3 (BatchNormalization)  (None, 14, 14, 256)  1024        ['conv_c_3[0][0]']               \n",
            "                                                                                                  \n",
            " relu_3_2 (Activation)          (None, 14, 14, 256)  0           ['norm_3_2[0][0]']               \n",
            "                                                                                                  \n",
            " relu_c_3 (Activation)          (None, 14, 14, 256)  0           ['norm_c_3[0][0]']               \n",
            "                                                                                                  \n",
            " concatenate_8 (Concatenate)    (None, 14, 14, 512)  0           ['relu_3_2[0][0]',               \n",
            "                                                                  'relu_c_3[0][0]']               \n",
            "                                                                                                  \n",
            " conv_4 (Conv2D)                (None, 14, 14, 256)  1179904     ['concatenate_8[0][0]']          \n",
            "                                                                                                  \n",
            " norm_4 (BatchNormalization)    (None, 14, 14, 256)  1024        ['conv_4[0][0]']                 \n",
            "                                                                                                  \n",
            " relu_4 (Activation)            (None, 14, 14, 256)  0           ['norm_4[0][0]']                 \n",
            "                                                                                                  \n",
            " conv_4_2 (Conv2D)              (None, 7, 7, 256)    590080      ['relu_4[0][0]']                 \n",
            "                                                                                                  \n",
            " norm_4_2 (BatchNormalization)  (None, 7, 7, 256)    1024        ['conv_4_2[0][0]']               \n",
            "                                                                                                  \n",
            " relu_4_2 (Activation)          (None, 7, 7, 256)    0           ['norm_4_2[0][0]']               \n",
            "                                                                                                  \n",
            " conv_5_1 (Conv2D)              (None, 7, 7, 128)    295040      ['relu_4_2[0][0]']               \n",
            "                                                                                                  \n",
            " norm_5_1 (BatchNormalization)  (None, 7, 7, 128)    512         ['conv_5_1[0][0]']               \n",
            "                                                                                                  \n",
            " relu_5_1 (Activation)          (None, 7, 7, 128)    0           ['norm_5_1[0][0]']               \n",
            "                                                                                                  \n",
            " avg_pool_5_1 (AveragePooling2D  (None, 7, 7, 128)   0           ['relu_5_1[0][0]']               \n",
            " )                                                                                                \n",
            "                                                                                                  \n",
            " flat_5_1 (Flatten)             (None, 6272)         0           ['avg_pool_5_1[0][0]']           \n",
            "                                                                                                  \n",
            " fc_5 (Dense)                   (None, 2048)         12847104    ['flat_5_1[0][0]']               \n",
            "                                                                                                  \n",
            " norm_5 (BatchNormalization)    (None, 2048)         8192        ['fc_5[0][0]']                   \n",
            "                                                                                                  \n",
            " relu_5 (Activation)            (None, 2048)         0           ['norm_5[0][0]']                 \n",
            "                                                                                                  \n",
            " drop_5 (Dropout)               (None, 2048)         0           ['relu_5[0][0]']                 \n",
            "                                                                                                  \n",
            " fc_6 (Dense)                   (None, 2048)         4196352     ['drop_5[0][0]']                 \n",
            "                                                                                                  \n",
            " norm_6 (BatchNormalization)    (None, 2048)         8192        ['fc_6[0][0]']                   \n",
            "                                                                                                  \n",
            " relu_6 (Activation)            (None, 2048)         0           ['norm_6[0][0]']                 \n",
            "                                                                                                  \n",
            " drop_6 (Dropout)               (None, 2048)         0           ['relu_6[0][0]']                 \n",
            "                                                                                                  \n",
            " fc_7 (Dense)                   (None, 2)            4098        ['drop_6[0][0]']                 \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 21,471,810\n",
            "Trainable params: 21,458,882\n",
            "Non-trainable params: 12,928\n",
            "__________________________________________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w6zpGp6flIEz"
      },
      "source": [
        "model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])"
      ],
      "execution_count": 55,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JKG4AAEDlWXm"
      },
      "source": [
        "!pip install -q kaggle\n",
        "!mkdir -p ~/.kaggle\n",
        "!cp kaggle.json ~/.kaggle/\n",
        "\n",
        "# This permissions change avoids a warning on Kaggle tool startup.\n",
        "!chmod 600 ~/.kaggle/kaggle.json"
      ],
      "execution_count": 41,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RQ7oNY98mc6q",
        "outputId": "820f61ca-ff90-4efb-9656-b94610841349",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "!kaggle datasets list -s cat"
      ],
      "execution_count": 42,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Warning: Looks like you're using an outdated API Version, please consider updating (server 1.5.12 / client 1.5.4)\n",
            "ref                                                     title                                      size  lastUpdated          downloadCount  \n",
            "------------------------------------------------------  ----------------------------------------  -----  -------------------  -------------  \n",
            "crawford/cat-dataset                                    Cat Dataset                                 4GB  2018-02-16 23:41:25          18237  \n",
            "tongpython/cat-and-dog                                  Cat and Dog                               218MB  2018-04-26 10:56:50          50735  \n",
            "d4rklucif3r/cat-and-dogs                                Cat & Dogs                                218MB  2021-06-07 11:39:13            670  \n",
            "ma7555/cat-breeds-dataset                               Cat Breeds Dataset                          2GB  2019-12-12 12:42:00           3347  \n",
            "mriganksingh/cat-images-dataset                         Cat images Dataset                          3MB  2019-04-13 20:24:18           1692  \n",
            "harriken/bias-media-cat                                 Bias Media CAT                             17MB  2017-10-27 22:46:25            375  \n",
            "andrewmvd/dog-and-cat-detection                         Dog and Cat Detection                       1GB  2020-05-12 20:46:17           1381  \n",
            "andrewmvd/cat-meow-classification                       Cat Meow Classification                    12MB  2021-11-22 07:16:46            176  \n",
            "uciml/ct-slice-localization                             CAT Scan Localization                      18MB  2017-09-06 22:01:03            513  \n",
            "hocop1/cat-and-dog-breeds-parameters                    Cat and dog breeds parameters              15KB  2019-01-26 16:39:25            852  \n",
            "arpitjain007/dog-vs-cat-fastai                          Dog vs Cat (fastai)                         2GB  2019-02-07 04:05:38           3910  \n",
            "ashishsaxena2209/animal-image-datasetdog-cat-and-panda  Animal Image Dataset(DOG, CAT and PANDA)  376MB  2019-05-26 06:54:26           8518  \n",
            "gpreda/cat-face-detection                               Cat Face Detection                        197KB  2020-01-12 17:53:49            194  \n",
            "timost1234/cat-individuals                              Cat Individual Images                      10GB  2020-06-08 12:53:03            117  \n",
            "tarunbisht11/yolo-animal-detection-small                Monkey, Cat and Dog detection              30MB  2020-08-25 06:31:26            770  \n",
            "karakaggle/kaggle-cat-vs-dog-dataset                    Kaggle: Cat VS Dog Dataset                787MB  2020-05-01 18:40:09           1248  \n",
            "muniryadi/cat-vs-rabbit                                 cat vs rabbit                              66MB  2021-07-28 19:04:56            175  \n",
            "erkamk/cat-and-dog-images-dataset                       Cat and Dog Images Dataset                128MB  2020-10-18 18:20:48            428  \n",
            "ashfakyeafi/cat-dog-images-for-classification           Cat & Dog images for Classification       545MB  2022-01-04 19:51:49            143  \n",
            "pybear/cats-vs-dogs                                     Cats -- VS -- Dogs                        788MB  2020-10-18 13:55:30           1210  \n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CvhyBIfrlb8G",
        "outputId": "729e20e3-3d70-4cea-8f88-aab2afc87539",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "!kaggle datasets download -d chetankv/dogs-cats-images"
      ],
      "execution_count": 43,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "dogs-cats-images.zip: Skipping, found more recently modified local copy (use --force to force download)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BDA9bhlQmvlX",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "e25c421e-68d6-4416-b254-29213e954c19"
      },
      "source": [
        "!unzip -qq /content/dogs-cats-images.zip"
      ],
      "execution_count": 44,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "replace dataset/test_set/cats/cat.4001.jpg? [y]es, [n]o, [A]ll, [N]one, [r]ename: n\n",
            "replace dataset/test_set/cats/cat.4002.jpg? [y]es, [n]o, [A]ll, [N]one, [r]ename: n\n",
            "replace dataset/test_set/cats/cat.4003.jpg? [y]es, [n]o, [A]ll, [N]one, [r]ename: A\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gDuuMOgEm-XS",
        "outputId": "ec5a8162-4db2-482a-eeae-e71e5f7ee78b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "import os\n",
        "data_set_path = '/content/dog vs cat/dataset'\n",
        "os.listdir(data_set_path)"
      ],
      "execution_count": 46,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['test_set', 'training_set']"
            ]
          },
          "metadata": {},
          "execution_count": 46
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ejmciqicnMnu",
        "outputId": "3a3d466f-f3e5-41ff-c6bc-86b72039dcf3",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "train_data_gen = ImageDataGenerator(\n",
        "\t#rescale=1./255,\n",
        "\tshear_range=0.1,\n",
        "\tzoom_range=0.1,\n",
        "\thorizontal_flip=True\n",
        ")\n",
        "val_data_gen = ImageDataGenerator(\n",
        "\t#rescale=1./255\n",
        ")\n",
        "\n",
        "train_img_dir = '/content/dog vs cat/dataset/training_set'\n",
        "val_img_dir   = '/content/dog vs cat/dataset/test_set'\n",
        "\n",
        "# def train & test generators\n",
        "train_generator = train_data_gen.flow_from_directory(\n",
        "\ttrain_img_dir,\n",
        "\ttarget_size=(224, 224),\n",
        "\tbatch_size=32,\n",
        "\tclass_mode='categorical',\n",
        "\tcolor_mode='grayscale'\n",
        ")\n",
        "val_generator = val_data_gen.flow_from_directory(\n",
        "\tval_img_dir,\n",
        "\ttarget_size=(224, 224),\n",
        "\tbatch_size=32,\n",
        "\tclass_mode='categorical',\n",
        "\tcolor_mode='grayscale'\n",
        ")"
      ],
      "execution_count": 56,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Found 8000 images belonging to 2 classes.\n",
            "Found 2000 images belonging to 2 classes.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "V2740Mdynf0w",
        "outputId": "a6f2188f-9794-4cdb-e5ac-23611053d5af",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "source": [
        "history = model.fit_generator(\n",
        "\ttrain_generator,\n",
        "\tsteps_per_epoch=8000//32,\n",
        "\tepochs=2,\n",
        "\tvalidation_data=val_generator,\n",
        "    validation_steps=2000//32,\n",
        "\tverbose=True)"
      ],
      "execution_count": 57,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:7: UserWarning: `Model.fit_generator` is deprecated and will be removed in a future version. Please use `Model.fit`, which supports generators.\n",
            "  import sys\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/2\n",
            "250/250 [==============================] - 87s 334ms/step - loss: 0.8852 - accuracy: 0.5634 - val_loss: 1.3827 - val_accuracy: 0.5968\n",
            "Epoch 2/2\n",
            "250/250 [==============================] - 81s 325ms/step - loss: 0.7297 - accuracy: 0.6409 - val_loss: 1.4619 - val_accuracy: 0.5938\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MI3d6ONirH6h",
        "outputId": "c9a07685-3188-49e3-c264-e94006b84b08",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 312
        }
      },
      "source": [
        "acc = history.history['accuracy']\n",
        "val_acc = history.history['val_accuracy']\n",
        "epochs = range(len(acc))\n",
        "\n",
        "plt.plot(acc, label='training accuracy')\n",
        "plt.plot(val_acc, label='validation accuracy')\n",
        "plt.title('Accuracy curve')\n",
        "plt.xlabel('epochs')\n",
        "plt.ylabel('accuracy')\n",
        "plt.legend()"
      ],
      "execution_count": 60,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7f54f35a9b10>"
            ]
          },
          "metadata": {},
          "execution_count": 60
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FjlFssvfrIwO",
        "outputId": "0fd04902-c5bc-49aa-e313-8262303cbf14",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 312
        }
      },
      "source": [
        "loss = history.history['loss']\n",
        "val_loss = history.history['val_loss']\n",
        "\n",
        "plt.plot(loss, label='training loss')\n",
        "plt.plot(val_loss, label='validation loss')\n",
        "plt.title('Loss curve')\n",
        "plt.xlabel('epochs')\n",
        "plt.ylabel('loss')\n",
        "plt.legend()"
      ],
      "execution_count": 61,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7f54f59c5b10>"
            ]
          },
          "metadata": {},
          "execution_count": 61
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        ""
      ],
      "metadata": {
        "id": "BLqbub6_Lenv"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}