06. VGGNet Architecture/Mini VGGNet.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Mini VGGNet.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "**INITIALIZATION:**\n",
        "- I use these three lines of code on top of my each notebooks because it will help to prevent any problems while reloading the same project. And the third line of code helps to make visualization within the notebook."
      ],
      "metadata": {
        "id": "5Vmy0mCIIvZy"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "5L3NmJeWFUEA"
      },
      "outputs": [],
      "source": [
        "#@ INITIALIZATION: \n",
        "%reload_ext autoreload\n",
        "%autoreload 2\n",
        "%matplotlib inline"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**LIBRARIES AND DEPENDENCIES:**\n",
        "- I have downloaded all the libraries and dependencies required for the project in one particular cell."
      ],
      "metadata": {
        "id": "tir9siflI3eF"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ IMPORTING NECESSARY LIBRARIES AND DEPENDENCIES:\n",
        "from keras.models import Sequential\n",
        "from keras.layers import BatchNormalization\n",
        "from keras.layers.convolutional import Conv2D\n",
        "from keras.layers.convolutional import MaxPooling2D\n",
        "from keras.layers.core import Activation\n",
        "from keras.layers.core import Flatten\n",
        "from keras.layers.core import Dense, Dropout\n",
        "from keras import backend as K\n",
        "from tensorflow.keras.optimizers import SGD\n",
        "from tensorflow.keras.datasets import cifar10\n",
        "from keras.callbacks import LearningRateScheduler\n",
        "\n",
        "from sklearn.preprocessing import LabelBinarizer\n",
        "from sklearn.model_selection import train_test_split\n",
        "from sklearn.metrics import classification_report\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np"
      ],
      "metadata": {
        "id": "Bzk_8g75I05a"
      },
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "**VGG ARCHITECTURE:**\n",
        "- I will define the build method of Mini VGGNet architecture below. It requires four parameters: width of input image, height of input image, depth of image, number of class labels in the classification task. The Sequential class, the building block of sequential networks sequentially stack one layer on top of the other layer initialized below. Batch Normalization operates over the channels, so in order to apply BN, we need to know which axis to normalize over. "
      ],
      "metadata": {
        "id": "5gbRFH9iLGKN"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ DEFINING VGGNET ARCHITECTURE:\n",
        "class MiniVGGNet:                                               # Defining VGG Network. \n",
        "    @staticmethod\n",
        "    def build(width, height, depth, classes):                   # Defining Build Method. \n",
        "        model = Sequential()                                    # Initializing Sequential Model.\n",
        "        inputShape = (width, height, depth)                     # Initializing Input Shape. \n",
        "        chanDim = -1                                            # Index of Channel Dimension.\n",
        "        if K.image_data_format() == \"channels_first\":\n",
        "            inputShape = (depth, width, height)                 # Initializing Input Shape. \n",
        "            chanDim = 1                                         # Index of Channel Dimension. \n",
        "        model.add(Conv2D(32, (3, 3), padding='same', \n",
        "                         input_shape=inputShape))               # Adding Convolutional Layer. \n",
        "        model.add(Activation(\"relu\"))                           # Adding RELU Activation Function. \n",
        "        model.add(BatchNormalization(axis=chanDim))             # Adding Batch Normalization Layer. \n",
        "        model.add(Conv2D(32, (3, 3), padding='same'))           # Adding Convolutional Layer. \n",
        "        model.add(Activation(\"relu\"))                           # Adding RELU Activation Function. \n",
        "        model.add(BatchNormalization(axis=chanDim))             # Adding Batch Normalization Layer. \n",
        "        model.add(MaxPooling2D(pool_size=(2, 2)))               # Adding Max Pooling Layer. \n",
        "        model.add(Dropout(0.25))                                # Adding Dropout Layer.\n",
        "        model.add(Conv2D(64, (3, 3), padding=\"same\"))           # Adding Convolutional Layer. \n",
        "        model.add(Activation(\"relu\"))                           # Adding RELU Activation Function. \n",
        "        model.add(BatchNormalization(axis=chanDim))             # Adding Batch Normalization Layer. \n",
        "        model.add(Conv2D(64, (3, 3), padding='same'))           # Adding Convolutional Layer. \n",
        "        model.add(Activation(\"relu\"))                           # Adding RELU Activation Function. \n",
        "        model.add(BatchNormalization(axis=chanDim))             # Adding Batch Normalization Layer. \n",
        "        model.add(MaxPooling2D(pool_size=(2, 2)))               # Adding Max Pooling Layer. \n",
        "        model.add(Dropout(0.25))                                # Adding Dropout Layer.  \n",
        "        model.add(Flatten())                                    # Adding Flatten Layer. \n",
        "        model.add(Dense(512))                                   # Adding FC Dense Layer. \n",
        "        model.add(Activation(\"relu\"))                           # Adding Activation Layer. \n",
        "        model.add(BatchNormalization())                         # Adding Batch Normalization Layer. \n",
        "        model.add(Dropout(0.5))                                 # Adding Dropout Layer. \n",
        "        model.add(Dense(classes))                               # Adding Dense Output Layer. \n",
        "        model.add(Activation(\"softmax\"))                        # Adding Softmax Layer. \n",
        "        return model"
      ],
      "metadata": {
        "id": "Ej2z0nJ-Kblr"
      },
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@ CUSTOM LEARNING RATE SCHEDULER: \n",
        "def step_decay(epoch):                          # Definig step decay function. \n",
        "    initAlpha = 0.01                            # Initializing initial LR.\n",
        "    factor = 0.25                               # Initializing drop factor. \n",
        "    dropEvery = 5                               # Initializing epochs to drop. \n",
        "    alpha = initAlpha*(factor ** np.floor((1 + epoch) / dropEvery))\n",
        "    return float(alpha)"
      ],
      "metadata": {
        "id": "KwzG_O1NlqVI"
      },
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "**VGGNET ON CIFAR10**"
      ],
      "metadata": {
        "id": "hNQiqb5gAlAU"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ GETTING THE DATASET:\n",
        "((trainX, trainY), (testX, testY)) = cifar10.load_data()            # Loading Dataset. \n",
        "trainX = trainX.astype(\"float\") / 255.0                             # Normalizing Dataset. \n",
        "testX = testX.astype(\"float\") / 255.0                               # Normalizing Dataset. \n",
        "\n",
        "#@ PREPARING THE DATASET:\n",
        "lb = LabelBinarizer()                                               # Initializing LabelBinarizer. \n",
        "trainY = lb.fit_transform(trainY)                                   # Converting Labels to Vectors. \n",
        "testY = lb.transform(testY)                                         # Converting Labels to Vectors. \n",
        "labelNames = [\"airplane\", \"automobile\", \"bird\", \"cat\", \"deer\", \n",
        "              \"dog\", \"frog\", \"horse\", \"ship\", \"truck\"]              # Initializing LabelNames."
      ],
      "metadata": {
        "id": "z-EnWB4J-8kl"
      },
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@ INITIALIZING OPTIMIZER AND MODEL: \n",
        "callbacks = [LearningRateScheduler(step_decay)]                        # Initializing Callbacks. \n",
        "opt = SGD(0.01, nesterov=True, momentum=0.9)                           # Initializing SGD Optimizer. \n",
        "model = MiniVGGNet.build(width=32, height=32, depth=3, classes=10)     # Initializing VGGNet Architecture. \n",
        "model.compile(loss=\"categorical_crossentropy\", optimizer=opt,\n",
        "              metrics=[\"accuracy\"])                                    # Compiling VGGNet Model. \n",
        "H = model.fit(trainX, trainY, \n",
        "              validation_data=(testX, testY), batch_size=64, \n",
        "              epochs=40, verbose=1, callbacks=callbacks)               # Training VGGNet Model."
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1lCPFWETBvgf",
        "outputId": "e1cdef1b-c28a-4e10-a935-5f602972acee"
      },
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/40\n",
            "782/782 [==============================] - 29s 21ms/step - loss: 1.6339 - accuracy: 0.4555 - val_loss: 1.1509 - val_accuracy: 0.5970 - lr: 0.0100\n",
            "Epoch 2/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 1.1813 - accuracy: 0.5932 - val_loss: 0.9222 - val_accuracy: 0.6733 - lr: 0.0100\n",
            "Epoch 3/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.9908 - accuracy: 0.6567 - val_loss: 0.8341 - val_accuracy: 0.7159 - lr: 0.0100\n",
            "Epoch 4/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.8854 - accuracy: 0.6945 - val_loss: 0.8282 - val_accuracy: 0.7167 - lr: 0.0100\n",
            "Epoch 5/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.7380 - accuracy: 0.7421 - val_loss: 0.6881 - val_accuracy: 0.7598 - lr: 0.0025\n",
            "Epoch 6/40\n",
            "782/782 [==============================] - 17s 21ms/step - loss: 0.6845 - accuracy: 0.7586 - val_loss: 0.6600 - val_accuracy: 0.7711 - lr: 0.0025\n",
            "Epoch 7/40\n",
            "782/782 [==============================] - 17s 21ms/step - loss: 0.6628 - accuracy: 0.7683 - val_loss: 0.6435 - val_accuracy: 0.7744 - lr: 0.0025\n",
            "Epoch 8/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.6391 - accuracy: 0.7755 - val_loss: 0.6362 - val_accuracy: 0.7784 - lr: 0.0025\n",
            "Epoch 9/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.6204 - accuracy: 0.7830 - val_loss: 0.6499 - val_accuracy: 0.7744 - lr: 0.0025\n",
            "Epoch 10/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5912 - accuracy: 0.7909 - val_loss: 0.6161 - val_accuracy: 0.7856 - lr: 6.2500e-04\n",
            "Epoch 11/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5812 - accuracy: 0.7936 - val_loss: 0.6054 - val_accuracy: 0.7879 - lr: 6.2500e-04\n",
            "Epoch 12/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5730 - accuracy: 0.7978 - val_loss: 0.5994 - val_accuracy: 0.7907 - lr: 6.2500e-04\n",
            "Epoch 13/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5698 - accuracy: 0.7974 - val_loss: 0.6013 - val_accuracy: 0.7882 - lr: 6.2500e-04\n",
            "Epoch 14/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5623 - accuracy: 0.8009 - val_loss: 0.5973 - val_accuracy: 0.7910 - lr: 6.2500e-04\n",
            "Epoch 15/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5496 - accuracy: 0.8064 - val_loss: 0.5961 - val_accuracy: 0.7905 - lr: 1.5625e-04\n",
            "Epoch 16/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5484 - accuracy: 0.8048 - val_loss: 0.5937 - val_accuracy: 0.7914 - lr: 1.5625e-04\n",
            "Epoch 17/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5573 - accuracy: 0.8037 - val_loss: 0.5950 - val_accuracy: 0.7902 - lr: 1.5625e-04\n",
            "Epoch 18/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5477 - accuracy: 0.8062 - val_loss: 0.5927 - val_accuracy: 0.7907 - lr: 1.5625e-04\n",
            "Epoch 19/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5446 - accuracy: 0.8073 - val_loss: 0.5904 - val_accuracy: 0.7923 - lr: 1.5625e-04\n",
            "Epoch 20/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5391 - accuracy: 0.8104 - val_loss: 0.5926 - val_accuracy: 0.7920 - lr: 3.9062e-05\n",
            "Epoch 21/40\n",
            "782/782 [==============================] - 17s 21ms/step - loss: 0.5419 - accuracy: 0.8080 - val_loss: 0.5915 - val_accuracy: 0.7929 - lr: 3.9062e-05\n",
            "Epoch 22/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5438 - accuracy: 0.8099 - val_loss: 0.5909 - val_accuracy: 0.7925 - lr: 3.9062e-05\n",
            "Epoch 23/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5467 - accuracy: 0.8075 - val_loss: 0.5914 - val_accuracy: 0.7919 - lr: 3.9062e-05\n",
            "Epoch 24/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5376 - accuracy: 0.8103 - val_loss: 0.5918 - val_accuracy: 0.7920 - lr: 3.9062e-05\n",
            "Epoch 25/40\n",
            "782/782 [==============================] - 17s 21ms/step - loss: 0.5410 - accuracy: 0.8085 - val_loss: 0.5923 - val_accuracy: 0.7917 - lr: 9.7656e-06\n",
            "Epoch 26/40\n",
            "782/782 [==============================] - 17s 21ms/step - loss: 0.5406 - accuracy: 0.8084 - val_loss: 0.5910 - val_accuracy: 0.7915 - lr: 9.7656e-06\n",
            "Epoch 27/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5384 - accuracy: 0.8097 - val_loss: 0.5901 - val_accuracy: 0.7919 - lr: 9.7656e-06\n",
            "Epoch 28/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5431 - accuracy: 0.8089 - val_loss: 0.5915 - val_accuracy: 0.7927 - lr: 9.7656e-06\n",
            "Epoch 29/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5417 - accuracy: 0.8095 - val_loss: 0.5921 - val_accuracy: 0.7925 - lr: 9.7656e-06\n",
            "Epoch 30/40\n",
            "782/782 [==============================] - 17s 21ms/step - loss: 0.5385 - accuracy: 0.8108 - val_loss: 0.5900 - val_accuracy: 0.7926 - lr: 2.4414e-06\n",
            "Epoch 31/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5451 - accuracy: 0.8073 - val_loss: 0.5910 - val_accuracy: 0.7923 - lr: 2.4414e-06\n",
            "Epoch 32/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5402 - accuracy: 0.8103 - val_loss: 0.5899 - val_accuracy: 0.7925 - lr: 2.4414e-06\n",
            "Epoch 33/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5405 - accuracy: 0.8091 - val_loss: 0.5909 - val_accuracy: 0.7928 - lr: 2.4414e-06\n",
            "Epoch 34/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5427 - accuracy: 0.8091 - val_loss: 0.5914 - val_accuracy: 0.7921 - lr: 2.4414e-06\n",
            "Epoch 35/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5416 - accuracy: 0.8105 - val_loss: 0.5906 - val_accuracy: 0.7928 - lr: 6.1035e-07\n",
            "Epoch 36/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5375 - accuracy: 0.8109 - val_loss: 0.5905 - val_accuracy: 0.7927 - lr: 6.1035e-07\n",
            "Epoch 37/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5372 - accuracy: 0.8092 - val_loss: 0.5900 - val_accuracy: 0.7923 - lr: 6.1035e-07\n",
            "Epoch 38/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5438 - accuracy: 0.8090 - val_loss: 0.5907 - val_accuracy: 0.7927 - lr: 6.1035e-07\n",
            "Epoch 39/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5424 - accuracy: 0.8097 - val_loss: 0.5906 - val_accuracy: 0.7922 - lr: 6.1035e-07\n",
            "Epoch 40/40\n",
            "782/782 [==============================] - 16s 21ms/step - loss: 0.5385 - accuracy: 0.8116 - val_loss: 0.5909 - val_accuracy: 0.7928 - lr: 1.5259e-07\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**MODEL EVALUATION:**"
      ],
      "metadata": {
        "id": "VYWybYDuELRf"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ INITIALIZING MODEL EVALUATION:\n",
        "predictions = model.predict(testX, batch_size=64)           # Getting Model Predictions. \n",
        "print(classification_report(testY.argmax(axis=1),\n",
        "                            predictions.argmax(axis=1), \n",
        "                            target_names=labelNames))       # Inspecting Classification Report."
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "HZXqORsIDAtg",
        "outputId": "103704a3-ddf5-4503-eb09-cf1fc33fa3f3"
      },
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "    airplane       0.85      0.79      0.82      1000\n",
            "  automobile       0.90      0.88      0.89      1000\n",
            "        bird       0.73      0.65      0.69      1000\n",
            "         cat       0.62      0.60      0.61      1000\n",
            "        deer       0.72      0.81      0.76      1000\n",
            "         dog       0.71      0.71      0.71      1000\n",
            "        frog       0.80      0.89      0.84      1000\n",
            "       horse       0.87      0.82      0.85      1000\n",
            "        ship       0.89      0.89      0.89      1000\n",
            "       truck       0.85      0.88      0.86      1000\n",
            "\n",
            "    accuracy                           0.79     10000\n",
            "   macro avg       0.79      0.79      0.79     10000\n",
            "weighted avg       0.79      0.79      0.79     10000\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@ INSPECTING TRAINING LOSS AND ACCURACY:\n",
        "plt.style.use(\"ggplot\")\n",
        "plt.figure()\n",
        "plt.plot(np.arange(0, 40), H.history[\"loss\"], label=\"train_loss\")\n",
        "plt.plot(np.arange(0, 40), H.history[\"val_loss\"], label=\"val_loss\")\n",
        "plt.plot(np.arange(0, 40), H.history[\"accuracy\"], label=\"train_acc\")\n",
        "plt.plot(np.arange(0, 40), H.history[\"val_accuracy\"], label=\"val_acc\")\n",
        "plt.title(\"Training Loss and Accuracy\")\n",
        "plt.xlabel(\"Epoch\")\n",
        "plt.ylabel(\"Loss/Accuracy\")\n",
        "plt.legend()\n",
        "plt.show();"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 299
        },
        "id": "eCeJOSaJEdLC",
        "outputId": "779e2ecc-6ae8-4164-a170-6f5e8085df88"
      },
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Note:**\n",
        "- Batch Normalization can lead to a faster, more stable convergence with higher accuracy. \n",
        "- Batch Normalization will require more wall time to train the network even though the network will obtain higher accuracy in less epochs. "
      ],
      "metadata": {
        "id": "cfm34boWToGw"
      }
    }
  ]
}