FYP-v3 - ResNet50 Copy.ipynb

fyp-v3-resnet50.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "FYP-v3 - ResNet50 Copy.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "mount_file_id": "178r8lSgq96Ukcy9IVZCv5IcQXciYg6Fv",
      "authorship_tag": "ABX9TyPjIs9oNL2VC1ntZ5eCYAKm",
      "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/gist/MuizU/e6e0bcd2d03e6773a7b743bfdbd9c9ff/fyp-v3-resnet50.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SLCHmfK1JOZm",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "2f85ec53-37c3-4978-9ef2-e1f657ffb327"
      },
      "source": [
        "!pip install split_folders\r\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting split_folders\n",
            "  Downloading https://files.pythonhosted.org/packages/b8/5f/3c2b2f7ea5e047c8cdc3bb00ae582c5438fcdbbedcc23b3cc1c2c7aae642/split_folders-0.4.3-py3-none-any.whl\n",
            "Installing collected packages: split-folders\n",
            "Successfully installed split-folders-0.4.3\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6cXxQKvcKuMC"
      },
      "source": [
        "from google.colab import drive\r\n",
        "import splitfolders\r\n",
        "import os\r\n",
        "import tensorflow as tf\r\n",
        "from distutils.dir_util import copy_tree\r\n",
        "from tensorflow.keras.layers import Conv2D, Flatten, Dense, MaxPool2D, BatchNormalization, GlobalAveragePooling2D\r\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator, load_img\r\n",
        "from keras.applications.resnet50 import ResNet50, preprocess_input, decode_predictions\r\n",
        "from tensorflow.keras.preprocessing import image\r\n",
        "from tensorflow.keras.models import  Sequential, Model\r\n",
        "import matplotlib.pyplot as plt\r\n",
        "import numpy as np"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pKW8FNx1_iHB"
      },
      "source": [
        "drive.mount('/content/drive')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3hdCTVb_9nCX"
      },
      "source": [
        "!unzip -uq \"/content/drive/My Drive/IIT/Datasets/MVNN/MM17-WeiboRumorSet.zip\" -d \"sample_data/datasets\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "AuZTLMs8Fc7f",
        "outputId": "3900e4e4-a50f-45f1-a75d-0c7206e8979d"
      },
      "source": [
        "# Filter out corrupted images\r\n",
        "\r\n",
        "\r\n",
        "num_skipped = 0\r\n",
        "for folder_name in (\"rumor_images\", \"nonrumor_images\"):\r\n",
        "    folder_path = os.path.join(\"sample_data/datasets/MM17-WeiboRumorSet\", folder_name)\r\n",
        "    for fname in os.listdir(folder_path):\r\n",
        "        fpath = os.path.join(folder_path, fname)\r\n",
        "        try:\r\n",
        "            fobj = open(fpath, \"rb\")\r\n",
        "            is_jfif = tf.compat.as_bytes(\"JFIF\") in fobj.peek(10)\r\n",
        "        finally:\r\n",
        "            fobj.close()\r\n",
        "\r\n",
        "        if not is_jfif:\r\n",
        "            num_skipped += 1\r\n",
        "            # Delete corrupted image\r\n",
        "            os.remove(fpath)\r\n",
        "\r\n",
        "print(\"Deleted %d images\" % num_skipped)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Deleted 95 images\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DoeCXMLnlyPQ"
      },
      "source": [
        "# Copying the image folders to the input dataset dir\r\n",
        "\r\n",
        "# Directory \r\n",
        "directory = \"input_dataset\"\r\n",
        "  \r\n",
        "# Parent Directory path \r\n",
        "parent_dir = \"sample_data/datasets/MM17-WeiboRumorSet\"\r\n",
        "  \r\n",
        "# Path \r\n",
        "path = os.path.join(parent_dir, directory) \r\n",
        "  \r\n",
        "# Create the directory \r\n",
        "os.mkdir(path) \r\n",
        "print(\"Directory '% s' created\" % directory)\r\n",
        "\r\n",
        "# copy subdirectory example\r\n",
        "fromDirectory = \"sample_data/datasets/MM17-WeiboRumorSet/rumor_images\"\r\n",
        "toDirectory = \"sample_data/datasets/MM17-WeiboRumorSet/input_dataset/rumor_images\"\r\n",
        "\r\n",
        "copy_tree(fromDirectory, toDirectory)\r\n",
        "fromDirectory = \"sample_data/datasets/MM17-WeiboRumorSet/nonrumor_images\"\r\n",
        "toDirectory = \"sample_data/datasets/MM17-WeiboRumorSet/input_dataset/nonrumor_images\"\r\n",
        "copy_tree(fromDirectory, toDirectory)\r\n",
        "\r\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0F9_4fmx6c0V",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7790479c-9af5-4b35-8eea-a6bf3b260eca"
      },
      "source": [
        "# Test train validation split \r\n",
        "\r\n",
        "input_folder = \"sample_data/datasets/MM17-WeiboRumorSet/input_dataset/\"\r\n",
        "#os.mkdir(\"sample_data/datasets/MM17-WeiboRumorSet/processed_data\")\r\n",
        "output_dataset = \"sample_data/datasets/MM17-WeiboRumorSet/processed_data\"\r\n",
        "splitfolders.ratio(input_folder, output_dataset, seed=1337, ratio=(.7, .1, .2))\r\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Copying files: 13177 files [00:05, 2585.44 files/s]\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qtw7RV3i7Dj_"
      },
      "source": [
        "img_height, img_width = (224, 224)\r\n",
        "batch_size=32\r\n",
        "\r\n",
        "train_dir = r\"sample_data/datasets/MM17-WeiboRumorSet/processed_data/train\"\r\n",
        "validation_dir = r\"sample_data/datasets/MM17-WeiboRumorSet/processed_data/val/\"\r\n",
        "test_dir = r\"sample_data/datasets/MM17-WeiboRumorSet/processed_data/test/\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "H8OJDxtl-1Jg",
        "outputId": "1abcb291-32e5-4084-c8fb-357c52e2559f"
      },
      "source": [
        "train_datagen = ImageDataGenerator(preprocessing_function=preprocess_input,\r\n",
        "                                   shear_range=0.2,\r\n",
        "                                   horizontal_flip=True,\r\n",
        "                                   validation_split=0.4\r\n",
        "                                   )\r\n",
        "\r\n",
        "train_generator = train_datagen.flow_from_directory(\r\n",
        "    train_dir,\r\n",
        "    target_size=(img_height, img_width),\r\n",
        "    batch_size=batch_size,\r\n",
        "    class_mode=\"binary\",\r\n",
        "    subset=\"training\"\r\n",
        ")\r\n",
        "\r\n",
        "validation_generator = train_datagen.flow_from_directory(\r\n",
        "    validation_dir,\r\n",
        "    batch_size=batch_size,\r\n",
        "    target_size=(img_height, img_width),\r\n",
        "    class_mode=\"binary\",\r\n",
        "    subset=\"validation\"\r\n",
        ")\r\n",
        "\r\n",
        "test_generator = train_datagen.flow_from_directory(\r\n",
        "    test_dir,\r\n",
        "    target_size=(img_height, img_width),\r\n",
        "    batch_size=1,\r\n",
        "    class_mode=\"binary\",\r\n",
        "    subset=\"validation\"\r\n",
        ")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Found 5535 images belonging to 2 classes.\n",
            "Found 526 images belonging to 2 classes.\n",
            "Found 1054 images belonging to 2 classes.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "89uDgRZ_BKeN",
        "outputId": "795dfba2-d4cf-49b1-fcbd-7bfa63b732c6"
      },
      "source": [
        "x,y = test_generator.next()\r\n",
        "x.shape"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(1, 224, 224, 3)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "70IY4VrzBSNQ"
      },
      "source": [
        "# defining the mode\r\n",
        "base_model = ResNet50(include_top=False, weights=\"imagenet\")\r\n",
        "x = base_model.output\r\n",
        "x = GlobalAveragePooling2D()(x)\r\n",
        "x = Dense(1024, activation=\"relu\")(x)\r\n",
        "predictions = Dense(train_generator.num_classes, activation=\"softmax\")(x)\r\n",
        "model = Model(inputs=base_model.input, outputs=predictions)\r\n",
        "for layer in base_model.layers:\r\n",
        "  layer.trainable=False\r\n",
        "\r\n",
        "\r\n",
        "Early\r\n",
        "\r\n",
        "\r\n",
        "model.compile(optimizer=\"adam\", callbacks=[callback], loss=\"sparse_categorical_crossentropy\", metrics = [\"accuracy\"])\r\n",
        "model.fit(train_generator, epochs=25)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PFBEFZFPE0Ub"
      },
      "source": [
        "os.mkdir(\"sample_data/models\")\r\n",
        "model.save(\"sample_data/models/ResNet50_Base_Model.h5\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YwDrfEmvH81N",
        "outputId": "e22d6bcf-e0d9-4961-8c7c-528f730dd0a8"
      },
      "source": [
        "test_losss, test_acc = model.evaluate(test_generator, verbose=2)\r\n",
        "print(\"\\nTest accuracy: \", test_acc)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1054/1054 - 21s - loss: 0.6630 - accuracy: 0.7476\n",
            "\n",
            "Test accuracy:  0.7476280927658081\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}