final_image_captioning_Yash Mishra.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "final_image_captioning_deepanshu.ipynb",
      "provenance": [],
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TphpR7Na36NR"
      },
      "source": [
        "# IMAGE CAPTIONING"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tvLMfGZi4Ap9"
      },
      "source": [
        "IMPORTING ALL LIBRARIES"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nEjegGtVN5ya"
      },
      "source": [
        "import tensorflow as tf\n",
        "from tqdm import tqdm\n",
        "from glob import glob\n",
        "import cv2\n",
        "import pickle\n",
        "import nltk\n",
        "import string\n",
        "import pickle\n",
        "import warnings\n",
        "from tensorflow.keras.preprocessing.text import Tokenizer\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "from tensorflow.keras.utils import to_categorical\n",
        "import numpy as np\n",
        "from tensorflow.keras.callbacks import ModelCheckpoint\n",
        "warnings.filterwarnings(\"ignore\")\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FnPvysEN35Q7"
      },
      "source": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "h-QCa8nEQN75",
        "outputId": "0ee83ffa-718e-4882-d615-97484d579ecd"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/gdrive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/gdrive; to attempt to forcibly remount, call drive.mount(\"/content/gdrive\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yAlqOGEpRyVN",
        "outputId": "09a29703-9633-4f5a-df5a-e01ee5b36e9c"
      },
      "source": [
        "cd /content/gdrive"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/gdrive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "dXwvky4GR2bE",
        "outputId": "df377eb7-85a5-4314-e90f-f1627e174b11"
      },
      "source": [
        "ls \n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[0m\u001b[01;34mMyDrive\u001b[0m/  \u001b[01;34mShareddrives\u001b[0m/\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "h2gWafCfW9XE",
        "outputId": "073b1eef-517b-41d7-a7e1-9d4f0d569e67"
      },
      "source": [
        "% cd MyDrive/"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/gdrive/MyDrive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ANAF9E-SXEkN",
        "outputId": "7ae151d9-eed6-4a60-9a22-324b02f05c7b"
      },
      "source": [
        "% cd kaggle/"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/gdrive/MyDrive/kaggle\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XtGFXIV0XJGd",
        "outputId": "e62a1ece-17b7-4e46-ed3a-3a9f9bf1d478"
      },
      "source": [
        "ls"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[0m\u001b[01;34mflickr30k_images\u001b[0m/  kaggle.json\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "HYuTPLXfvVed",
        "outputId": "761f5e60-447a-4928-d15d-880cbcc1bc42"
      },
      "source": [
        "% cd flickr30k_images"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/gdrive/MyDrive/kaggle/flickr30k_images\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "jJp5zhtgveLq",
        "outputId": "1ac28efe-1b69-4575-9d88-7ab0036f9261"
      },
      "source": [
        "ls"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Flickr30k_cap.txt  \u001b[0m\u001b[01;34mflickr30k_images\u001b[0m/  Flickr30k_IMG.txt  results.csv\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RIsQUTf4_sRv",
        "outputId": "31123787-5938-4e47-83bd-ad816911b4fd"
      },
      "source": [
        "%cd flickr30k_images/"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/gdrive/MyDrive/kaggle/flickr30k_images/flickr30k_images\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ia9b3Lpf4Jw0"
      },
      "source": [
        "# DECLARING CONSTANTS"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "32sUJ3MWu9UC"
      },
      "source": [
        "# Batch size for the encoder \n",
        "BATCH_SIZE_ENC = 50\n",
        "# Batch size for the decoder\n",
        "BATCH_SIZE_DEC = 100\n",
        "# Buffer size\n",
        "BUFFER_SIZE = 1000"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TRljBIWux0Jj"
      },
      "source": [
        "# image height according the the model architecture\n",
        "IMG_H = 224\n",
        "# image width according the the model architecture\n",
        "IMG_W = 224"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mJs-ime3OLea"
      },
      "source": [
        "# Dir which contains all the images. \n",
        "IMG_FILE_PATH = \"/content/gdrive/My Drive/kaggle/flickr30k_images/flickr30k_images/\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pqAUE43KybFJ"
      },
      "source": [
        "IMG_NAMES_TEXT_PATH = r\"/content/gdrive/MyDrive/kaggle/flickr30k_images/Flickr30k_IMG.txt\"\n",
        "CAP_TEXT_PATH = r\"/content/gdrive/MyDrive/kaggle/flickr30k_images/Flickr30k_cap.txt\""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4NAc14ON5E1k"
      },
      "source": [
        "# LOADING ALL THE FUNCTIONS"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RPYFp-pxQvoe"
      },
      "source": [
        "images_path_list = glob(IMG_FILE_PATH + '*.jpg')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bLo4719fYJ-p",
        "outputId": "865d2faf-bfea-41ee-c6e4-80e33025881d"
      },
      "source": [
        "image_model = tf.keras.applications.VGG16(include_top=False, weights='imagenet')\n",
        "new_input = image_model.input\n",
        "hidden_layer = image_model.layers[-1].output\n",
        "\n",
        "image_features_extract_model = tf.keras.Model(new_input, hidden_layer)\n",
        "\n",
        "print(image_features_extract_model.summary())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "input_1 (InputLayer)         [(None, None, None, 3)]   0         \n",
            "_________________________________________________________________\n",
            "block1_conv1 (Conv2D)        (None, None, None, 64)    1792      \n",
            "_________________________________________________________________\n",
            "block1_conv2 (Conv2D)        (None, None, None, 64)    36928     \n",
            "_________________________________________________________________\n",
            "block1_pool (MaxPooling2D)   (None, None, None, 64)    0         \n",
            "_________________________________________________________________\n",
            "block2_conv1 (Conv2D)        (None, None, None, 128)   73856     \n",
            "_________________________________________________________________\n",
            "block2_conv2 (Conv2D)        (None, None, None, 128)   147584    \n",
            "_________________________________________________________________\n",
            "block2_pool (MaxPooling2D)   (None, None, None, 128)   0         \n",
            "_________________________________________________________________\n",
            "block3_conv1 (Conv2D)        (None, None, None, 256)   295168    \n",
            "_________________________________________________________________\n",
            "block3_conv2 (Conv2D)        (None, None, None, 256)   590080    \n",
            "_________________________________________________________________\n",
            "block3_conv3 (Conv2D)        (None, None, None, 256)   590080    \n",
            "_________________________________________________________________\n",
            "block3_pool (MaxPooling2D)   (None, None, None, 256)   0         \n",
            "_________________________________________________________________\n",
            "block4_conv1 (Conv2D)        (None, None, None, 512)   1180160   \n",
            "_________________________________________________________________\n",
            "block4_conv2 (Conv2D)        (None, None, None, 512)   2359808   \n",
            "_________________________________________________________________\n",
            "block4_conv3 (Conv2D)        (None, None, None, 512)   2359808   \n",
            "_________________________________________________________________\n",
            "block4_pool (MaxPooling2D)   (None, None, None, 512)   0         \n",
            "_________________________________________________________________\n",
            "block5_conv1 (Conv2D)        (None, None, None, 512)   2359808   \n",
            "_________________________________________________________________\n",
            "block5_conv2 (Conv2D)        (None, None, None, 512)   2359808   \n",
            "_________________________________________________________________\n",
            "block5_conv3 (Conv2D)        (None, None, None, 512)   2359808   \n",
            "_________________________________________________________________\n",
            "block5_pool (MaxPooling2D)   (None, None, None, 512)   0         \n",
            "=================================================================\n",
            "Total params: 14,714,688\n",
            "Trainable params: 14,714,688\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "VxnJOpy5RnVv",
        "outputId": "0428e43f-57e6-43f2-96da-03a80e79d62d"
      },
      "source": [
        "print(f\"{len(images_path_list)} images found from {IMG_FILE_PATH}.\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "31783 images found from /content/gdrive/My Drive/kaggle/flickr30k_images/flickr30k_images/.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4nUn1rUYRrUX"
      },
      "source": [
        "def load_img(image_path: str, img_h: str = IMG_H, img_w: str = IMG_W) -> (object, str):\n",
        "    \"\"\"\n",
        "    Returns the numpy image and the image path\n",
        "    :param image_path: the path of the image\n",
        "    :param img_w: the width of image taken into the cnn\n",
        "    :param img_h: the height of image taken into the cnn\n",
        "    :return: image and image_path\n",
        "    \"\"\"\n",
        "    # Read the image\n",
        "    img = tf.io.read_file(image_path)\n",
        "    img = tf.image.decode_jpeg(img, channels=3)\n",
        "    # Resizing the image according the the model architecture\n",
        "    img = tf.image.resize(img, (img_h, img_w))\n",
        "    img = tf.keras.applications.vgg16.preprocess_input(img)\n",
        "    return img, image_path"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o5sdyfDAuBIi"
      },
      "source": [
        "def load_doc(filename: str) -> str:\n",
        "    \"\"\"\n",
        "    to open the file as read only\n",
        "    :param filename: name of the file\n",
        "    :return: the entire text as a string\n",
        "    \"\"\"\n",
        "    with open(filename, 'r') as file:\n",
        "        text = file.read()\n",
        "    return text"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EZ3G2-U6uKNc"
      },
      "source": [
        "def img_name_2_path(image_name: str, img_file_path: str = IMG_FILE_PATH, ext: str = r\".jpg\") -> str:\n",
        "    \"\"\"\n",
        "    Converts the name of the image to image path\n",
        "    :param image_name: The name of the image\n",
        "    :param img_file_path: The path where the image is stored\n",
        "    :param ext:The extension of the image default is .jpg\n",
        "    :return: The image path\n",
        "    \"\"\"\n",
        "    image_path = img_file_path + str(image_name) + ext\n",
        "    return image_path"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "29TAKgphP6Iw"
      },
      "source": [
        "def load_img_dataset(txt_path: str, batch_size=BATCH_SIZE_ENC):\n",
        "    \"\"\"\n",
        "    To load the train or test dataset\n",
        "    :param txt_path: The text file which has the names of the training images\n",
        "    :param batch_size: The batch size for processing\n",
        "    :return: image_dataset\n",
        "    \"\"\"\n",
        "    img_name_vector = load_set(txt_path)\n",
        "    img_path_list = map(img_name_2_path, img_name_vector)\n",
        "    encode_train = sorted(img_path_list)\n",
        "    image_dataset = tf.data.Dataset.from_tensor_slices(encode_train)\n",
        "    image_dataset = image_dataset.map(load_img, num_parallel_calls=tf.data.experimental.AUTOTUNE).batch(batch_size)\n",
        "    return image_dataset"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lRO6gIQ6umsu"
      },
      "source": [
        "def load_set(text_file_path: str) -> set:\n",
        "    \"\"\"\n",
        "    to load a pre-defined list of photo names\n",
        "    returns the names of the images form the set_text_file\n",
        "    :param text_file_path:\n",
        "    :return:\n",
        "    \"\"\"\n",
        "    # user defined to load the document\n",
        "    doc = load_doc(text_file_path)\n",
        "    dataset = doc.split('\\n')\n",
        "    return set(dataset)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RRw8L51FUOPn"
      },
      "source": [
        "image_train_dataset = load_img_dataset(IMG_NAMES_TEXT_PATH)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2kJekuaqrneR"
      },
      "source": [
        "# Extracting features "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QKZxcpIqeJER"
      },
      "source": [
        "if str(input(\"Do you want to extract the features of the images[y/Y]: \")).casefold() == 'y':\n",
        "    for img, path in tqdm(image_train_dataset):\n",
        "        batch_features = image_features_extract_model(img)\n",
        "        batch_features = tf.reshape(batch_features, (batch_features.shape[0], -1, batch_features.shape[3]))\n",
        "        for bf, p in zip(batch_features, path):\n",
        "            path_of_feature = p.numpy().decode(\"utf-8\")\n",
        "            np.save(path_of_feature, bf.numpy())"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "n10_EdAhUnYD",
        "outputId": "10c2788b-99d3-4968-e0fd-e4c2eadc0913"
      },
      "source": [
        "image_train_dataset"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<BatchDataset shapes: ((None, 224, 224, 3), (None,)), types: (tf.float32, tf.string)>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 25
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MiMxnnSzoJn3"
      },
      "source": [
        "#captions\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pS3amXMnskXD"
      },
      "source": [
        "def load_clean_cap(caption_txt_path: str, dataset: set) -> dict:\n",
        "    \"\"\"\n",
        "    load clean descriptions into memory\n",
        "    :param caption_txt_path: The path where the clean captions were saved\n",
        "    :param dataset: the img names of the tining or test data set\n",
        "    :return: dict of captions mapped with its curr name\n",
        "    \"\"\"\n",
        "    doc = load_doc(caption_txt_path)\n",
        "    clean_captions = dict()\n",
        "    for line in doc.split('\\n'):\n",
        "        # split line by white space\n",
        "        tokens = line.split()\n",
        "        # split name and captions\n",
        "        image_name, image_cap = tokens[0], \" \".join(tokens[1:])\n",
        "        # skip the images which are not in the set.\n",
        "        if image_name in dataset:\n",
        "            # creating a list\n",
        "            if image_name not in clean_captions:\n",
        "                clean_captions[image_name] = list()\n",
        "            clean_captions[image_name].append(image_cap)\n",
        "    return clean_captions\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "id": "TjmbtmAGvz5G",
        "outputId": "b20b72a0-4a14-450e-cbe3-d0be40cac2ad"
      },
      "source": [
        "IMG_NAMES_TEXT_PATH"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'/content/gdrive/MyDrive/kaggle/flickr30k_images/Flickr30k_IMG.txt'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Gv0Qyyrvv3lg"
      },
      "source": [
        "train_img_names = sorted(load_set(IMG_NAMES_TEXT_PATH))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xLVVzp0hv9Z0",
        "outputId": "5c70786e-35e3-4c50-abb3-45bddb6b8d91"
      },
      "source": [
        "train_img_names[:10]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['1000092795',\n",
              " '10002456',\n",
              " '1000268201',\n",
              " '1000344755',\n",
              " '1000366164',\n",
              " '1000523639',\n",
              " '1000919630',\n",
              " '10010052',\n",
              " '1001465944',\n",
              " '1001545525']"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 29
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "id": "ir3Y60NTv_cz",
        "outputId": "d862557a-a2be-450e-c911-8051eb6f179f"
      },
      "source": [
        "CAP_TEXT_PATH"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'/content/gdrive/MyDrive/kaggle/flickr30k_images/Flickr30k_cap.txt'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 30
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rwZLAIrYwDx0"
      },
      "source": [
        "train_img_cap = load_clean_cap(CAP_TEXT_PATH, train_img_names)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "MvelKAiSwJK0",
        "outputId": "853365ed-0c6a-4353-a6ad-fcac8d3ed3ef"
      },
      "source": [
        "len(train_img_cap)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "31783"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 32
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wlfL7_oLwNuf"
      },
      "source": [
        "def max_len(clean_captions: dict) -> int:\n",
        "    \"\"\"\n",
        "    Returns the length of the caption with most words\n",
        "    :param clean_captions: a dictionary of captions\n",
        "    :return: length of the longest caption\n",
        "    \"\"\"\n",
        "    # Converts a dictionary of clean captions and returns a list of captions.\n",
        "    clean_captions_list = [caption.split() for captions in clean_captions.values()\n",
        "                           for caption in captions]\n",
        "    return max(len(caption) for caption in clean_captions_list)\n",
        "    "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "P4Af9MQNwRBm",
        "outputId": "17cbb570-fd77-4dbb-af44-b4a9a930a5ac"
      },
      "source": [
        "MAX_CAP_LEN = max_len(train_img_cap)\n",
        "print(MAX_CAP_LEN)\n",
        "MAX_CAP_LEN = 50"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "74\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BlonqeZG0vIU"
      },
      "source": [
        "tokenizing words\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rVBTbeIugKsL"
      },
      "source": [
        "def create_tokenizer(captions_dict: dict, top_k: int = 5000) -> Tokenizer:\n",
        "    \"\"\"\n",
        "    Fit a tokenizer given caption descriptions\n",
        "    :param captions_dict: (dict) of clean captions\n",
        "    :param top_k: number of words in vocabulary\n",
        "    :return: tokenizer object\n",
        "    \"\"\"\n",
        "    clean_captions_list = [caption for captions in captions_dict.values()\n",
        "                           for caption in captions]\n",
        "    tokenizer = Tokenizer(num_words=top_k, oov_token=\"<unk>\")\n",
        "    # Map '<pad>' to '0'\n",
        "    tokenizer.word_index['<pad>'] = 0\n",
        "    tokenizer.index_word[0] = '<pad>'\n",
        "    tokenizer.fit_on_texts(clean_captions_list)\n",
        "    return tokenizer\n",
        "\n",
        "\n",
        "    "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6zGcjAthgQ3t"
      },
      "source": [
        "tokenizer = create_tokenizer(train_img_cap)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3zLKHZyc_C_P",
        "outputId": "667904d4-6666-407e-e4bc-6b088e0334ce"
      },
      "source": [
        "VOCAB_SIZE = len(tokenizer.word_index) + 1\n",
        "print(VOCAB_SIZE)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "16860\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mFHP4zWu_Lug"
      },
      "source": [
        "def tokenize_cap(tokenizer: Tokenizer, captions_dict: dict, pad_len: int = MAX_CAP_LEN) -> dict:\n",
        "    \"\"\"\n",
        "    Tokenizes the captions and\n",
        "    :param pad_len: The maximum caption length in the whole dataset\n",
        "    (should include both train and test dataset)\n",
        "    :param tokenizer: Tokenizer object\n",
        "    :param captions_dict: The dict of train/test cap which have to be tokenized\n",
        "    :return: A dict of tokenized captions\n",
        "    \"\"\"\n",
        "    pad_caps_dict = {img_name: pad_sequences(tokenizer.texts_to_sequences(captions), maxlen=pad_len, padding='post',truncating='post')\n",
        "                     for img_name, captions in captions_dict.items()}\n",
        "    return pad_caps_dict\n",
        "      "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YVIUJLfw0k1u"
      },
      "source": [
        "def path_cap_list(img_names_set: set, tokenizer: Tokenizer, captions_dict) -> (list, list):\n",
        "    \"\"\"\n",
        "    a list of image paths and a list of captions for images with corresponding values\n",
        "    Note: the captions will be tokenized and padded in this function\n",
        "    :param img_names_set: The set on which the processing is done\n",
        "    :param tokenizer: tokenizer\n",
        "    :param captions_dict: clean captions for that set without any tokenization\n",
        "    \"\"\"\n",
        "    tokenized_caps_dict = tokenize_cap(tokenizer, captions_dict)\n",
        "    image_name_list = sorted(img_names_set)\n",
        "    capt_list = [cap for name in image_name_list for cap in tokenized_caps_dict[name]]\n",
        "    img_path_list = [img_name_2_path(name) for name in image_name_list for i in range(len(tokenized_caps_dict[name]))]\n",
        "    return img_path_list, capt_list"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FLU_NBcP0nDF"
      },
      "source": [
        "def create_dataset(img_path_list: str, cap_list: str) -> object:\n",
        "    \"\"\"\n",
        "    :param img_path_list: The ordered list of img paths with duplication acc to number of captions  \n",
        "    :param cap_list: the padded caption list with the curr order \n",
        "    :return: dataset \n",
        "    \"\"\"\n",
        "    dataset = tf.data.Dataset.from_tensor_slices((img_path_list, cap_list))\n",
        "    # Use map to load the numpy files in parallel\n",
        "    dataset = dataset.map(lambda item1, item2: tf.numpy_function(load_npy, [item1, item2], [tf.float32, tf.int32]),\n",
        "                          num_parallel_calls=tf.data.experimental.AUTOTUNE)\n",
        "    # Shuffle and batch\n",
        "    dataset = dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE_DEC).prefetch(buffer_size=tf.data.experimental.AUTOTUNE)\n",
        "    return dataset"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9_w96bsmwkhf"
      },
      "source": [
        "def load_npy(image_path: str, cap: str) -> (str, str):\n",
        "    \"\"\"\n",
        "    :returns image tensor vector with the image path \n",
        "    :param image_path: \n",
        "    :param cap: \n",
        "    \"\"\"\n",
        "    img_tensor = np.load(image_path.decode('utf-8') + '.npy')\n",
        "    return img_tensor, cap"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "R63hJwZuwzdq"
      },
      "source": [
        "# train"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lkPkMRLBw4ua"
      },
      "source": [
        "img_name_train, caption_train = path_cap_list(train_img_names, tokenizer, train_img_cap)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dsUdMg9P1Hd7"
      },
      "source": [
        "train_dataset = create_dataset(img_name_train, caption_train)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Gmb1hJmq1x7h"
      },
      "source": [
        "embedding_dim = 256\n",
        "units = 512\n",
        "vocab_size = len(tokenizer.word_index) + 1 \n",
        "num_steps = len(train_img_names) // BATCH_SIZE_DEC \n",
        "EPOCHS = 20\n",
        "# Shape from last layer of VGG-16 :(7,7,512)\n",
        "# So, say there are 49 pixel locations now and each pixel is 512 dimensional\n",
        "features_shape = 512\n",
        "attention_features_shape = 49 "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_BmwJlkM2nxr"
      },
      "source": [
        "# MODEL\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sjpwCzCS2roF"
      },
      "source": [
        "class BahdanauAttention(tf.keras.Model):\n",
        "  def __init__(self, units):\n",
        "    super(BahdanauAttention, self).__init__()\n",
        "    self.W1 = tf.keras.layers.Dense(units)\n",
        "    self.W2 = tf.keras.layers.Dense(units)\n",
        "    self.V = tf.keras.layers.Dense(1)\n",
        "\n",
        "  def call(self, features, hidden):\n",
        "    # features(CNN_encoder output) shape == (batch_size, 64, embedding_dim)\n",
        "\n",
        "    # hidden shape == (batch_size, hidden_size)\n",
        "    # hidden_with_time_axis shape == (batch_size, 1, hidden_size)\n",
        "    hidden_with_time_axis = tf.expand_dims(hidden, 1)\n",
        "\n",
        "    # attention_hidden_layer shape == (batch_size, 64, units)\n",
        "    attention_hidden_layer = (tf.nn.tanh(self.W1(features) +\n",
        "                                         self.W2(hidden_with_time_axis)))\n",
        "\n",
        "    # score shape == (batch_size, 64, 1)\n",
        "    # This gives you an unnormalized score for each image feature.\n",
        "    score = self.V(attention_hidden_layer)\n",
        "\n",
        "    # attention_weights shape == (batch_size, 64, 1)\n",
        "    attention_weights = tf.nn.softmax(score, axis=1)\n",
        "\n",
        "    # context_vector shape after sum == (batch_size, hidden_size)\n",
        "    context_vector = attention_weights * features\n",
        "    context_vector = tf.reduce_sum(context_vector, axis=1)\n",
        "\n",
        "    return context_vector, attention_weights\n",
        "    "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tv1OCLSe2zv1"
      },
      "source": [
        "class CNN_Encoder(tf.keras.Model):\n",
        "    # Since you have already extracted the features and dumped it\n",
        "    # This encoder passes those features through a Fully connected layer\n",
        "    def __init__(self, embedding_dim):\n",
        "        super(CNN_Encoder, self).__init__()\n",
        "        # shape after fc == (batch_size, 49, embedding_dim)\n",
        "        self.fc = tf.keras.layers.Dense(embedding_dim)\n",
        "\n",
        "    def call(self, x):\n",
        "        x = self.fc(x)\n",
        "        x = tf.nn.relu(x)\n",
        "        return x"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FTL-wo6x20l1"
      },
      "source": [
        "class RNN_Decoder(tf.keras.Model):\n",
        "  def __init__(self, embedding_dim, units, vocab_size):\n",
        "    super(RNN_Decoder, self).__init__()\n",
        "    self.units = units\n",
        "\n",
        "    self.embedding = tf.keras.layers.Embedding(vocab_size, embedding_dim)\n",
        "#     self.cudnn = tf.compat.v1.keras.layers.CuDNNLSTM(self.units, \n",
        "#                                         return_sequences=True, \n",
        "#                                         return_state=True, \n",
        "#                                         recurrent_initializer='glorot_uniform')\n",
        "    self.gru = tf.compat.v1.keras.layers.CuDNNGRU(self.units,\n",
        "                                   return_sequences=True,\n",
        "                                   return_state=True,\n",
        "                                   recurrent_initializer='glorot_uniform')\n",
        "    self.fc1 = tf.keras.layers.Dense(self.units)\n",
        "    self.fc2 = tf.keras.layers.Dense(vocab_size)\n",
        "\n",
        "    self.attention = BahdanauAttention(self.units)\n",
        "\n",
        "  def call(self, x, features, hidden):\n",
        "    # defining attention as a separate model\n",
        "    context_vector, attention_weights = self.attention(features, hidden)\n",
        "\n",
        "    # x shape after passing through embedding == (batch_size, 1, embedding_dim)\n",
        "    x = self.embedding(x)\n",
        "\n",
        "    # x shape after concatenation == (batch_size, 1, embedding_dim + hidden_size)\n",
        "    x = tf.concat([tf.expand_dims(context_vector, 1), x], axis=-1)\n",
        "\n",
        "    # passing the concatenated vector to the GRU\n",
        "    output, state = self.gru(x)\n",
        "\n",
        "    # shape == (batch_size, MAX_CAP_LEN, hidden_size)\n",
        "    x = self.fc1(output)\n",
        "\n",
        "    # x shape == (batch_size * MAX_CAP_LEN, hidden_size)\n",
        "    x = tf.reshape(x, (-1, x.shape[2]))\n",
        "\n",
        "    # output shape == (batch_size * MAX_CAP_LEN, vocab)\n",
        "    x = self.fc2(x)\n",
        "\n",
        "    return x, state, attention_weights\n",
        "\n",
        "  def reset_state(self, batch_size):\n",
        "    return tf.zeros((batch_size, self.units))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0sA97GP520jm"
      },
      "source": [
        "encoder = CNN_Encoder(embedding_dim)\n",
        "decoder = RNN_Decoder(embedding_dim, units, vocab_size)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_Q-eaw3x20go",
        "outputId": "8f157212-9658-4ff2-a946-923075ed7cca"
      },
      "source": [
        "decoder.gru"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.layers.cudnn_recurrent.CuDNNGRU at 0x7ff502617810>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 49
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "aYEflGF120I6",
        "outputId": "3854298a-61d8-4de8-e3b2-e09b769f37f7"
      },
      "source": [
        "encoder"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<__main__.CNN_Encoder at 0x7ff5034eb050>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 50
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZYezcYV43GTy"
      },
      "source": [
        "optimizer = tf.keras.optimizers.Adam()\n",
        "loss_object = tf.keras.losses.SparseCategoricalCrossentropy(\n",
        "    from_logits=True, reduction='none')\n",
        "\n",
        "\n",
        "def loss_function(real, pred):\n",
        "  mask = tf.math.logical_not(tf.math.equal(real, 0))\n",
        "  loss_ = loss_object(real, pred)\n",
        "\n",
        "  mask = tf.cast(mask, dtype=loss_.dtype)\n",
        "  loss_ *= mask\n",
        "\n",
        "  return tf.reduce_mean(loss_)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uxjLQMrM3MQJ"
      },
      "source": [
        "# checkpoint\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XcvSCvM23S8A"
      },
      "source": [
        "checkpoint_path = \"./checkpoints/train\"\n",
        "ckpt = tf.train.Checkpoint(encoder=encoder,\n",
        "                           decoder=decoder,\n",
        "                           optimizer=optimizer)\n",
        "ckpt_manager = tf.train.CheckpointManager(ckpt, checkpoint_path, max_to_keep=5)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "scWhKGOs3S0F"
      },
      "source": [
        "start_epoch = 0\n",
        "if ckpt_manager.latest_checkpoint:\n",
        "  start_epoch = int(ckpt_manager.latest_checkpoint.split('-')[-1])\n",
        "  # restoring the latest checkpoint in checkpoint_path\n",
        "  ckpt.restore(ckpt_manager.latest_checkpoint)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HYWs8Eqb3a8W"
      },
      "source": [
        "# TRAIN"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3mKj7Kng3fgm"
      },
      "source": [
        "# adding this in a separate cell because if you run the training cell\n",
        "# many times, the loss_plot array will be reset\n",
        "loss_plot = []"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UU7CSYPY3iWB"
      },
      "source": [
        "@tf.function\n",
        "def train_step(img_tensor, target):\n",
        "  loss = 0\n",
        "\n",
        "  # initializing the hidden state for each batch\n",
        "  # because the captions are not related from image to image\n",
        "  hidden = decoder.reset_state(batch_size=target.shape[0])\n",
        "\n",
        "  dec_input = tf.expand_dims([tokenizer.word_index['startofseq']] * target.shape[0], 1)\n",
        "\n",
        "  with tf.GradientTape() as tape:\n",
        "      features = encoder(img_tensor)\n",
        "\n",
        "      for i in range(1, target.shape[1]):\n",
        "          # passing the features through the decoder\n",
        "          predictions, hidden, _ = decoder(dec_input, features, hidden)\n",
        "\n",
        "          loss += loss_function(target[:, i], predictions)\n",
        "\n",
        "          # using teacher forcing\n",
        "          dec_input = tf.expand_dims(target[:, i], 1)\n",
        "\n",
        "  total_loss = (loss / int(target.shape[1]))\n",
        "\n",
        "  trainable_variables = encoder.trainable_variables + decoder.trainable_variables\n",
        "\n",
        "  gradients = tape.gradient(loss, trainable_variables)\n",
        "\n",
        "  optimizer.apply_gradients(zip(gradients, trainable_variables))\n",
        "\n",
        "  return loss, total_loss"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JKF0by952xsI"
      },
      "source": [
        "# Loading weights \n",
        "encoder.load_weights(\"Encoder_train_1_30k.h5\")\n",
        "decoder.load_weights(\"Decoder_train_1_30k.h5\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zW3O4Uzp4Zc6"
      },
      "source": [
        "import time"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "s0WylpZqzEb8",
        "outputId": "14efc0df-cd82-43fe-846b-87573643a128"
      },
      "source": [
        "EPOCHS = 1\n",
        "\n",
        "for epoch in range(start_epoch, EPOCHS):\n",
        "    start = time.time()\n",
        "    total_loss = 0\n",
        "\n",
        "    for (batch, (img_tensor, target)) in enumerate(train_dataset):\n",
        "        batch_loss, t_loss = train_step(img_tensor, target)\n",
        "        total_loss += t_loss\n",
        "\n",
        "        if batch % 100 == 0:\n",
        "            average_batch_loss = batch_loss.numpy()/int(target.shape[1])\n",
        "            print(f'Epoch {epoch+1} Batch {batch} Loss {average_batch_loss:.4f}')\n",
        "    # storing the epoch end loss value to plot later\n",
        "    loss_plot.append(total_loss / num_steps)\n",
        "\n",
        "    if epoch % 5 == 0:\n",
        "      ckpt_manager.save()\n",
        "\n",
        "    print(f'Epoch {epoch+1} Loss {total_loss/num_steps:.6f}')\n",
        "    print(f'Time taken for 1 epoch {time.time()-start:.2f} sec\\n')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1 Batch 0 Loss 0.9522\n",
            "Epoch 1 Batch 100 Loss 0.9352\n",
            "Epoch 1 Batch 200 Loss 0.7898\n",
            "Epoch 1 Batch 300 Loss 0.7913\n",
            "Epoch 1 Batch 400 Loss 0.8793\n",
            "Epoch 1 Batch 500 Loss 0.7482\n",
            "Epoch 1 Batch 600 Loss 0.7342\n",
            "Epoch 1 Batch 700 Loss 0.7428\n",
            "Epoch 1 Batch 800 Loss 0.6772\n",
            "Epoch 1 Batch 900 Loss 0.7049\n",
            "Epoch 1 Batch 1000 Loss 0.7158\n",
            "Epoch 1 Batch 1100 Loss 0.6745\n",
            "Epoch 1 Batch 1200 Loss 0.7305\n",
            "Epoch 1 Batch 1300 Loss 0.8418\n",
            "Epoch 1 Batch 1400 Loss 0.7757\n",
            "Epoch 1 Batch 1500 Loss 0.8156\n",
            "Epoch 1 Batch 1600 Loss 0.9013\n",
            "Epoch 1 Batch 1700 Loss 0.7170\n",
            "Epoch 1 Batch 1800 Loss 0.8516\n",
            "Epoch 1 Batch 1900 Loss 0.8143\n",
            "Epoch 1 Batch 2000 Loss 0.7565\n",
            "Epoch 1 Batch 2100 Loss 0.7624\n",
            "Epoch 1 Batch 2200 Loss 0.8069\n",
            "Epoch 1 Batch 2300 Loss 0.8972\n",
            "Epoch 1 Batch 2400 Loss 0.9720\n",
            "Epoch 1 Batch 2500 Loss 0.8362\n",
            "Epoch 1 Batch 2600 Loss 0.8402\n",
            "Epoch 1 Loss 6.785010\n",
            "Time taken for 1 epoch 1690.62 sec\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VEbTK0D1PLol"
      },
      "source": [
        "# CAPTION"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OOICtYMvJAfT"
      },
      "source": [
        "def evaluate(image):\n",
        "    attention_plot = np.zeros((MAX_CAP_LEN, attention_features_shape))\n",
        "\n",
        "    hidden = decoder.reset_state(batch_size=1)\n",
        "\n",
        "    temp_input = tf.expand_dims(load_img(image)[0], 0)\n",
        "    img_tensor_val = image_features_extract_model(temp_input)\n",
        "    img_tensor_val = tf.reshape(img_tensor_val, (img_tensor_val.shape[0],\n",
        "                                                 -1,\n",
        "                                                 img_tensor_val.shape[3]))\n",
        "    features = encoder(img_tensor_val)\n",
        "\n",
        "    dec_input = tf.expand_dims([tokenizer.word_index['startofseq']], 0)\n",
        "    result = []\n",
        "\n",
        "    for i in range(MAX_CAP_LEN):\n",
        "        predictions, hidden, attention_weights = decoder(dec_input,\n",
        "                                                         features,\n",
        "                                                         hidden)\n",
        "\n",
        "        attention_plot[i] = tf.reshape(attention_weights, (-1, )).numpy()\n",
        "\n",
        "        predicted_id = tf.random.categorical(predictions, 1)[0][0].numpy()\n",
        "        result.append(tokenizer.index_word[predicted_id])\n",
        "\n",
        "        if tokenizer.index_word[predicted_id] == 'endofseq':\n",
        "            return result, attention_plot\n",
        "\n",
        "        dec_input = tf.expand_dims([predicted_id], 0)\n",
        "\n",
        "    attention_plot = attention_plot[:len(result), :]\n",
        "    return result, attention_plot"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jm0oXRB0PVBi"
      },
      "source": [
        "def plot_attention(image, result, attention_plot):\n",
        "    temp_image = np.array(Image.open(image))\n",
        "\n",
        "    fig = plt.figure(figsize=(10, 10))\n",
        "\n",
        "    len_result = len(result)\n",
        "    for i in range(len_result):\n",
        "        temp_att = np.resize(attention_plot[i], (8, 8))\n",
        "        grid_size = max(np.ceil(len_result/2), 2)\n",
        "        ax = fig.add_subplot(grid_size, grid_size, i+1)\n",
        "        ax.set_title(result[i])\n",
        "        img = ax.imshow(temp_image)\n",
        "        ax.imshow(temp_att, cmap='gray', alpha=0.6, extent=img.get_extent())\n",
        "\n",
        "    plt.tight_layout()\n",
        "    plt.show()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OFvfc4moPU5l"
      },
      "source": [
        "image_name = train_img_names[np.random.randint(0, len(train_img_cap))]\n",
        "img_path = img_name_2_path(image_name)\n",
        "real_captions = train_img_cap[image_name]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yYFq475W6rzk"
      },
      "source": [
        "from PIL import Image\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 321
        },
        "id": "sGAL_pGWPbgL",
        "outputId": "467bdabf-69a1-442f-88ca-d0b0755cc266"
      },
      "source": [
        "result, attention_plot = evaluate(img_path)\n",
        "print('Prediction Caption:', ' '.join(result))\n",
        "print('Real Captions:')\n",
        "for i, cap in enumerate(real_captions):\n",
        "    print(f\"{i+1}) {cap}\")\n",
        "plot_attention(img_path, result, attention_plot)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Prediction Caption: an older woman stand next to the two singer is around center of an audience endofseq\n",
            "Real Captions:\n",
            "1) startofseq woman in knit cap and man in peacoat sit on subway laden with bag endofseq\n",
            "2) startofseq two people wearing black sit on bench and look down endofseq\n",
            "3) startofseq two woman are sitting with their purse in their lap endofseq\n",
            "4) startofseq people in dark clothing sit together endofseq\n",
            "5) startofseq three people sitting on bench endofseq\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x720 with 16 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nqbc7XRmPeL2"
      },
      "source": [
        "# WEIGHTS"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2nq9iAkhPgdt"
      },
      "source": [
        "if os.path.isfile(\"Encoder_train_1_30k.h5\") is False:\n",
        "    encoder.save_weights(\"Encoder_train_1_30k.h5\")\n",
        "    print(\"Encoder saved\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oZGzgLo-Pjf2"
      },
      "source": [
        "if os.path.isfile(\"Decoder_train_1_30k.h5\") is False:\n",
        "    decoder.save_weights(\"Decoder_train_1_30k.h5\")\n",
        "    print(\"Decoder saved\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nVrT0nbb4qGI"
      },
      "source": [
        "# TESTING "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9CJdHO7e4srN"
      },
      "source": [
        "img_path = str(input(\"Enter the image path: \"))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u87riJqM4xOk"
      },
      "source": [
        "img_path"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rTd8KJx84y3P"
      },
      "source": [
        "result, attention_plot = evaluate(img_path)\n",
        "print('Prediction Caption:', ' '.join(result))\n",
        "plot_attention(img_path, result, attention_plot)\n",
        "# opening the image\n",
        "Image.open(img_path)"
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}