Tensorflow_Train_Fruits_Classification_model.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Tensorflow Train Fruits Classification model.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Tz3ELsXMizxN"
      },
      "source": [
        "# Tensorflow Fruits Classification\n",
        "heavily based on [retrain_classification_ptq_tf2.ipynb](https://github.com/google-coral/tutorials/blob/master/retrain_classification_ptq_tf2.ipynb)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_YdXX79vq0JC"
      },
      "source": [
        "## Import required libraries"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-pzqyb38iuIM"
      },
      "source": [
        "import tensorflow as tf\n",
        "\n",
        "import os\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7-VJv4_vq4vl"
      },
      "source": [
        "## Get data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9bblT1ptqykk",
        "outputId": "f3fd3bd7-38fe-4590-8cf0-ca7dfc0b4aab"
      },
      "source": [
        "# Install Kaggle API\n",
        "!pip install -q kaggle\n",
        "!pip install -q kaggle-cli"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[K     |████████████████████████████████| 81kB 8.9MB/s \n",
            "\u001b[K     |████████████████████████████████| 4.2MB 38.8MB/s \n",
            "\u001b[K     |████████████████████████████████| 51kB 8.7MB/s \n",
            "\u001b[K     |████████████████████████████████| 143kB 54.3MB/s \n",
            "\u001b[K     |████████████████████████████████| 112kB 58.7MB/s \n",
            "\u001b[?25h  Building wheel for kaggle-cli (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Building wheel for lxml (setup.py) ... \u001b[?25lerror\n",
            "\u001b[31m  ERROR: Failed building wheel for lxml\u001b[0m\n",
            "\u001b[?25h  Building wheel for PrettyTable (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Building wheel for pyperclip (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "    Running setup.py install for lxml ... \u001b[?25l\u001b[?25herror\n",
            "\u001b[31mERROR: Command errored out with exit status 1: /usr/bin/python3 -u -c 'import sys, setuptools, tokenize; sys.argv[0] = '\"'\"'/tmp/pip-install-13onbglm/lxml/setup.py'\"'\"'; __file__='\"'\"'/tmp/pip-install-13onbglm/lxml/setup.py'\"'\"';f=getattr(tokenize, '\"'\"'open'\"'\"', open)(__file__);code=f.read().replace('\"'\"'\\r\\n'\"'\"', '\"'\"'\\n'\"'\"');f.close();exec(compile(code, __file__, '\"'\"'exec'\"'\"'))' install --record /tmp/pip-record-eo5xcwb7/install-record.txt --single-version-externally-managed --compile Check the logs for full command output.\u001b[0m\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MVgCfgXVrDYd"
      },
      "source": [
        "# only for google colab\n",
        "import os\n",
        "os.environ['KAGGLE_USERNAME'] = \"<username>\" \n",
        "os.environ['KAGGLE_KEY'] = \"<key>\""
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xdllmoLWrEeF",
        "outputId": "df24b8dc-1d77-4db8-9e70-1a83fd54185a"
      },
      "source": [
        "!kaggle datasets download -d moltean/fruits --unzip"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading fruits.zip to /content\n",
            " 98% 746M/760M [00:05<00:00, 163MB/s]\n",
            "100% 760M/760M [00:05<00:00, 133MB/s]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "S_Pzodiss3Un"
      },
      "source": [
        "## Prepare the training data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Fiw70yUusyok",
        "outputId": "8e6a57bb-b022-4a90-ab3f-262dc60b2d5c"
      },
      "source": [
        "IMAGE_SIZE = 100\n",
        "BATCH_SIZE = 64\n",
        "\n",
        "datagen = tf.keras.preprocessing.image.ImageDataGenerator(\n",
        "    rescale=1./255, \n",
        "    validation_split=0.2)\n",
        "\n",
        "train_generator = datagen.flow_from_directory(\n",
        "    'fruits-360/Training',\n",
        "    target_size=(IMAGE_SIZE, IMAGE_SIZE),\n",
        "    batch_size=BATCH_SIZE)\n",
        "\n",
        "val_generator = datagen.flow_from_directory(\n",
        "    'fruits-360/Test',\n",
        "    target_size=(IMAGE_SIZE, IMAGE_SIZE),\n",
        "    batch_size=BATCH_SIZE)"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Found 67692 images belonging to 131 classes.\n",
            "Found 22688 images belonging to 131 classes.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "oqJLts3Ttp25",
        "outputId": "cc5d1422-86c9-44ee-fb89-0bd80b7312d5"
      },
      "source": [
        "image_batch, label_batch = next(val_generator)\n",
        "image_batch.shape, label_batch.shape"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "((64, 100, 100, 3), (64, 131))"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "JgX_PBgwtqGO",
        "outputId": "0934f3a1-86d9-461a-9fb4-3a3e218e12b8"
      },
      "source": [
        "print (train_generator.class_indices)\n",
        "\n",
        "labels = '\\n'.join(sorted(train_generator.class_indices.keys()))\n",
        "\n",
        "with open('fruit_labels.txt', 'w') as f:\n",
        "  f.write(labels)"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "{'Apple Braeburn': 0, 'Apple Crimson Snow': 1, 'Apple Golden 1': 2, 'Apple Golden 2': 3, 'Apple Golden 3': 4, 'Apple Granny Smith': 5, 'Apple Pink Lady': 6, 'Apple Red 1': 7, 'Apple Red 2': 8, 'Apple Red 3': 9, 'Apple Red Delicious': 10, 'Apple Red Yellow 1': 11, 'Apple Red Yellow 2': 12, 'Apricot': 13, 'Avocado': 14, 'Avocado ripe': 15, 'Banana': 16, 'Banana Lady Finger': 17, 'Banana Red': 18, 'Beetroot': 19, 'Blueberry': 20, 'Cactus fruit': 21, 'Cantaloupe 1': 22, 'Cantaloupe 2': 23, 'Carambula': 24, 'Cauliflower': 25, 'Cherry 1': 26, 'Cherry 2': 27, 'Cherry Rainier': 28, 'Cherry Wax Black': 29, 'Cherry Wax Red': 30, 'Cherry Wax Yellow': 31, 'Chestnut': 32, 'Clementine': 33, 'Cocos': 34, 'Corn': 35, 'Corn Husk': 36, 'Cucumber Ripe': 37, 'Cucumber Ripe 2': 38, 'Dates': 39, 'Eggplant': 40, 'Fig': 41, 'Ginger Root': 42, 'Granadilla': 43, 'Grape Blue': 44, 'Grape Pink': 45, 'Grape White': 46, 'Grape White 2': 47, 'Grape White 3': 48, 'Grape White 4': 49, 'Grapefruit Pink': 50, 'Grapefruit White': 51, 'Guava': 52, 'Hazelnut': 53, 'Huckleberry': 54, 'Kaki': 55, 'Kiwi': 56, 'Kohlrabi': 57, 'Kumquats': 58, 'Lemon': 59, 'Lemon Meyer': 60, 'Limes': 61, 'Lychee': 62, 'Mandarine': 63, 'Mango': 64, 'Mango Red': 65, 'Mangostan': 66, 'Maracuja': 67, 'Melon Piel de Sapo': 68, 'Mulberry': 69, 'Nectarine': 70, 'Nectarine Flat': 71, 'Nut Forest': 72, 'Nut Pecan': 73, 'Onion Red': 74, 'Onion Red Peeled': 75, 'Onion White': 76, 'Orange': 77, 'Papaya': 78, 'Passion Fruit': 79, 'Peach': 80, 'Peach 2': 81, 'Peach Flat': 82, 'Pear': 83, 'Pear 2': 84, 'Pear Abate': 85, 'Pear Forelle': 86, 'Pear Kaiser': 87, 'Pear Monster': 88, 'Pear Red': 89, 'Pear Stone': 90, 'Pear Williams': 91, 'Pepino': 92, 'Pepper Green': 93, 'Pepper Orange': 94, 'Pepper Red': 95, 'Pepper Yellow': 96, 'Physalis': 97, 'Physalis with Husk': 98, 'Pineapple': 99, 'Pineapple Mini': 100, 'Pitahaya Red': 101, 'Plum': 102, 'Plum 2': 103, 'Plum 3': 104, 'Pomegranate': 105, 'Pomelo Sweetie': 106, 'Potato Red': 107, 'Potato Red Washed': 108, 'Potato Sweet': 109, 'Potato White': 110, 'Quince': 111, 'Rambutan': 112, 'Raspberry': 113, 'Redcurrant': 114, 'Salak': 115, 'Strawberry': 116, 'Strawberry Wedge': 117, 'Tamarillo': 118, 'Tangelo': 119, 'Tomato 1': 120, 'Tomato 2': 121, 'Tomato 3': 122, 'Tomato 4': 123, 'Tomato Cherry Red': 124, 'Tomato Heart': 125, 'Tomato Maroon': 126, 'Tomato Yellow': 127, 'Tomato not Ripened': 128, 'Walnut': 129, 'Watermelon': 130}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1eKUoqJ9tsDg",
        "outputId": "fcebdaae-86d2-44c3-c905-c9e2c693fcc1"
      },
      "source": [
        "!cat fruit_labels.txt"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Apple Braeburn\n",
            "Apple Crimson Snow\n",
            "Apple Golden 1\n",
            "Apple Golden 2\n",
            "Apple Golden 3\n",
            "Apple Granny Smith\n",
            "Apple Pink Lady\n",
            "Apple Red 1\n",
            "Apple Red 2\n",
            "Apple Red 3\n",
            "Apple Red Delicious\n",
            "Apple Red Yellow 1\n",
            "Apple Red Yellow 2\n",
            "Apricot\n",
            "Avocado\n",
            "Avocado ripe\n",
            "Banana\n",
            "Banana Lady Finger\n",
            "Banana Red\n",
            "Beetroot\n",
            "Blueberry\n",
            "Cactus fruit\n",
            "Cantaloupe 1\n",
            "Cantaloupe 2\n",
            "Carambula\n",
            "Cauliflower\n",
            "Cherry 1\n",
            "Cherry 2\n",
            "Cherry Rainier\n",
            "Cherry Wax Black\n",
            "Cherry Wax Red\n",
            "Cherry Wax Yellow\n",
            "Chestnut\n",
            "Clementine\n",
            "Cocos\n",
            "Corn\n",
            "Corn Husk\n",
            "Cucumber Ripe\n",
            "Cucumber Ripe 2\n",
            "Dates\n",
            "Eggplant\n",
            "Fig\n",
            "Ginger Root\n",
            "Granadilla\n",
            "Grape Blue\n",
            "Grape Pink\n",
            "Grape White\n",
            "Grape White 2\n",
            "Grape White 3\n",
            "Grape White 4\n",
            "Grapefruit Pink\n",
            "Grapefruit White\n",
            "Guava\n",
            "Hazelnut\n",
            "Huckleberry\n",
            "Kaki\n",
            "Kiwi\n",
            "Kohlrabi\n",
            "Kumquats\n",
            "Lemon\n",
            "Lemon Meyer\n",
            "Limes\n",
            "Lychee\n",
            "Mandarine\n",
            "Mango\n",
            "Mango Red\n",
            "Mangostan\n",
            "Maracuja\n",
            "Melon Piel de Sapo\n",
            "Mulberry\n",
            "Nectarine\n",
            "Nectarine Flat\n",
            "Nut Forest\n",
            "Nut Pecan\n",
            "Onion Red\n",
            "Onion Red Peeled\n",
            "Onion White\n",
            "Orange\n",
            "Papaya\n",
            "Passion Fruit\n",
            "Peach\n",
            "Peach 2\n",
            "Peach Flat\n",
            "Pear\n",
            "Pear 2\n",
            "Pear Abate\n",
            "Pear Forelle\n",
            "Pear Kaiser\n",
            "Pear Monster\n",
            "Pear Red\n",
            "Pear Stone\n",
            "Pear Williams\n",
            "Pepino\n",
            "Pepper Green\n",
            "Pepper Orange\n",
            "Pepper Red\n",
            "Pepper Yellow\n",
            "Physalis\n",
            "Physalis with Husk\n",
            "Pineapple\n",
            "Pineapple Mini\n",
            "Pitahaya Red\n",
            "Plum\n",
            "Plum 2\n",
            "Plum 3\n",
            "Pomegranate\n",
            "Pomelo Sweetie\n",
            "Potato Red\n",
            "Potato Red Washed\n",
            "Potato Sweet\n",
            "Potato White\n",
            "Quince\n",
            "Rambutan\n",
            "Raspberry\n",
            "Redcurrant\n",
            "Salak\n",
            "Strawberry\n",
            "Strawberry Wedge\n",
            "Tamarillo\n",
            "Tangelo\n",
            "Tomato 1\n",
            "Tomato 2\n",
            "Tomato 3\n",
            "Tomato 4\n",
            "Tomato Cherry Red\n",
            "Tomato Heart\n",
            "Tomato Maroon\n",
            "Tomato Yellow\n",
            "Tomato not Ripened\n",
            "Walnut\n",
            "Watermelon"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2v4GYQ2btwo3"
      },
      "source": [
        "## Build the model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hKjmUC_rvUt5"
      },
      "source": [
        "### Create the base model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Az-PuiB0tyay",
        "outputId": "bfdab14b-429a-4472-87b1-7e974f0e8c03"
      },
      "source": [
        "IMG_SHAPE = (IMAGE_SIZE, IMAGE_SIZE, 3)\n",
        "\n",
        "# Create the base model from the pre-trained MobileNet V2\n",
        "base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE,\n",
        "                                              include_top=False, \n",
        "                                              weights='imagenet')\n",
        "base_model.trainable = False"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:`input_shape` is undefined or non-square, or `rows` is not in [96, 128, 160, 192, 224]. Weights for input shape (224, 224) will be loaded as the default.\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/mobilenet_v2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_224_no_top.h5\n",
            "9412608/9406464 [==============================] - 0s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_19bNQvjvXHF"
      },
      "source": [
        "### Add a classification head"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2080xt56t051"
      },
      "source": [
        "model = tf.keras.Sequential([\n",
        "  base_model,\n",
        "  tf.keras.layers.Conv2D(filters=32, kernel_size=3, activation='relu'),\n",
        "  tf.keras.layers.Dropout(0.2),\n",
        "  tf.keras.layers.GlobalAveragePooling2D(),\n",
        "  tf.keras.layers.Dense(units=len(labels.split('\\n')), activation='softmax')\n",
        "])"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BRbeLXOlvZxU"
      },
      "source": [
        "### Configure the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XGTfn_POuDAh"
      },
      "source": [
        "model.compile(optimizer='adam', \n",
        "              loss='categorical_crossentropy', \n",
        "              metrics=['accuracy'])"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "CE6XMOfRuEO5",
        "outputId": "e7f04ebb-208c-48a6-e421-bf3d06b09101"
      },
      "source": [
        "model.summary()"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "mobilenetv2_1.00_224 (Functi (None, 4, 4, 1280)        2257984   \n",
            "_________________________________________________________________\n",
            "conv2d (Conv2D)              (None, 2, 2, 32)          368672    \n",
            "_________________________________________________________________\n",
            "dropout (Dropout)            (None, 2, 2, 32)          0         \n",
            "_________________________________________________________________\n",
            "global_average_pooling2d (Gl (None, 32)                0         \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 131)               4323      \n",
            "=================================================================\n",
            "Total params: 2,630,979\n",
            "Trainable params: 372,995\n",
            "Non-trainable params: 2,257,984\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9Md7AkzSuFMl",
        "outputId": "94f7c2ab-a686-4cf4-9bcc-a77577156989"
      },
      "source": [
        "print('Number of trainable weights = {}'.format(len(model.trainable_weights)))"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of trainable weights = 4\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7mHjfEaxvb0V"
      },
      "source": [
        "## Train the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fPdrGZuJuGnJ",
        "outputId": "a410b094-2c3a-4274-bb21-86a7a08adf19"
      },
      "source": [
        "history = model.fit(train_generator,\n",
        "                    steps_per_epoch=len(train_generator), \n",
        "                    epochs=10,\n",
        "                    validation_data=val_generator,\n",
        "                    validation_steps=len(val_generator))"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/10\n",
            "1058/1058 [==============================] - 90s 53ms/step - loss: 1.8755 - accuracy: 0.5774 - val_loss: 0.3641 - val_accuracy: 0.9101\n",
            "Epoch 2/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0828 - accuracy: 0.9765 - val_loss: 0.2915 - val_accuracy: 0.9300\n",
            "Epoch 3/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0569 - accuracy: 0.9826 - val_loss: 0.3511 - val_accuracy: 0.9350\n",
            "Epoch 4/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0453 - accuracy: 0.9863 - val_loss: 0.4108 - val_accuracy: 0.9219\n",
            "Epoch 5/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0434 - accuracy: 0.9868 - val_loss: 0.4279 - val_accuracy: 0.9338\n",
            "Epoch 6/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0379 - accuracy: 0.9889 - val_loss: 0.4040 - val_accuracy: 0.9375\n",
            "Epoch 7/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0291 - accuracy: 0.9911 - val_loss: 0.3863 - val_accuracy: 0.9438\n",
            "Epoch 8/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0338 - accuracy: 0.9903 - val_loss: 0.5382 - val_accuracy: 0.9367\n",
            "Epoch 9/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0319 - accuracy: 0.9901 - val_loss: 0.4649 - val_accuracy: 0.9460\n",
            "Epoch 10/10\n",
            "1058/1058 [==============================] - 54s 51ms/step - loss: 0.0283 - accuracy: 0.9921 - val_loss: 0.3770 - val_accuracy: 0.9483\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SqW3X2QDveRN"
      },
      "source": [
        "### Review the learning curves"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 513
        },
        "id": "Fcsb3cxWuI-1",
        "outputId": "d34ec248-96fc-4472-e36a-b901fc2a821b"
      },
      "source": [
        "acc = history.history['accuracy']\n",
        "val_acc = history.history['val_accuracy']\n",
        "\n",
        "loss = history.history['loss']\n",
        "val_loss = history.history['val_loss']\n",
        "\n",
        "plt.figure(figsize=(8, 8))\n",
        "plt.subplot(2, 1, 1)\n",
        "plt.plot(acc, label='Training Accuracy')\n",
        "plt.plot(val_acc, label='Validation Accuracy')\n",
        "plt.legend(loc='lower right')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.ylim([min(plt.ylim()),1])\n",
        "plt.title('Training and Validation Accuracy')\n",
        "\n",
        "plt.subplot(2, 1, 2)\n",
        "plt.plot(loss, label='Training Loss')\n",
        "plt.plot(val_loss, label='Validation Loss')\n",
        "plt.legend(loc='upper right')\n",
        "plt.ylabel('Cross Entropy')\n",
        "plt.ylim([0,1.0])\n",
        "plt.title('Training and Validation Loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.show()"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 576x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5-9VUXHbvk4X"
      },
      "source": [
        "## Fine tune the base model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vf_JkoD8vmms"
      },
      "source": [
        "### Un-freeze more layers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "o1GUOIYIuJ-N",
        "outputId": "0a35f595-3bb9-4549-cb1a-5a503cc37641"
      },
      "source": [
        "print(\"Number of layers in the base model: \", len(base_model.layers))"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of layers in the base model:  154\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Lh81YSZuuLGV"
      },
      "source": [
        "base_model.trainable = True\n",
        "fine_tune_at = 100\n",
        "\n",
        "# Freeze all the layers before the `fine_tune_at` layer\n",
        "for layer in base_model.layers[:fine_tune_at]:\n",
        "  layer.trainable =  False"
      ],
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gT96Z5YSvq2s"
      },
      "source": [
        "### Reconfigure the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yUaPhNpFuMHR"
      },
      "source": [
        "model.compile(optimizer=tf.keras.optimizers.Adam(1e-5),\n",
        "              loss='categorical_crossentropy',\n",
        "              metrics=['accuracy'])"
      ],
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "j-dFvD_cuNV9",
        "outputId": "35c5dc37-9790-419e-8ef8-3a1fce0ba97a"
      },
      "source": [
        "model.summary()"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "mobilenetv2_1.00_224 (Functi (None, 4, 4, 1280)        2257984   \n",
            "_________________________________________________________________\n",
            "conv2d (Conv2D)              (None, 2, 2, 32)          368672    \n",
            "_________________________________________________________________\n",
            "dropout (Dropout)            (None, 2, 2, 32)          0         \n",
            "_________________________________________________________________\n",
            "global_average_pooling2d (Gl (None, 32)                0         \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 131)               4323      \n",
            "=================================================================\n",
            "Total params: 2,630,979\n",
            "Trainable params: 2,234,435\n",
            "Non-trainable params: 396,544\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XULSDMRmuOM5",
        "outputId": "8f354d65-2058-4a85-e1f5-fdcb0705f9e9"
      },
      "source": [
        "print('Number of trainable weights = {}'.format(len(model.trainable_weights)))"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of trainable weights = 58\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pI92ceQgvs7j"
      },
      "source": [
        "### Continue training"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pS1Z1pfguPdQ",
        "outputId": "3a4d7aac-0dba-4047-a737-4335ac5e28fc"
      },
      "source": [
        "history_fine = model.fit(train_generator,\n",
        "                         steps_per_epoch=len(train_generator), \n",
        "                         epochs=5,\n",
        "                         validation_data=val_generator,\n",
        "                         validation_steps=len(val_generator))"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/5\n",
            "1058/1058 [==============================] - 62s 56ms/step - loss: 6.7897 - accuracy: 0.3573 - val_loss: 1.0206 - val_accuracy: 0.8095\n",
            "Epoch 2/5\n",
            "1058/1058 [==============================] - 58s 55ms/step - loss: 0.6361 - accuracy: 0.8196 - val_loss: 0.6520 - val_accuracy: 0.8633\n",
            "Epoch 3/5\n",
            "1058/1058 [==============================] - 58s 55ms/step - loss: 0.2380 - accuracy: 0.9277 - val_loss: 0.5152 - val_accuracy: 0.9006\n",
            "Epoch 4/5\n",
            "1058/1058 [==============================] - 58s 55ms/step - loss: 0.1081 - accuracy: 0.9661 - val_loss: 0.4363 - val_accuracy: 0.9194\n",
            "Epoch 5/5\n",
            "1058/1058 [==============================] - 59s 55ms/step - loss: 0.0640 - accuracy: 0.9793 - val_loss: 0.3801 - val_accuracy: 0.9343\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ip-1WYeHvuuW"
      },
      "source": [
        "### Review the new learning curves"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 513
        },
        "id": "afgidc7kuQO5",
        "outputId": "77b37572-af1d-4684-9291-2da5dc0e0e93"
      },
      "source": [
        "acc = history_fine.history['accuracy']\n",
        "val_acc = history_fine.history['val_accuracy']\n",
        "\n",
        "loss = history_fine.history['loss']\n",
        "val_loss = history_fine.history['val_loss']\n",
        "\n",
        "plt.figure(figsize=(8, 8))\n",
        "plt.subplot(2, 1, 1)\n",
        "plt.plot(acc, label='Training Accuracy')\n",
        "plt.plot(val_acc, label='Validation Accuracy')\n",
        "plt.legend(loc='lower right')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.ylim([min(plt.ylim()),1])\n",
        "plt.title('Training and Validation Accuracy')\n",
        "\n",
        "plt.subplot(2, 1, 2)\n",
        "plt.plot(loss, label='Training Loss')\n",
        "plt.plot(val_loss, label='Validation Loss')\n",
        "plt.legend(loc='upper right')\n",
        "plt.ylabel('Cross Entropy')\n",
        "plt.ylim([0,1.0])\n",
        "plt.title('Training and Validation Loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.show()"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 576x576 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "F2ALYxNtvza6"
      },
      "source": [
        "## Convert to TFLite"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "cY8KYBH0uRC9",
        "outputId": "bee409ce-ad78-4a56-d444-0ebef6cbff0b"
      },
      "source": [
        "converter = tf.lite.TFLiteConverter.from_keras_model(model)\n",
        "tflite_model = converter.convert()\n",
        "\n",
        "with open('mobilenet_v2_1.0_224.tflite', 'wb') as f:\n",
        "  f.write(tflite_model)"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Assets written to: /tmp/tmp9be3300d/assets\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "anbcAZ22uXeR",
        "outputId": "470827c6-d2ad-46e5-91eb-d8f88d6b31c4"
      },
      "source": [
        "# A generator that provides a representative dataset\n",
        "def representative_data_gen():\n",
        "  dataset_list = tf.data.Dataset.list_files('fruits-360/*/*.jpg')\n",
        "  for i in range(100):\n",
        "    image = next(iter(dataset_list))\n",
        "    image = tf.io.read_file(image)\n",
        "    image = tf.io.decode_jpeg(image, channels=3)\n",
        "    image = tf.image.resize(image, [IMAGE_SIZE, IMAGE_SIZE])\n",
        "    image = tf.cast(image / 255., tf.float32)\n",
        "    image = tf.expand_dims(image, 0)\n",
        "    yield [image]\n",
        "\n",
        "converter = tf.lite.TFLiteConverter.from_keras_model(model)\n",
        "# This enables quantization\n",
        "converter.optimizations = [tf.lite.Optimize.DEFAULT]\n",
        "# This sets the representative dataset for quantization\n",
        "converter.representative_dataset = representative_data_gen\n",
        "# This ensures that if any ops can't be quantized, the converter throws an error\n",
        "converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]\n",
        "# For full integer quantization, though supported types defaults to int8 only, we explicitly declare it for clarity.\n",
        "converter.target_spec.supported_types = [tf.int8]\n",
        "# These set the input and output tensors to uint8 (added in r2.3)\n",
        "converter.inference_input_type = tf.uint8\n",
        "converter.inference_output_type = tf.uint8\n",
        "tflite_model = converter.convert()\n",
        "\n",
        "with open('mobilenet_v2_1.0_224_quant.tflite', 'wb') as f:\n",
        "  f.write(tflite_model)"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Assets written to: /tmp/tmpy6n9w1ly/assets\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Assets written to: /tmp/tmpy6n9w1ly/assets\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pIZFHRZxv2XJ"
      },
      "source": [
        "### Compare the accuracy"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pXKqPWaCuglI",
        "outputId": "6a2253aa-3fc5-4e58-f7d0-2b853bba6384"
      },
      "source": [
        "batch_images, batch_labels = next(val_generator)\n",
        "\n",
        "logits = model(batch_images)\n",
        "prediction = np.argmax(logits, axis=1)\n",
        "truth = np.argmax(batch_labels, axis=1)\n",
        "\n",
        "keras_accuracy = tf.keras.metrics.Accuracy()\n",
        "keras_accuracy(prediction, truth)\n",
        "\n",
        "print(\"Raw model accuracy: {:.3%}\".format(keras_accuracy.result()))"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Raw model accuracy: 96.875%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "IgwXPIczuiF8",
        "outputId": "1653595f-1773-41a8-9d8a-5f0e303cc5bd"
      },
      "source": [
        "def set_input_tensor(interpreter, input):\n",
        "  input_details = interpreter.get_input_details()[0]\n",
        "  tensor_index = input_details['index']\n",
        "  input_tensor = interpreter.tensor(tensor_index)()[0]\n",
        "  # Inputs for the TFLite model must be uint8, so we quantize our input data.\n",
        "  # NOTE: This step is necessary only because we're receiving input data from\n",
        "  # ImageDataGenerator, which rescaled all image data to float [0,1]. When using\n",
        "  # bitmap inputs, they're already uint8 [0,255] so this can be replaced with:\n",
        "  #   input_tensor[:, :] = input\n",
        "  scale, zero_point = input_details['quantization']\n",
        "  input_tensor[:, :] = np.uint8(input / scale + zero_point)\n",
        "\n",
        "def classify_image(interpreter, input):\n",
        "  set_input_tensor(interpreter, input)\n",
        "  interpreter.invoke()\n",
        "  output_details = interpreter.get_output_details()[0]\n",
        "  output = interpreter.get_tensor(output_details['index'])\n",
        "  # Outputs from the TFLite model are uint8, so we dequantize the results:\n",
        "  scale, zero_point = output_details['quantization']\n",
        "  output = scale * (output - zero_point)\n",
        "  top_1 = np.argmax(output)\n",
        "  return top_1\n",
        "\n",
        "interpreter = tf.lite.Interpreter('mobilenet_v2_1.0_224_quant.tflite')\n",
        "interpreter.allocate_tensors()\n",
        "\n",
        "# Collect all inference predictions in a list\n",
        "batch_prediction = []\n",
        "batch_truth = np.argmax(batch_labels, axis=1)\n",
        "\n",
        "for i in range(len(batch_images)):\n",
        "  prediction = classify_image(interpreter, batch_images[i])\n",
        "  batch_prediction.append(prediction)\n",
        "\n",
        "# Compare all predictions to the ground truth\n",
        "tflite_accuracy = tf.keras.metrics.Accuracy()\n",
        "tflite_accuracy(batch_prediction, batch_truth)\n",
        "print(\"Quant TF Lite accuracy: {:.3%}\".format(tflite_accuracy.result()))"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Quant TF Lite accuracy: 89.062%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OxUoQ3p-v5g9"
      },
      "source": [
        "## Compile for the Edge TPU"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yn3ZI5BwujMU",
        "outputId": "af1fa362-89bf-4e43-bb36-87ea5f8f8c82"
      },
      "source": [
        "! curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -\n",
        "\n",
        "! echo \"deb https://packages.cloud.google.com/apt coral-edgetpu-stable main\" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list\n",
        "\n",
        "! sudo apt-get update\n",
        "\n",
        "! sudo apt-get install edgetpu-compiler\t"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current\n",
            "                                 Dload  Upload   Total   Spent    Left  Speed\n",
            "\r  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0\r100  1974  100  1974    0     0  19352      0 --:--:-- --:--:-- --:--:-- 19352\n",
            "OK\n",
            "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main\n",
            "Get:1 http://security.ubuntu.com/ubuntu bionic-security InRelease [88.7 kB]\n",
            "Get:2 https://cloud.r-project.org/bin/linux/ubuntu bionic-cran40/ InRelease [3,626 B]\n",
            "Ign:3 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64  InRelease\n",
            "Get:4 https://cloud.r-project.org/bin/linux/ubuntu bionic-cran40/ Packages [52.7 kB]\n",
            "Get:5 https://packages.cloud.google.com/apt coral-edgetpu-stable InRelease [6,722 B]\n",
            "Get:6 http://ppa.launchpad.net/c2d4u.team/c2d4u4.0+/ubuntu bionic InRelease [15.9 kB]\n",
            "Hit:7 http://archive.ubuntu.com/ubuntu bionic InRelease\n",
            "Ign:8 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  InRelease\n",
            "Hit:9 https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64  Release\n",
            "Hit:10 https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  Release\n",
            "Get:11 http://security.ubuntu.com/ubuntu bionic-security/main amd64 Packages [2,045 kB]\n",
            "Get:12 http://archive.ubuntu.com/ubuntu bionic-updates InRelease [88.7 kB]\n",
            "Get:13 http://security.ubuntu.com/ubuntu bionic-security/universe amd64 Packages [1,402 kB]\n",
            "Hit:14 http://ppa.launchpad.net/cran/libgit2/ubuntu bionic InRelease\n",
            "Get:15 https://packages.cloud.google.com/apt coral-edgetpu-stable/main amd64 Packages [2,174 B]\n",
            "Get:16 http://archive.ubuntu.com/ubuntu bionic-backports InRelease [74.6 kB]\n",
            "Hit:17 http://ppa.launchpad.net/deadsnakes/ppa/ubuntu bionic InRelease\n",
            "Hit:19 http://ppa.launchpad.net/graphics-drivers/ppa/ubuntu bionic InRelease\n",
            "Get:21 http://ppa.launchpad.net/c2d4u.team/c2d4u4.0+/ubuntu bionic/main Sources [1,749 kB]\n",
            "Get:22 http://archive.ubuntu.com/ubuntu bionic-updates/main amd64 Packages [2,475 kB]\n",
            "Get:23 http://archive.ubuntu.com/ubuntu bionic-updates/universe amd64 Packages [2,170 kB]\n",
            "Get:24 http://ppa.launchpad.net/c2d4u.team/c2d4u4.0+/ubuntu bionic/main amd64 Packages [895 kB]\n",
            "Fetched 11.1 MB in 2s (5,188 kB/s)\n",
            "Reading package lists... Done\n",
            "Reading package lists... Done\n",
            "Building dependency tree       \n",
            "Reading state information... Done\n",
            "The following NEW packages will be installed:\n",
            "  edgetpu-compiler\n",
            "0 upgraded, 1 newly installed, 0 to remove and 48 not upgraded.\n",
            "Need to get 7,865 kB of archives.\n",
            "After this operation, 31.2 MB of additional disk space will be used.\n",
            "Get:1 https://packages.cloud.google.com/apt coral-edgetpu-stable/main amd64 edgetpu-compiler amd64 15.0 [7,865 kB]\n",
            "Fetched 7,865 kB in 0s (20.2 MB/s)\n",
            "debconf: unable to initialize frontend: Dialog\n",
            "debconf: (No usable dialog-like program is installed, so the dialog based frontend cannot be used. at /usr/share/perl5/Debconf/FrontEnd/Dialog.pm line 76, <> line 1.)\n",
            "debconf: falling back to frontend: Readline\n",
            "debconf: unable to initialize frontend: Readline\n",
            "debconf: (This frontend requires a controlling tty.)\n",
            "debconf: falling back to frontend: Teletype\n",
            "dpkg-preconfigure: unable to re-open stdin: \n",
            "Selecting previously unselected package edgetpu-compiler.\n",
            "(Reading database ... 160980 files and directories currently installed.)\n",
            "Preparing to unpack .../edgetpu-compiler_15.0_amd64.deb ...\n",
            "Unpacking edgetpu-compiler (15.0) ...\n",
            "Setting up edgetpu-compiler (15.0) ...\n",
            "Processing triggers for libc-bin (2.27-3ubuntu1.2) ...\n",
            "/sbin/ldconfig.real: /usr/local/lib/python3.7/dist-packages/ideep4py/lib/libmkldnn.so.0 is not a symbolic link\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wPFc26ZhumFl",
        "outputId": "4bb6d245-6b83-4f65-9ce3-f19bb139ec61"
      },
      "source": [
        "! edgetpu_compiler mobilenet_v2_1.0_224_quant.tflite"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Edge TPU Compiler version 15.0.340273435\n",
            "\n",
            "Model compiled successfully in 757 ms.\n",
            "\n",
            "Input model: mobilenet_v2_1.0_224_quant.tflite\n",
            "Input size: 3.08MiB\n",
            "Output model: mobilenet_v2_1.0_224_quant_edgetpu.tflite\n",
            "Output size: 3.15MiB\n",
            "On-chip memory used for caching model parameters: 3.36MiB\n",
            "On-chip memory remaining for caching model parameters: 4.45MiB\n",
            "Off-chip memory used for streaming uncached model parameters: 0.00B\n",
            "Number of Edge TPU subgraphs: 1\n",
            "Total number of operations: 72\n",
            "Operation log: mobilenet_v2_1.0_224_quant_edgetpu.log\n",
            "See the operation log file for individual operation details.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IQRa86n_v8le"
      },
      "source": [
        "## Download the model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 17
        },
        "id": "c6BUjuWdumTn",
        "outputId": "491d13ec-0b05-4970-b62a-ec58c1e1273e"
      },
      "source": [
        "from google.colab import files\n",
        "\n",
        "files.download('mobilenet_v2_1.0_224_quant_edgetpu.tflite')\n",
        "files.download('fruit_labels.txt')"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "\n",
              "    async function download(id, filename, size) {\n",
              "      if (!google.colab.kernel.accessAllowed) {\n",
              "        return;\n",
              "      }\n",
              "      const div = document.createElement('div');\n",
              "      const label = document.createElement('label');\n",
              "      label.textContent = `Downloading \"${filename}\": `;\n",
              "      div.appendChild(label);\n",
              "      const progress = document.createElement('progress');\n",
              "      progress.max = size;\n",
              "      div.appendChild(progress);\n",
              "      document.body.appendChild(div);\n",
              "\n",
              "      const buffers = [];\n",
              "      let downloaded = 0;\n",
              "\n",
              "      const channel = await google.colab.kernel.comms.open(id);\n",
              "      // Send a message to notify the kernel that we're ready.\n",
              "      channel.send({})\n",
              "\n",
              "      for await (const message of channel.messages) {\n",
              "        // Send a message to notify the kernel that we're ready.\n",
              "        channel.send({})\n",
              "        if (message.buffers) {\n",
              "          for (const buffer of message.buffers) {\n",
              "            buffers.push(buffer);\n",
              "            downloaded += buffer.byteLength;\n",
              "            progress.value = downloaded;\n",
              "          }\n",
              "        }\n",
              "      }\n",
              "      const blob = new Blob(buffers, {type: 'application/binary'});\n",
              "      const a = document.createElement('a');\n",
              "      a.href = window.URL.createObjectURL(blob);\n",
              "      a.download = filename;\n",
              "      div.appendChild(a);\n",
              "      a.click();\n",
              "      div.remove();\n",
              "    }\n",
              "  "
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "download(\"download_84e00a10-e7dd-48ab-ada6-545bc8c9c121\", \"mobilenet_v2_1.0_224_quant_edgetpu.tflite\", 3297840)"
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "\n",
              "    async function download(id, filename, size) {\n",
              "      if (!google.colab.kernel.accessAllowed) {\n",
              "        return;\n",
              "      }\n",
              "      const div = document.createElement('div');\n",
              "      const label = document.createElement('label');\n",
              "      label.textContent = `Downloading \"${filename}\": `;\n",
              "      div.appendChild(label);\n",
              "      const progress = document.createElement('progress');\n",
              "      progress.max = size;\n",
              "      div.appendChild(progress);\n",
              "      document.body.appendChild(div);\n",
              "\n",
              "      const buffers = [];\n",
              "      let downloaded = 0;\n",
              "\n",
              "      const channel = await google.colab.kernel.comms.open(id);\n",
              "      // Send a message to notify the kernel that we're ready.\n",
              "      channel.send({})\n",
              "\n",
              "      for await (const message of channel.messages) {\n",
              "        // Send a message to notify the kernel that we're ready.\n",
              "        channel.send({})\n",
              "        if (message.buffers) {\n",
              "          for (const buffer of message.buffers) {\n",
              "            buffers.push(buffer);\n",
              "            downloaded += buffer.byteLength;\n",
              "            progress.value = downloaded;\n",
              "          }\n",
              "        }\n",
              "      }\n",
              "      const blob = new Blob(buffers, {type: 'application/binary'});\n",
              "      const a = document.createElement('a');\n",
              "      a.href = window.URL.createObjectURL(blob);\n",
              "      a.download = filename;\n",
              "      div.appendChild(a);\n",
              "      a.click();\n",
              "      div.remove();\n",
              "    }\n",
              "  "
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "download(\"download_1d522fbb-a21b-4bf8-9989-cd0e528a435d\", \"fruit_labels.txt\", 1497)"
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3BXeKEJ6wCO-"
      },
      "source": [
        "## Run the model on the Edge TPU"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9e-_eiU2wELB"
      },
      "source": [
        "You can now run the model on your Coral device with acceleration on the Edge TPU.\n",
        "\n",
        "To get started, try using your `.tflite` model with [this code for image classification with the TensorFlow Lite API](https://github.com/google-coral/tflite/tree/master/python/examples/classification). \n",
        "\n",
        "Just follow the instructions on that page to set up your device, copy the `mobilenet_v2_1.0_224_quant_edgetpu.tflite` and `fruit_labels.txt` files to your Coral Dev Board or device with a Coral Accelerator, and pass it a fruit photo like this:\n",
        "\n",
        "```\n",
        "python3 classify_image.py \\\n",
        "  --model mobilenet_v2_1.0_224_quant_edgetpu.tflite \\\n",
        "  --labels fruit_labels.txt \\\n",
        "  --input fruit.jpg\n",
        "```\n",
        "\n",
        "Check out more examples for running inference at [coral.ai/examples](https://coral.ai/examples/#code-examples/)."
      ]
    }
  ]
}