ch06/06_karaage_ai_book_tflite_convert.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "06_karaage_ai_book_tflite_convert.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyPojOjJJ44OfXxILyhBQx+K",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/karaage0703/karaage-ai-book/blob/master/ch06/06_karaage_ai_book_tflite_convert.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dBJYhtDxkQom"
      },
      "source": [
        "# TensorFlow Lite軽量化\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MR8zDuZpcrsa"
      },
      "source": [
        "## 事前準備"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OtYbzcsPkg6e",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "97b6746a-0c97-45fa-c81e-62406783449f"
      },
      "source": [
        "%tensorflow_version 2.x\n",
        "import tensorflow as tf\n",
        "print(tf.__version__)"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2.3.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NAzaEMEZluLy"
      },
      "source": [
        "from tensorflow.keras.preprocessing.image import img_to_array, load_img\n",
        "import numpy as np\n",
        "import time"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UzeTecp8dABE"
      },
      "source": [
        "テスト画像をダウンロードします。"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JFU75I-Xl86d",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "cfee18fe-8565-4044-c495-1e19465555ca"
      },
      "source": [
        "!wget https://raw.githubusercontent.com/karaage0703/janken_dataset/master/choki/choki_01.jpg"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2020-11-22 14:33:15--  https://raw.githubusercontent.com/karaage0703/janken_dataset/master/choki/choki_01.jpg\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.0.133, 151.101.64.133, 151.101.128.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.0.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 27653 (27K) [image/jpeg]\n",
            "Saving to: ‘choki_01.jpg’\n",
            "\n",
            "\rchoki_01.jpg          0%[                    ]       0  --.-KB/s               \rchoki_01.jpg        100%[===================>]  27.00K  --.-KB/s    in 0s      \n",
            "\n",
            "2020-11-22 14:33:16 (123 MB/s) - ‘choki_01.jpg’ saved [27653/27653]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FSTC7kWRqJtU",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 257
        },
        "outputId": "ebafe6a1-b0ea-4288-d6c2-415e7cee78d4"
      },
      "source": [
        "from IPython.display import Image as IPImage\n",
        "from IPython.display import display_jpeg\n",
        "display_jpeg(IPImage('choki_01.jpg'))"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/jpeg": "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\n"
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "A3UvGkLmkjHN"
      },
      "source": [
        "モデルとラベルをGoogle Driveからコピーします。"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "aNBqDFwukme5",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a1176b42-1497-49a1-daf0-dc457d1f2666"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mounted at /content/drive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sRmnz5FjkvyS"
      },
      "source": [
        "!cp '/content/drive/My Drive/karaage-ai-book/my_model.h5' './my_model.h5'\n",
        "!cp '/content/drive/My Drive/karaage-ai-book/efficientnet_model.h5' './efficientnet_model.h5'\n",
        "!cp '/content/drive/My Drive/karaage-ai-book/labels.txt' './labels.txt'"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HAMrF1KhzSwl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a6c6a857-285a-46d7-ca16-265f3434f85b"
      },
      "source": [
        "!ls -lh"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "total 124M\n",
            "-rw-r--r-- 1 root root  28K Nov 22 14:33 choki_01.jpg\n",
            "drwx------ 5 root root 4.0K Nov 22 14:33 drive\n",
            "-rw------- 1 root root  34M Nov 22 14:33 efficientnet_model.h5\n",
            "-rw------- 1 root root   12 Nov 22 14:33 labels.txt\n",
            "-rw------- 1 root root  91M Nov 22 14:33 my_model.h5\n",
            "drwxr-xr-x 1 root root 4.0K Nov 13 17:33 sample_data\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2I4uwnHmsRl7"
      },
      "source": [
        "ラベルを読み込みます"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DL3IxYtmsTAe",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9e920afd-1a4d-4637-9cf0-b0183ea03845"
      },
      "source": [
        "labels = []\n",
        "with open('./labels.txt','r') as f:\n",
        "  for line in f:\n",
        "    labels.append(line.rstrip())\n",
        "print(labels)\n",
        "\n",
        "NUM_CLASSES = len(labels)"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['choki', 'gu', 'pa']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "J8OumIjRdJUj"
      },
      "source": [
        "## モデルの変換"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_Czkj9kElCe9"
      },
      "source": [
        "モデルを読み込みます"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "h8l9f28heJpu"
      },
      "source": [
        "MNISTベースの自作モデルを読み込みます。"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KqTp0rFWzflm"
      },
      "source": [
        "keras_mnist_model = tf.keras.models.load_model(\"my_model.h5\")"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6nNzMfKHmCyD"
      },
      "source": [
        "TFLiteモデルに変換します。"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Vhq-lrIcl2qu",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5379b807-4b3b-4aef-d8f9-32152a134472"
      },
      "source": [
        "converter = tf.lite.TFLiteConverter.from_keras_model(keras_mnist_model)\n",
        "\n",
        "tflite_model = converter.convert()\n",
        "open(\"mnist_model.tflite\", \"wb\").write(tflite_model)"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/tracking/tracking.py:111: Model.state_updates (from tensorflow.python.keras.engine.training) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "This property should not be used in TensorFlow 2.0, as updates are applied automatically.\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/training/tracking/tracking.py:111: Layer.updates (from tensorflow.python.keras.engine.base_layer) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "This property should not be used in TensorFlow 2.0, as updates are applied automatically.\n",
            "INFO:tensorflow:Assets written to: /tmp/tmpw9dr7w4c/assets\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "31572588"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3Dz5nlv_eMEW"
      },
      "source": [
        "EfficientNetモデルを読み込みます。"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hn73mVWZ1pSK",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4f782858-9c36-4637-c65d-7ba08e62207c"
      },
      "source": [
        "# keras_model = tf.keras.models.load_model(\"efficientnet_model.h5\")\n",
        "\n",
        "import tensorflow_hub as hub\n",
        "keras_efficientnet_model = tf.keras.models.load_model('efficientnet_model.h5', custom_objects={'KerasLayer':hub.KerasLayer})"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lEFCHu-w1p39"
      },
      "source": [
        "https://github.com/tensorflow/tensorflow/issues/26835"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aRCz5pd8dUcW"
      },
      "source": [
        "モデルを変換します。"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "flKqGWXok8xt",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "068c7884-4816-46d6-f8d5-bc24fb761721"
      },
      "source": [
        "converter = tf.lite.TFLiteConverter.from_keras_model(keras_efficientnet_model)\n",
        "\n",
        "tflite_model = converter.convert()\n",
        "open(\"efficientnet_model.tflite\", \"wb\").write(tflite_model)"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Assets written to: /tmp/tmpx2bx2m0_/assets\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "INFO:tensorflow:Assets written to: /tmp/tmpx2bx2m0_/assets\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "16113268"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NQRoDZQslQLf"
      },
      "source": [
        "## モデルの性能（推論時間）を比較"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dVve1kGyqYSI"
      },
      "source": [
        "### 軽量化前のモデル"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wTL8ooI2qa54"
      },
      "source": [
        "データの変換"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qP7B_Pz-k-nm"
      },
      "source": [
        "img = img_to_array(load_img('choki_01.jpg', target_size=(64,64)))\n",
        "img = img / 255.0\n",
        "img = img[np.newaxis]\n",
        "img = np.asarray(img)"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pD0ebXA0qcyP"
      },
      "source": [
        "MNISTベースモデルの推論"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fai20ZDsWDz0",
        "outputId": "e25247af-a9f4-4de7-8e4c-0068ed5ef8ae"
      },
      "source": [
        "%%time\n",
        "preds = keras_mnist_model.predict(img)\n",
        "result = np.argmax(preds[0])\n",
        "print(labels[result])"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "choki\n",
            "CPU times: user 48.4 ms, sys: 0 ns, total: 48.4 ms\n",
            "Wall time: 51.1 ms\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Bf3Ah_TSYl8M"
      },
      "source": [
        "EfficientNetモデルの推論"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nJuf0jLjWRds",
        "outputId": "d118b444-38de-423d-bbe3-b19d9dac3b97"
      },
      "source": [
        "%%time\n",
        "preds = keras_efficientnet_model.predict(img)\n",
        "result = np.argmax(preds[0])\n",
        "print(labels[result])"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "choki\n",
            "CPU times: user 50.1 ms, sys: 6.03 ms, total: 56.2 ms\n",
            "Wall time: 55.2 ms\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TiGau3GCqv-Z"
      },
      "source": [
        "### 軽量化したモデル"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1XggpyrPmq56"
      },
      "source": [
        "MNISTモデルのTFLiteファイルを読み込みます。"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YiBMrGlPoCmj"
      },
      "source": [
        "interpreter = tf.lite.Interpreter(model_path=\"mnist_model.tflite\")\n",
        "interpreter.allocate_tensors()\n",
        "input_details = interpreter.get_input_details()\n",
        "output_details = interpreter.get_output_details()"
      ],
      "execution_count": 28,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nt-KgsYUq5ik",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "0afd3a97-1e4a-4dd4-b3c6-f8b0b2df2119"
      },
      "source": [
        "print(input_details)"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[{'name': 'conv2d_2_input', 'index': 0, 'shape': array([ 1, 64, 64,  3], dtype=int32), 'shape_signature': array([-1, 64, 64,  3], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Wf_I7_5Xq8zf",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4a219141-b5f7-4d42-966a-bcdabca68487"
      },
      "source": [
        "print(output_details)"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[{'name': 'Identity', 'index': 16, 'shape': array([1, 3], dtype=int32), 'shape_signature': array([-1,  3], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uBadLvnArFE0"
      },
      "source": [
        "データの変換"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1jXIUhj9q3Op"
      },
      "source": [
        "img = img_to_array(load_img('choki_01.jpg', target_size=(64,64)))\n",
        "img = img / 255.0\n",
        "img = img[np.newaxis]\n",
        "img = np.asarray(img)\n",
        "input_tensor = tf.convert_to_tensor(img)"
      ],
      "execution_count": 31,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sL_ghlcwYz9u"
      },
      "source": [
        "推論"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "u0Z_hqsJW8p_",
        "outputId": "330cd1ac-a080-41e0-84d3-8a4992b2747d"
      },
      "source": [
        "%%time\n",
        "# set input tensor\n",
        "interpreter.set_tensor(input_details[0]['index'], input_tensor)\n",
        "\n",
        "# Inference\n",
        "interpreter.invoke()\n",
        "\n",
        "scores = interpreter.get_tensor(output_details[0]['index'])\n",
        "result = np.argmax(scores[0])\n",
        "print(labels[result])"
      ],
      "execution_count": 34,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "choki\n",
            "CPU times: user 10.9 ms, sys: 79 µs, total: 11 ms\n",
            "Wall time: 11.8 ms\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ixxm34kEm1De"
      },
      "source": [
        "EfficientNetモデルのTFLiteファイルを読み込みます。"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8pG4Wsqym4UR"
      },
      "source": [
        "interpreter = tf.lite.Interpreter(model_path=\"efficientnet_model.tflite\")\n",
        "interpreter.allocate_tensors()\n",
        "input_details = interpreter.get_input_details()\n",
        "output_details = interpreter.get_output_details()"
      ],
      "execution_count": 35,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UtuVhVcZXQYw"
      },
      "source": [
        "推論"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bnKtO0xVXPv-",
        "outputId": "fc878c5c-d9ec-476d-aedd-bfd732577e89"
      },
      "source": [
        "%%time\n",
        "# set input tensor\n",
        "interpreter.set_tensor(input_details[0]['index'], input_tensor)\n",
        "\n",
        "# Inference\n",
        "interpreter.invoke()\n",
        "\n",
        "scores = interpreter.get_tensor(output_details[0]['index'])\n",
        "result = np.argmax(scores[0])\n",
        "print(labels[result])"
      ],
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "choki\n",
            "CPU times: user 12.6 ms, sys: 0 ns, total: 12.6 ms\n",
            "Wall time: 10.9 ms\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "O257maL9kW7k"
      },
      "source": [
        "# 参考リンク\n",
        "\n",
        "- https://qiita.com/iwatake2222/items/d63aa67e5c700fcea70a\n",
        "- https://qiita.com/PINTO/items/008c54536fca690e0572#4-2-5-tensorflowkeras-h5json-%E3%81%8B%E3%82%89%E3%81%AE%E9%87%8F%E5%AD%90%E5%8C%96"
      ]
    }
  ]
}