openmv_mobilenet_v2.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/SingTown/openmv_tensorflow_training_scripts/blob/main/openmv_mobilenet_v2.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "d6CEOrWoWLG8"
      },
      "source": [
        "# Train MobilenetV2 and Save to OpenMV\n",
        "\n",
        "This Code is for TensorFlow 2"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zgYMJFPRWGI5"
      },
      "source": [
        "## Download Cats vs Dogs dataset from github\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xFwlZAzEWaLR",
        "outputId": "04642add-fc4b-4db0-ce6f-31d26ea5a7a5"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cloning into 'cat_dog'...\n",
            "remote: Enumerating objects: 10018, done.\u001b[K\n",
            "remote: Counting objects: 100% (10018/10018), done.\u001b[K\n",
            "remote: Compressing objects: 100% (10015/10015), done.\u001b[K\n",
            "remote: Total 10018 (delta 3), reused 10017 (delta 3), pack-reused 0\u001b[K\n",
            "Receiving objects: 100% (10018/10018), 224.07 MiB | 30.54 MiB/s, done.\n",
            "Resolving deltas: 100% (3/3), done.\n",
            "Checking out files: 100% (10011/10011), done.\n"
          ]
        }
      ],
      "source": [
        "!git clone --branch master --depth=1 https://github.com/haritha91/Cats-Dogs-Classifier---Keras.git cat_dog"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "W4gluAgGWfbP"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "import numpy as np\n",
        "from tensorflow import keras, lite\n",
        "\n",
        "CLASS_NUM = 2 # dog and cat\n",
        "INPUT_SIZE = 96 #image size is 96 * 96 *3"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "I9LKDHKgWhAB",
        "outputId": "0b596725-f95b-4791-d6e2-37b55f355ae2"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Found 8000 images belonging to 2 classes.\n",
            "Found 2000 images belonging to 2 classes.\n"
          ]
        }
      ],
      "source": [
        "from keras.preprocessing.image import ImageDataGenerator\n",
        "\n",
        "train_datagen = ImageDataGenerator(rescale = 1./255, #All the pixel values would be 0-1\n",
        "                    shear_range = 0.2,\n",
        "                    zoom_range = 0.2,\n",
        "                    horizontal_flip = True)\n",
        "\n",
        "test_datagen = ImageDataGenerator(rescale = 1./255)\n",
        "\n",
        "training_set = train_datagen.flow_from_directory('cat_dog/dataset/training_set',\n",
        "                                                 target_size = (INPUT_SIZE, INPUT_SIZE),\n",
        "                                                 batch_size = 32)\n",
        "\n",
        "test_set = test_datagen.flow_from_directory('cat_dog/dataset/test_set',\n",
        "                                            target_size = (INPUT_SIZE, INPUT_SIZE),\n",
        "                                            batch_size = 32)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7xc0UswWW2CI"
      },
      "source": [
        "## Define Model\n",
        "\n",
        "alpha=0.35 for lower size\n",
        "\n",
        "Note: Do NOT use GlobalAveragePooling2D Which is NOT supported by OpenMV"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6oCUM3g9W4Zo",
        "outputId": "1449f216-7b32-4526-c4fd-5cfc2fc5c246"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/mobilenet_v2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_96_no_top.h5\n",
            "2023424/2019640 [==============================] - 0s 0us/step\n",
            "2031616/2019640 [==============================] - 0s 0us/step\n"
          ]
        }
      ],
      "source": [
        "base = keras.applications.MobileNetV2(\n",
        "    include_top=False,\n",
        "    alpha=0.35,\n",
        "    weights=\"imagenet\",\n",
        "    input_shape=(INPUT_SIZE, INPUT_SIZE, 3)\n",
        ")\n",
        "base.trainable = False\n",
        "\n",
        "x = base.output\n",
        "flatten = keras.layers.Flatten()(x) \n",
        "predictions = keras.layers.Dense(CLASS_NUM, activation='softmax')(flatten)\n",
        "model = keras.models.Model(inputs=base.input, outputs=predictions)\n",
        "#print(model.summary())"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Xzjg_Z5_XBwW"
      },
      "source": [
        "## Train model\n",
        "After 1 epoches, accuracy is about 88% "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RvZ1SA0LXFKy",
        "outputId": "a160a077-c2cb-4e3c-bd13-58f3291f6bf6"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "250/250 [==============================] - 72s 276ms/step - loss: 0.5161 - accuracy: 0.8346 - val_loss: 0.4570 - val_accuracy: 0.8695\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7f700988da10>"
            ]
          },
          "metadata": {},
          "execution_count": 5
        }
      ],
      "source": [
        "model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])\n",
        "model.fit(training_set, epochs = 1, validation_data = test_set)\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LMw0ywl2XNd3"
      },
      "source": [
        "## Full Integer Quantization\n",
        "\n",
        "ref: https://www.tensorflow.org/lite/performance/post_training_quantization#full_integer_quantization\n",
        "\n",
        "TODO: use real representative_dataset"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "TXdcxbctXkYi",
        "outputId": "e94489b2-2497-437a-bab2-eee392dd9205"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "INFO:tensorflow:Assets written to: /tmp/tmpd4sco1b9/assets\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/tensorflow/lite/python/convert.py:746: UserWarning: Statistics for quantized inputs were expected, but not specified; continuing anyway.\n",
            "  warnings.warn(\"Statistics for quantized inputs were expected, but not \"\n",
            "WARNING:absl:Buffer deduplication procedure will be skipped when flatbuffer library is not properly loaded\n"
          ]
        }
      ],
      "source": [
        "def representative_dataset():\n",
        "    for _ in range(100):\n",
        "        data = np.random.rand(1, INPUT_SIZE, INPUT_SIZE, 3)\n",
        "        yield [data.astype(np.float32)]\n",
        "\n",
        "# Convert the tflite.\n",
        "converter = lite.TFLiteConverter.from_keras_model(model)\n",
        "converter.optimizations = [lite.Optimize.DEFAULT]\n",
        "converter.representative_dataset = representative_dataset\n",
        "converter.target_spec.supported_ops = [lite.OpsSet.TFLITE_BUILTINS_INT8]\n",
        "converter.inference_input_type = tf.int8\n",
        "converter.inference_output_type = tf.int8\n",
        "tflite_quant_model = converter.convert()\n",
        "\n",
        "# Save the model.\n",
        "with open('trained.tflite', 'wb') as f:\n",
        "  f.write(tflite_quant_model)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Succeed\n",
        "\n",
        "Copy trained.tflite to OpenMV4 H7 Plus"
      ],
      "metadata": {
        "id": "79I1bWyyH9cv"
      }
    }
  ],
  "metadata": {
    "colab": {
      "collapsed_sections": [],
      "name": "openmv_mobilenet_v2.ipynb",
      "provenance": [],
      "authorship_tag": "ABX9TyPWSzEP6JG5tpBFikz/iFuy",
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}