keypoint_classification_EN.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "id": "igMyGnjE9hEp"
   },
   "outputs": [],
   "source": [
    "import csv\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "RANDOM_SEED = 42"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "t2HDvhIu9hEr"
   },
   "source": [
    "# Specify each path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "id": "9NvZP2Zn9hEy"
   },
   "outputs": [],
   "source": [
    "dataset = 'model/keypoint_classifier/keypoint.csv'\n",
    "model_save_path = 'model/keypoint_classifier/keypoint_classifier.hdf5'\n",
    "tflite_save_path = 'model/keypoint_classifier/keypoint_classifier.tflite'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "s5oMH7x19hEz"
   },
   "source": [
    "# Set number of classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "id": "du4kodXL9hEz"
   },
   "outputs": [],
   "source": [
    "NUM_CLASSES = 71"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "XjnL0uso9hEz"
   },
   "source": [
    "# Dataset reading"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "id": "QT5ZqtEz9hE0"
   },
   "outputs": [],
   "source": [
    "X_dataset = np.loadtxt(dataset, delimiter=',', dtype='float32', usecols=list(range(1, (21*2 + 21*2 + 2*1) + 1)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "id": "QmoKFsp49hE0"
   },
   "outputs": [],
   "source": [
    "y_dataset = np.loadtxt(dataset, delimiter=',', dtype='int32', usecols=(0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "id": "xQU7JTZ_9hE0"
   },
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X_dataset, y_dataset, train_size=0.75, random_state=RANDOM_SEED)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "mxK_lETT9hE0"
   },
   "source": [
    "# Model building"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "id": "vHBmUf1t9hE1"
   },
   "outputs": [],
   "source": [
    "# model = tf.keras.models.Sequential([\n",
    "#     tf.keras.layers.Input((21*2 + 21*2 + 2*1 )),\n",
    "#     tf.keras.layers.Dropout(0.2),\n",
    "#     tf.keras.layers.Dense(20, activation='relu'),\n",
    "#     tf.keras.layers.Dropout(0.4),\n",
    "#     tf.keras.layers.Dense(10, activation='relu'),\n",
    "#     tf.keras.layers.Dense(NUM_CLASSES, activation='softmax')\n",
    "# ])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = tf.keras.models.Sequential([\n",
    "    tf.keras.layers.Input((21*2 + 21*2 + 2*1 )),\n",
    "    tf.keras.layers.Dense(90, activation='relu'),\n",
    "    tf.keras.layers.Dropout(0.2),\n",
    "    tf.keras.layers.Dense(120, activation='relu'),\n",
    "    tf.keras.layers.Dropout(0.4),\n",
    "    tf.keras.layers.Dense(50, activation='relu'),\n",
    "    tf.keras.layers.Dense(NUM_CLASSES, activation='softmax')\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model = tf.keras.models.Sequential([\n",
    "#     tf.keras.layers.Input((21*2 + 21*2 + 2*1 )),\n",
    "#     tf.keras.layers.Dropout(0.2),\n",
    "#     tf.keras.layers.Dense(100, activation='relu'),\n",
    "#     tf.keras.layers.Dropout(0.4),\n",
    "#     tf.keras.layers.Dense(50, activation='relu'),\n",
    "#     tf.keras.layers.Dense(NUM_CLASSES, activation='softmax')\n",
    "# ])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "ypqky9tc9hE1",
    "outputId": "5db082bb-30e3-4110-bf63-a1ee777ecd46"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " dense_4 (Dense)             (None, 90)                7830      \n",
      "                                                                 \n",
      " dropout_2 (Dropout)         (None, 90)                0         \n",
      "                                                                 \n",
      " dense_5 (Dense)             (None, 120)               10920     \n",
      "                                                                 \n",
      " dropout_3 (Dropout)         (None, 120)               0         \n",
      "                                                                 \n",
      " dense_6 (Dense)             (None, 50)                6050      \n",
      "                                                                 \n",
      " dense_7 (Dense)             (None, 71)                3621      \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 28421 (111.02 KB)\n",
      "Trainable params: 28421 (111.02 KB)\n",
      "Non-trainable params: 0 (0.00 Byte)\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()  # tf.keras.utils.plot_model(model, show_shapes=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "id": "MbMjOflQ9hE1"
   },
   "outputs": [],
   "source": [
    "# Model checkpoint callback\n",
    "cp_callback = tf.keras.callbacks.ModelCheckpoint(\n",
    "    model_save_path, verbose=1, save_weights_only=False)\n",
    "# Callback for early stopping\n",
    "es_callback = tf.keras.callbacks.EarlyStopping(patience=20, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "id": "c3Dac0M_9hE2"
   },
   "outputs": [],
   "source": [
    "# Model compilation\n",
    "model.compile(\n",
    "    optimizer='adam',\n",
    "    loss='sparse_categorical_crossentropy',\n",
    "    metrics=['accuracy']\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "7XI0j1Iu9hE2"
   },
   "source": [
    "# Model training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "WirBl-JE9hE3",
    "outputId": "71b30ca2-8294-4d9d-8aa2-800d90d399de",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 4.0612 - accuracy: 0.0487\n",
      "Epoch 1: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 3s 23ms/step - loss: 4.0452 - accuracy: 0.0510 - val_loss: 3.5061 - val_accuracy: 0.1159\n",
      "Epoch 2/3550\n",
      "12/50 [======>.......................] - ETA: 0s - loss: 3.5221 - accuracy: 0.1230"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Sanjay\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\keras\\src\\engine\\training.py:3000: UserWarning: You are saving your model as an HDF5 file via `model.save()`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')`.\n",
      "  saving_api.save_model(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "45/50 [==========================>...] - ETA: 0s - loss: 3.1740 - accuracy: 0.1550\n",
      "Epoch 2: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 3.1299 - accuracy: 0.1595 - val_loss: 2.3816 - val_accuracy: 0.3250\n",
      "Epoch 3/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 2.3964 - accuracy: 0.2751\n",
      "Epoch 3: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 2.3878 - accuracy: 0.2752 - val_loss: 1.8129 - val_accuracy: 0.5047\n",
      "Epoch 4/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 1.9943 - accuracy: 0.3676\n",
      "Epoch 4: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 1.9943 - accuracy: 0.3676 - val_loss: 1.4936 - val_accuracy: 0.6098\n",
      "Epoch 5/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 1.7288 - accuracy: 0.4410\n",
      "Epoch 5: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 1.7252 - accuracy: 0.4418 - val_loss: 1.2766 - val_accuracy: 0.6318\n",
      "Epoch 6/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 1.5282 - accuracy: 0.4927\n",
      "Epoch 6: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 1.5282 - accuracy: 0.4927 - val_loss: 1.1113 - val_accuracy: 0.6876\n",
      "Epoch 7/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 1.3755 - accuracy: 0.5394\n",
      "Epoch 7: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 1.3752 - accuracy: 0.5399 - val_loss: 1.0008 - val_accuracy: 0.7322\n",
      "Epoch 8/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 1.2651 - accuracy: 0.5844\n",
      "Epoch 8: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 1.2652 - accuracy: 0.5877 - val_loss: 0.9211 - val_accuracy: 0.7528\n",
      "Epoch 9/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 1.1628 - accuracy: 0.6141\n",
      "Epoch 9: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 1.1628 - accuracy: 0.6141 - val_loss: 0.8121 - val_accuracy: 0.7739\n",
      "Epoch 10/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 1.0825 - accuracy: 0.6445\n",
      "Epoch 10: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 1.0745 - accuracy: 0.6462 - val_loss: 0.7500 - val_accuracy: 0.7955\n",
      "Epoch 11/3550\n",
      "44/50 [=========================>....] - ETA: 0s - loss: 1.0058 - accuracy: 0.6648\n",
      "Epoch 11: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 1.0077 - accuracy: 0.6656 - val_loss: 0.6851 - val_accuracy: 0.7913\n",
      "Epoch 12/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.9150 - accuracy: 0.6968\n",
      "Epoch 12: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.9155 - accuracy: 0.6964 - val_loss: 0.6252 - val_accuracy: 0.8246\n",
      "Epoch 13/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.8659 - accuracy: 0.7122\n",
      "Epoch 13: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.8618 - accuracy: 0.7129 - val_loss: 0.5830 - val_accuracy: 0.8372\n",
      "Epoch 14/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.8221 - accuracy: 0.7114\n",
      "Epoch 14: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.8197 - accuracy: 0.7129 - val_loss: 0.5426 - val_accuracy: 0.8518\n",
      "Epoch 15/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.7628 - accuracy: 0.7420\n",
      "Epoch 15: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.7580 - accuracy: 0.7430 - val_loss: 0.5083 - val_accuracy: 0.8555\n",
      "Epoch 16/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.7279 - accuracy: 0.7475\n",
      "Epoch 16: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.7265 - accuracy: 0.7478 - val_loss: 0.4743 - val_accuracy: 0.8630\n",
      "Epoch 17/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.6852 - accuracy: 0.7633\n",
      "Epoch 17: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.6803 - accuracy: 0.7639 - val_loss: 0.4426 - val_accuracy: 0.8870\n",
      "Epoch 18/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.6427 - accuracy: 0.7785\n",
      "Epoch 18: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.6493 - accuracy: 0.7766 - val_loss: 0.4349 - val_accuracy: 0.8823\n",
      "Epoch 19/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.6267 - accuracy: 0.7842\n",
      "Epoch 19: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.6255 - accuracy: 0.7836 - val_loss: 0.4106 - val_accuracy: 0.8818\n",
      "Epoch 20/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.5933 - accuracy: 0.7938\n",
      "Epoch 20: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.5933 - accuracy: 0.7938 - val_loss: 0.3843 - val_accuracy: 0.9024\n",
      "Epoch 21/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.5632 - accuracy: 0.8016\n",
      "Epoch 21: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.5636 - accuracy: 0.8033 - val_loss: 0.3666 - val_accuracy: 0.9081\n",
      "Epoch 22/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.5344 - accuracy: 0.8187\n",
      "Epoch 22: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.5334 - accuracy: 0.8188 - val_loss: 0.3470 - val_accuracy: 0.9095\n",
      "Epoch 23/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.5167 - accuracy: 0.8200\n",
      "Epoch 23: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.5192 - accuracy: 0.8191 - val_loss: 0.3339 - val_accuracy: 0.9132\n",
      "Epoch 24/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.5021 - accuracy: 0.8256\n",
      "Epoch 24: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.5018 - accuracy: 0.8238 - val_loss: 0.3194 - val_accuracy: 0.9174\n",
      "Epoch 25/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.4872 - accuracy: 0.8330\n",
      "Epoch 25: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.4872 - accuracy: 0.8330 - val_loss: 0.3071 - val_accuracy: 0.9137\n",
      "Epoch 26/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.4654 - accuracy: 0.8371\n",
      "Epoch 26: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.4637 - accuracy: 0.8379 - val_loss: 0.3032 - val_accuracy: 0.9090\n",
      "Epoch 27/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.4484 - accuracy: 0.8438\n",
      "Epoch 27: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.4530 - accuracy: 0.8426 - val_loss: 0.2788 - val_accuracy: 0.9329\n",
      "Epoch 28/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.4335 - accuracy: 0.8480\n",
      "Epoch 28: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.4359 - accuracy: 0.8463 - val_loss: 0.2686 - val_accuracy: 0.9325\n",
      "Epoch 29/3550\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "46/50 [==========================>...] - ETA: 0s - loss: 0.4192 - accuracy: 0.8544\n",
      "Epoch 29: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.4172 - accuracy: 0.8552 - val_loss: 0.2555 - val_accuracy: 0.9381\n",
      "Epoch 30/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.4107 - accuracy: 0.8614\n",
      "Epoch 30: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.4076 - accuracy: 0.8598 - val_loss: 0.2428 - val_accuracy: 0.9423\n",
      "Epoch 31/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.4028 - accuracy: 0.8564\n",
      "Epoch 31: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.4001 - accuracy: 0.8582 - val_loss: 0.2577 - val_accuracy: 0.9240\n",
      "Epoch 32/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.3902 - accuracy: 0.8624\n",
      "Epoch 32: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.3900 - accuracy: 0.8621 - val_loss: 0.2463 - val_accuracy: 0.9212\n",
      "Epoch 33/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.3740 - accuracy: 0.8666\n",
      "Epoch 33: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.3740 - accuracy: 0.8666 - val_loss: 0.2391 - val_accuracy: 0.9475\n",
      "Epoch 34/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.3606 - accuracy: 0.8731\n",
      "Epoch 34: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.3661 - accuracy: 0.8704 - val_loss: 0.2289 - val_accuracy: 0.9320\n",
      "Epoch 35/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.3617 - accuracy: 0.8737\n",
      "Epoch 35: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 16ms/step - loss: 0.3585 - accuracy: 0.8734 - val_loss: 0.2230 - val_accuracy: 0.9432\n",
      "Epoch 36/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.3462 - accuracy: 0.8786\n",
      "Epoch 36: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.3443 - accuracy: 0.8790 - val_loss: 0.2051 - val_accuracy: 0.9545\n",
      "Epoch 37/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.3292 - accuracy: 0.8857\n",
      "Epoch 37: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.3287 - accuracy: 0.8862 - val_loss: 0.2084 - val_accuracy: 0.9451\n",
      "Epoch 38/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.3182 - accuracy: 0.8913\n",
      "Epoch 38: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.3219 - accuracy: 0.8906 - val_loss: 0.2045 - val_accuracy: 0.9465\n",
      "Epoch 39/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.3152 - accuracy: 0.8931\n",
      "Epoch 39: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.3168 - accuracy: 0.8923 - val_loss: 0.1986 - val_accuracy: 0.9554\n",
      "Epoch 40/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.3083 - accuracy: 0.8895\n",
      "Epoch 40: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.3087 - accuracy: 0.8890 - val_loss: 0.1929 - val_accuracy: 0.9447\n",
      "Epoch 41/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.2998 - accuracy: 0.8967\n",
      "Epoch 41: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2998 - accuracy: 0.8963 - val_loss: 0.1875 - val_accuracy: 0.9442\n",
      "Epoch 42/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.3012 - accuracy: 0.8932\n",
      "Epoch 42: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.3017 - accuracy: 0.8931 - val_loss: 0.1926 - val_accuracy: 0.9461\n",
      "Epoch 43/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.2956 - accuracy: 0.8976\n",
      "Epoch 43: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2953 - accuracy: 0.8965 - val_loss: 0.1982 - val_accuracy: 0.9432\n",
      "Epoch 44/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.2906 - accuracy: 0.8939\n",
      "Epoch 44: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2927 - accuracy: 0.8937 - val_loss: 0.1804 - val_accuracy: 0.9498\n",
      "Epoch 45/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.2870 - accuracy: 0.9012\n",
      "Epoch 45: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2870 - accuracy: 0.9012 - val_loss: 0.1864 - val_accuracy: 0.9475\n",
      "Epoch 46/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.2790 - accuracy: 0.9012\n",
      "Epoch 46: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2790 - accuracy: 0.9012 - val_loss: 0.1837 - val_accuracy: 0.9522\n",
      "Epoch 47/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.2730 - accuracy: 0.9024\n",
      "Epoch 47: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 16ms/step - loss: 0.2730 - accuracy: 0.9024 - val_loss: 0.1762 - val_accuracy: 0.9536\n",
      "Epoch 48/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.2884 - accuracy: 0.9020\n",
      "Epoch 48: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2833 - accuracy: 0.9037 - val_loss: 0.1728 - val_accuracy: 0.9592\n",
      "Epoch 49/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.2649 - accuracy: 0.9104\n",
      "Epoch 49: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2627 - accuracy: 0.9099 - val_loss: 0.1711 - val_accuracy: 0.9508\n",
      "Epoch 50/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.2705 - accuracy: 0.9079\n",
      "Epoch 50: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2685 - accuracy: 0.9079 - val_loss: 0.1592 - val_accuracy: 0.9611\n",
      "Epoch 51/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.2443 - accuracy: 0.9137\n",
      "Epoch 51: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.2451 - accuracy: 0.9131 - val_loss: 0.1559 - val_accuracy: 0.9644\n",
      "Epoch 52/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.2405 - accuracy: 0.9173\n",
      "Epoch 52: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.2397 - accuracy: 0.9174 - val_loss: 0.1524 - val_accuracy: 0.9648\n",
      "Epoch 53/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.2376 - accuracy: 0.9158\n",
      "Epoch 53: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.2386 - accuracy: 0.9148 - val_loss: 0.1619 - val_accuracy: 0.9634\n",
      "Epoch 54/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.2268 - accuracy: 0.9201\n",
      "Epoch 54: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.2273 - accuracy: 0.9196 - val_loss: 0.1492 - val_accuracy: 0.9667\n",
      "Epoch 55/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.2405 - accuracy: 0.9124\n",
      "Epoch 55: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.2423 - accuracy: 0.9121 - val_loss: 0.1594 - val_accuracy: 0.9545\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 56/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.2260 - accuracy: 0.9215\n",
      "Epoch 56: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.2275 - accuracy: 0.9207 - val_loss: 0.1513 - val_accuracy: 0.9634\n",
      "Epoch 57/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.2121 - accuracy: 0.9264\n",
      "Epoch 57: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.2121 - accuracy: 0.9256 - val_loss: 0.1489 - val_accuracy: 0.9620\n",
      "Epoch 58/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.2139 - accuracy: 0.9243\n",
      "Epoch 58: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 16ms/step - loss: 0.2144 - accuracy: 0.9240 - val_loss: 0.1531 - val_accuracy: 0.9526\n",
      "Epoch 59/3550\n",
      "40/50 [=======================>......] - ETA: 0s - loss: 0.2200 - accuracy: 0.9275\n",
      "Epoch 59: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 0s 7ms/step - loss: 0.2212 - accuracy: 0.9239 - val_loss: 0.1424 - val_accuracy: 0.9681\n",
      "Epoch 60/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.2177 - accuracy: 0.9217\n",
      "Epoch 60: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 12ms/step - loss: 0.2192 - accuracy: 0.9210 - val_loss: 0.1459 - val_accuracy: 0.9568\n",
      "Epoch 61/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.2141 - accuracy: 0.9255\n",
      "Epoch 61: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.2128 - accuracy: 0.9259 - val_loss: 0.1495 - val_accuracy: 0.9629\n",
      "Epoch 62/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.2087 - accuracy: 0.9217\n",
      "Epoch 62: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 12ms/step - loss: 0.2070 - accuracy: 0.9225 - val_loss: 0.1439 - val_accuracy: 0.9700\n",
      "Epoch 63/3550\n",
      "43/50 [========================>.....] - ETA: 0s - loss: 0.2056 - accuracy: 0.9284\n",
      "Epoch 63: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 13ms/step - loss: 0.2012 - accuracy: 0.9303 - val_loss: 0.1382 - val_accuracy: 0.9662\n",
      "Epoch 64/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.2002 - accuracy: 0.9302\n",
      "Epoch 64: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 13ms/step - loss: 0.1992 - accuracy: 0.9306 - val_loss: 0.1460 - val_accuracy: 0.9648\n",
      "Epoch 65/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1920 - accuracy: 0.9345\n",
      "Epoch 65: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 13ms/step - loss: 0.1912 - accuracy: 0.9345 - val_loss: 0.1348 - val_accuracy: 0.9695\n",
      "Epoch 66/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1943 - accuracy: 0.9310\n",
      "Epoch 66: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1948 - accuracy: 0.9307 - val_loss: 0.1343 - val_accuracy: 0.9615\n",
      "Epoch 67/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1979 - accuracy: 0.9314\n",
      "Epoch 67: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1980 - accuracy: 0.9315 - val_loss: 0.1361 - val_accuracy: 0.9705\n",
      "Epoch 68/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1848 - accuracy: 0.9343\n",
      "Epoch 68: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1848 - accuracy: 0.9343 - val_loss: 0.1308 - val_accuracy: 0.9728\n",
      "Epoch 69/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.1806 - accuracy: 0.9375\n",
      "Epoch 69: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1827 - accuracy: 0.9370 - val_loss: 0.1339 - val_accuracy: 0.9719\n",
      "Epoch 70/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1910 - accuracy: 0.9299\n",
      "Epoch 70: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1900 - accuracy: 0.9303 - val_loss: 0.1349 - val_accuracy: 0.9737\n",
      "Epoch 71/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1831 - accuracy: 0.9372\n",
      "Epoch 71: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1843 - accuracy: 0.9370 - val_loss: 0.1330 - val_accuracy: 0.9672\n",
      "Epoch 72/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1753 - accuracy: 0.9375\n",
      "Epoch 72: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1777 - accuracy: 0.9371 - val_loss: 0.1290 - val_accuracy: 0.9714\n",
      "Epoch 73/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1824 - accuracy: 0.9348\n",
      "Epoch 73: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1817 - accuracy: 0.9345 - val_loss: 0.1232 - val_accuracy: 0.9723\n",
      "Epoch 74/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1786 - accuracy: 0.9362\n",
      "Epoch 74: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1786 - accuracy: 0.9362 - val_loss: 0.1372 - val_accuracy: 0.9676\n",
      "Epoch 75/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1778 - accuracy: 0.9356\n",
      "Epoch 75: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1779 - accuracy: 0.9351 - val_loss: 0.1275 - val_accuracy: 0.9737\n",
      "Epoch 76/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1781 - accuracy: 0.9354\n",
      "Epoch 76: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1799 - accuracy: 0.9346 - val_loss: 0.1205 - val_accuracy: 0.9775\n",
      "Epoch 77/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.1631 - accuracy: 0.9457\n",
      "Epoch 77: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1639 - accuracy: 0.9451 - val_loss: 0.1362 - val_accuracy: 0.9606\n",
      "Epoch 78/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.1786 - accuracy: 0.9370\n",
      "Epoch 78: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1753 - accuracy: 0.9379 - val_loss: 0.1149 - val_accuracy: 0.9765\n",
      "Epoch 79/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1567 - accuracy: 0.9446\n",
      "Epoch 79: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1571 - accuracy: 0.9436 - val_loss: 0.1253 - val_accuracy: 0.9761\n",
      "Epoch 80/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1719 - accuracy: 0.9429\n",
      "Epoch 80: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 16ms/step - loss: 0.1719 - accuracy: 0.9429 - val_loss: 0.1187 - val_accuracy: 0.9765\n",
      "Epoch 81/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1586 - accuracy: 0.9466\n",
      "Epoch 81: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1573 - accuracy: 0.9470 - val_loss: 0.1223 - val_accuracy: 0.9686\n",
      "Epoch 82/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.1609 - accuracy: 0.9431\n",
      "Epoch 82: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1598 - accuracy: 0.9431 - val_loss: 0.1236 - val_accuracy: 0.9695\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 83/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1557 - accuracy: 0.9448\n",
      "Epoch 83: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1561 - accuracy: 0.9445 - val_loss: 0.1224 - val_accuracy: 0.9780\n",
      "Epoch 84/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1544 - accuracy: 0.9461\n",
      "Epoch 84: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1543 - accuracy: 0.9457 - val_loss: 0.1095 - val_accuracy: 0.9756\n",
      "Epoch 85/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1573 - accuracy: 0.9476\n",
      "Epoch 85: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1566 - accuracy: 0.9479 - val_loss: 0.1179 - val_accuracy: 0.9775\n",
      "Epoch 86/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.1359 - accuracy: 0.9501\n",
      "Epoch 86: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1347 - accuracy: 0.9514 - val_loss: 0.1098 - val_accuracy: 0.9808\n",
      "Epoch 87/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1464 - accuracy: 0.9485\n",
      "Epoch 87: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1468 - accuracy: 0.9484 - val_loss: 0.1140 - val_accuracy: 0.9803\n",
      "Epoch 88/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1516 - accuracy: 0.9492\n",
      "Epoch 88: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1516 - accuracy: 0.9492 - val_loss: 0.1067 - val_accuracy: 0.9808\n",
      "Epoch 89/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1449 - accuracy: 0.9514\n",
      "Epoch 89: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1438 - accuracy: 0.9520 - val_loss: 0.1057 - val_accuracy: 0.9817\n",
      "Epoch 90/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1461 - accuracy: 0.9461\n",
      "Epoch 90: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1474 - accuracy: 0.9456 - val_loss: 0.1062 - val_accuracy: 0.9822\n",
      "Epoch 91/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1355 - accuracy: 0.9556\n",
      "Epoch 91: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1347 - accuracy: 0.9558 - val_loss: 0.1048 - val_accuracy: 0.9822\n",
      "Epoch 92/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1460 - accuracy: 0.9530\n",
      "Epoch 92: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1440 - accuracy: 0.9533 - val_loss: 0.1041 - val_accuracy: 0.9836\n",
      "Epoch 93/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.1371 - accuracy: 0.9552\n",
      "Epoch 93: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1376 - accuracy: 0.9554 - val_loss: 0.1011 - val_accuracy: 0.9798\n",
      "Epoch 94/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.1277 - accuracy: 0.9596\n",
      "Epoch 94: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1301 - accuracy: 0.9584 - val_loss: 0.1077 - val_accuracy: 0.9826\n",
      "Epoch 95/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1368 - accuracy: 0.9528\n",
      "Epoch 95: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1400 - accuracy: 0.9517 - val_loss: 0.0994 - val_accuracy: 0.9812\n",
      "Epoch 96/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1333 - accuracy: 0.9558\n",
      "Epoch 96: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1332 - accuracy: 0.9561 - val_loss: 0.1091 - val_accuracy: 0.9812\n",
      "Epoch 97/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1318 - accuracy: 0.9539\n",
      "Epoch 97: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1311 - accuracy: 0.9540 - val_loss: 0.1006 - val_accuracy: 0.9826\n",
      "Epoch 98/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1244 - accuracy: 0.9570\n",
      "Epoch 98: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1244 - accuracy: 0.9570 - val_loss: 0.1012 - val_accuracy: 0.9841\n",
      "Epoch 99/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1226 - accuracy: 0.9613\n",
      "Epoch 99: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1221 - accuracy: 0.9611 - val_loss: 0.0961 - val_accuracy: 0.9841\n",
      "Epoch 100/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.1262 - accuracy: 0.9573\n",
      "Epoch 100: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 13ms/step - loss: 0.1285 - accuracy: 0.9575 - val_loss: 0.0907 - val_accuracy: 0.9859\n",
      "Epoch 101/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1213 - accuracy: 0.9589\n",
      "Epoch 101: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1205 - accuracy: 0.9592 - val_loss: 0.1000 - val_accuracy: 0.9808\n",
      "Epoch 102/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1210 - accuracy: 0.9587\n",
      "Epoch 102: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1203 - accuracy: 0.9586 - val_loss: 0.0956 - val_accuracy: 0.9841\n",
      "Epoch 103/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1319 - accuracy: 0.9560\n",
      "Epoch 103: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1318 - accuracy: 0.9568 - val_loss: 0.0985 - val_accuracy: 0.9831\n",
      "Epoch 104/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.1222 - accuracy: 0.9572\n",
      "Epoch 104: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1217 - accuracy: 0.9576 - val_loss: 0.1016 - val_accuracy: 0.9826\n",
      "Epoch 105/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1148 - accuracy: 0.9620\n",
      "Epoch 105: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1148 - accuracy: 0.9620 - val_loss: 0.0914 - val_accuracy: 0.9850\n",
      "Epoch 106/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1133 - accuracy: 0.9627\n",
      "Epoch 106: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1126 - accuracy: 0.9626 - val_loss: 0.0935 - val_accuracy: 0.9836\n",
      "Epoch 107/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1006 - accuracy: 0.9655\n",
      "Epoch 107: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1031 - accuracy: 0.9647 - val_loss: 0.0902 - val_accuracy: 0.9859\n",
      "Epoch 108/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1102 - accuracy: 0.9662\n",
      "Epoch 108: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1102 - accuracy: 0.9662 - val_loss: 0.0940 - val_accuracy: 0.9831\n",
      "Epoch 109/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.1116 - accuracy: 0.9622\n",
      "Epoch 109: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1124 - accuracy: 0.9611 - val_loss: 0.0970 - val_accuracy: 0.9845\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 110/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1191 - accuracy: 0.9611\n",
      "Epoch 110: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1183 - accuracy: 0.9611 - val_loss: 0.0892 - val_accuracy: 0.9845\n",
      "Epoch 111/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1042 - accuracy: 0.9664\n",
      "Epoch 111: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1048 - accuracy: 0.9658 - val_loss: 0.0943 - val_accuracy: 0.9836\n",
      "Epoch 112/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1025 - accuracy: 0.9664\n",
      "Epoch 112: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1024 - accuracy: 0.9662 - val_loss: 0.0920 - val_accuracy: 0.9831\n",
      "Epoch 113/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.1060 - accuracy: 0.9650\n",
      "Epoch 113: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1061 - accuracy: 0.9648 - val_loss: 0.0912 - val_accuracy: 0.9822\n",
      "Epoch 114/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0961 - accuracy: 0.9689\n",
      "Epoch 114: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0963 - accuracy: 0.9690 - val_loss: 0.0848 - val_accuracy: 0.9873\n",
      "Epoch 115/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0954 - accuracy: 0.9691\n",
      "Epoch 115: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0959 - accuracy: 0.9692 - val_loss: 0.0920 - val_accuracy: 0.9831\n",
      "Epoch 116/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1041 - accuracy: 0.9670\n",
      "Epoch 116: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1041 - accuracy: 0.9670 - val_loss: 0.0918 - val_accuracy: 0.9822\n",
      "Epoch 117/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.1041 - accuracy: 0.9650\n",
      "Epoch 117: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1045 - accuracy: 0.9647 - val_loss: 0.0861 - val_accuracy: 0.9855\n",
      "Epoch 118/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1090 - accuracy: 0.9662\n",
      "Epoch 118: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1084 - accuracy: 0.9662 - val_loss: 0.0836 - val_accuracy: 0.9873\n",
      "Epoch 119/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.1040 - accuracy: 0.9679\n",
      "Epoch 119: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1040 - accuracy: 0.9679 - val_loss: 0.0815 - val_accuracy: 0.9864\n",
      "Epoch 120/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.1005 - accuracy: 0.9680\n",
      "Epoch 120: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.1006 - accuracy: 0.9678 - val_loss: 0.0828 - val_accuracy: 0.9855\n",
      "Epoch 121/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0974 - accuracy: 0.9681\n",
      "Epoch 121: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0979 - accuracy: 0.9679 - val_loss: 0.0812 - val_accuracy: 0.9859\n",
      "Epoch 122/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0958 - accuracy: 0.9689\n",
      "Epoch 122: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0962 - accuracy: 0.9686 - val_loss: 0.0857 - val_accuracy: 0.9855\n",
      "Epoch 123/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0936 - accuracy: 0.9669\n",
      "Epoch 123: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0951 - accuracy: 0.9665 - val_loss: 0.0888 - val_accuracy: 0.9864\n",
      "Epoch 124/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.1017 - accuracy: 0.9666\n",
      "Epoch 124: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1010 - accuracy: 0.9667 - val_loss: 0.0843 - val_accuracy: 0.9845\n",
      "Epoch 125/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0906 - accuracy: 0.9689\n",
      "Epoch 125: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0905 - accuracy: 0.9687 - val_loss: 0.0781 - val_accuracy: 0.9887\n",
      "Epoch 126/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0993 - accuracy: 0.9686\n",
      "Epoch 126: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.0983 - accuracy: 0.9689 - val_loss: 0.0800 - val_accuracy: 0.9878\n",
      "Epoch 127/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0994 - accuracy: 0.9692\n",
      "Epoch 127: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.1003 - accuracy: 0.9689 - val_loss: 0.0779 - val_accuracy: 0.9859\n",
      "Epoch 128/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0845 - accuracy: 0.9736\n",
      "Epoch 128: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0853 - accuracy: 0.9736 - val_loss: 0.0836 - val_accuracy: 0.9845\n",
      "Epoch 129/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0864 - accuracy: 0.9718\n",
      "Epoch 129: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0865 - accuracy: 0.9715 - val_loss: 0.0843 - val_accuracy: 0.9859\n",
      "Epoch 130/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0922 - accuracy: 0.9693\n",
      "Epoch 130: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0906 - accuracy: 0.9701 - val_loss: 0.0821 - val_accuracy: 0.9873\n",
      "Epoch 131/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.0853 - accuracy: 0.9709\n",
      "Epoch 131: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0853 - accuracy: 0.9709 - val_loss: 0.0781 - val_accuracy: 0.9873\n",
      "Epoch 132/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0863 - accuracy: 0.9731\n",
      "Epoch 132: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0863 - accuracy: 0.9730 - val_loss: 0.0801 - val_accuracy: 0.9902\n",
      "Epoch 133/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.0864 - accuracy: 0.9705\n",
      "Epoch 133: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.0872 - accuracy: 0.9703 - val_loss: 0.0807 - val_accuracy: 0.9883\n",
      "Epoch 134/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0909 - accuracy: 0.9722\n",
      "Epoch 134: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.0900 - accuracy: 0.9726 - val_loss: 0.0805 - val_accuracy: 0.9887\n",
      "Epoch 135/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0804 - accuracy: 0.9739\n",
      "Epoch 135: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.0805 - accuracy: 0.9734 - val_loss: 0.0860 - val_accuracy: 0.9850\n",
      "Epoch 136/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.0851 - accuracy: 0.9722\n",
      "Epoch 136: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0840 - accuracy: 0.9728 - val_loss: 0.0816 - val_accuracy: 0.9878\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 137/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0940 - accuracy: 0.9694\n",
      "Epoch 137: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0926 - accuracy: 0.9698 - val_loss: 0.0863 - val_accuracy: 0.9878\n",
      "Epoch 138/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0941 - accuracy: 0.9701\n",
      "Epoch 138: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.0930 - accuracy: 0.9703 - val_loss: 0.0866 - val_accuracy: 0.9878\n",
      "Epoch 139/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0799 - accuracy: 0.9743\n",
      "Epoch 139: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0805 - accuracy: 0.9739 - val_loss: 0.0777 - val_accuracy: 0.9897\n",
      "Epoch 140/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0771 - accuracy: 0.9756\n",
      "Epoch 140: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0774 - accuracy: 0.9756 - val_loss: 0.0821 - val_accuracy: 0.9906\n",
      "Epoch 141/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.0760 - accuracy: 0.9755\n",
      "Epoch 141: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0760 - accuracy: 0.9759 - val_loss: 0.0776 - val_accuracy: 0.9883\n",
      "Epoch 142/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0734 - accuracy: 0.9762\n",
      "Epoch 142: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.0739 - accuracy: 0.9759 - val_loss: 0.0790 - val_accuracy: 0.9892\n",
      "Epoch 143/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0790 - accuracy: 0.9754\n",
      "Epoch 143: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0797 - accuracy: 0.9750 - val_loss: 0.0794 - val_accuracy: 0.9883\n",
      "Epoch 144/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0872 - accuracy: 0.9715\n",
      "Epoch 144: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0886 - accuracy: 0.9715 - val_loss: 0.0921 - val_accuracy: 0.9864\n",
      "Epoch 145/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0841 - accuracy: 0.9717\n",
      "Epoch 145: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0844 - accuracy: 0.9717 - val_loss: 0.0776 - val_accuracy: 0.9892\n",
      "Epoch 146/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0765 - accuracy: 0.9771\n",
      "Epoch 146: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0778 - accuracy: 0.9769 - val_loss: 0.0755 - val_accuracy: 0.9878\n",
      "Epoch 147/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0828 - accuracy: 0.9714\n",
      "Epoch 147: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0836 - accuracy: 0.9712 - val_loss: 0.0751 - val_accuracy: 0.9892\n",
      "Epoch 148/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0753 - accuracy: 0.9759\n",
      "Epoch 148: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0773 - accuracy: 0.9755 - val_loss: 0.0758 - val_accuracy: 0.9902\n",
      "Epoch 149/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0660 - accuracy: 0.9774\n",
      "Epoch 149: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0675 - accuracy: 0.9765 - val_loss: 0.0710 - val_accuracy: 0.9892\n",
      "Epoch 150/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.0747 - accuracy: 0.9766\n",
      "Epoch 150: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0753 - accuracy: 0.9764 - val_loss: 0.0727 - val_accuracy: 0.9911\n",
      "Epoch 151/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0708 - accuracy: 0.9801\n",
      "Epoch 151: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0725 - accuracy: 0.9795 - val_loss: 0.0697 - val_accuracy: 0.9911\n",
      "Epoch 152/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0773 - accuracy: 0.9741\n",
      "Epoch 152: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0788 - accuracy: 0.9739 - val_loss: 0.0793 - val_accuracy: 0.9892\n",
      "Epoch 153/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0668 - accuracy: 0.9764\n",
      "Epoch 153: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0681 - accuracy: 0.9759 - val_loss: 0.0720 - val_accuracy: 0.9883\n",
      "Epoch 154/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0773 - accuracy: 0.9754\n",
      "Epoch 154: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0763 - accuracy: 0.9755 - val_loss: 0.0809 - val_accuracy: 0.9887\n",
      "Epoch 155/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0733 - accuracy: 0.9755\n",
      "Epoch 155: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0740 - accuracy: 0.9755 - val_loss: 0.0785 - val_accuracy: 0.9883\n",
      "Epoch 156/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.0705 - accuracy: 0.9759\n",
      "Epoch 156: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0705 - accuracy: 0.9759 - val_loss: 0.0728 - val_accuracy: 0.9897\n",
      "Epoch 157/3550\n",
      "50/50 [==============================] - ETA: 0s - loss: 0.0731 - accuracy: 0.9767\n",
      "Epoch 157: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0731 - accuracy: 0.9767 - val_loss: 0.0724 - val_accuracy: 0.9916\n",
      "Epoch 158/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0673 - accuracy: 0.9789\n",
      "Epoch 158: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0699 - accuracy: 0.9780 - val_loss: 0.0796 - val_accuracy: 0.9902\n",
      "Epoch 159/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0685 - accuracy: 0.9769\n",
      "Epoch 159: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0681 - accuracy: 0.9772 - val_loss: 0.0766 - val_accuracy: 0.9906\n",
      "Epoch 160/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.0597 - accuracy: 0.9800\n",
      "Epoch 160: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0583 - accuracy: 0.9808 - val_loss: 0.0732 - val_accuracy: 0.9906\n",
      "Epoch 161/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0772 - accuracy: 0.9756\n",
      "Epoch 161: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0785 - accuracy: 0.9748 - val_loss: 0.0725 - val_accuracy: 0.9902\n",
      "Epoch 162/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.0708 - accuracy: 0.9776\n",
      "Epoch 162: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0716 - accuracy: 0.9767 - val_loss: 0.0702 - val_accuracy: 0.9916\n",
      "Epoch 163/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0699 - accuracy: 0.9759\n",
      "Epoch 163: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0703 - accuracy: 0.9761 - val_loss: 0.0744 - val_accuracy: 0.9897\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 164/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0625 - accuracy: 0.9794\n",
      "Epoch 164: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0608 - accuracy: 0.9795 - val_loss: 0.0807 - val_accuracy: 0.9897\n",
      "Epoch 165/3550\n",
      "49/50 [============================>.] - ETA: 0s - loss: 0.0660 - accuracy: 0.9780\n",
      "Epoch 165: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0659 - accuracy: 0.9781 - val_loss: 0.0805 - val_accuracy: 0.9883\n",
      "Epoch 166/3550\n",
      "48/50 [===========================>..] - ETA: 0s - loss: 0.0705 - accuracy: 0.9762\n",
      "Epoch 166: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0697 - accuracy: 0.9765 - val_loss: 0.0747 - val_accuracy: 0.9897\n",
      "Epoch 167/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0743 - accuracy: 0.9754\n",
      "Epoch 167: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0718 - accuracy: 0.9761 - val_loss: 0.0744 - val_accuracy: 0.9892\n",
      "Epoch 168/3550\n",
      "46/50 [==========================>...] - ETA: 0s - loss: 0.0782 - accuracy: 0.9747\n",
      "Epoch 168: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0768 - accuracy: 0.9750 - val_loss: 0.0852 - val_accuracy: 0.9883\n",
      "Epoch 169/3550\n",
      "47/50 [===========================>..] - ETA: 0s - loss: 0.0665 - accuracy: 0.9804\n",
      "Epoch 169: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 15ms/step - loss: 0.0670 - accuracy: 0.9801 - val_loss: 0.0760 - val_accuracy: 0.9892\n",
      "Epoch 170/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.0551 - accuracy: 0.9849\n",
      "Epoch 170: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0576 - accuracy: 0.9831 - val_loss: 0.0732 - val_accuracy: 0.9902\n",
      "Epoch 171/3550\n",
      "45/50 [==========================>...] - ETA: 0s - loss: 0.0686 - accuracy: 0.9767\n",
      "Epoch 171: saving model to model/keypoint_classifier\\keypoint_classifier.hdf5\n",
      "50/50 [==============================] - 1s 14ms/step - loss: 0.0689 - accuracy: 0.9772 - val_loss: 0.0770 - val_accuracy: 0.9902\n",
      "Epoch 171: early stopping\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.History at 0x287cb666be0>"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(\n",
    "    X_train,\n",
    "    y_train,\n",
    "    epochs=50*NUM_CLASSES,\n",
    "    batch_size=128,\n",
    "    validation_data=(X_test, y_test),\n",
    "    callbacks=[cp_callback, es_callback]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "pxvb2Y299hE3",
    "outputId": "59eb3185-2e37-4b9e-bc9d-ab1b8ac29b7f"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "17/17 [==============================] - 0s 5ms/step - loss: 0.0770 - accuracy: 0.9902\n"
     ]
    }
   ],
   "source": [
    "# Model evaluation\n",
    "val_loss, val_acc = model.evaluate(X_test, y_test, batch_size=128)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "id": "RBkmDeUW9hE4"
   },
   "outputs": [],
   "source": [
    "# Loading the saved model\n",
    "model = tf.keras.models.load_model(model_save_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "tFz9Tb0I9hE4",
    "outputId": "1c3b3528-54ae-4ee2-ab04-77429211cbef"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 [==============================] - 0s 173ms/step\n",
      "[4.97138734e-13 8.19166834e-16 9.63972738e-12 1.81294512e-04\n",
      " 5.25819877e-09 2.44161811e-08 1.89900509e-11 8.69887773e-11\n",
      " 7.12164272e-10 2.14848782e-07 5.90851881e-11 3.20831305e-07\n",
      " 4.09778439e-13 9.80282744e-10 2.96740232e-09 4.32316145e-18\n",
      " 7.64147990e-06 1.27531950e-13 1.92607800e-12 1.55236890e-08\n",
      " 9.99694109e-01 7.44881973e-23 8.64655938e-29 1.02457244e-16\n",
      " 4.65681604e-10 1.09707656e-07 3.43592182e-10 1.20187346e-10\n",
      " 4.92973067e-12 2.01103561e-11 1.69600172e-23 8.23120188e-23\n",
      " 1.08141318e-16 2.29283318e-21 1.21358635e-15 7.93935772e-14\n",
      " 1.05230662e-04 1.36382333e-12 1.77896960e-14 6.63542066e-09\n",
      " 3.72422785e-17 7.96509380e-15 6.51027054e-09 5.29633178e-11\n",
      " 8.34433390e-12 8.16829677e-12 1.09654143e-25 4.91255199e-20\n",
      " 6.73132562e-22 5.11984965e-14 2.12976507e-14 4.48700700e-22\n",
      " 7.37199635e-13 2.30111212e-07 5.89436474e-24 5.16257792e-09\n",
      " 1.32155838e-15 1.52377382e-14 8.26029918e-15 5.94566516e-19\n",
      " 5.14749077e-18 2.48096205e-12 1.49191095e-08 1.04018309e-05\n",
      " 2.41757603e-16 3.81504661e-10 1.04312851e-22 4.70201087e-07\n",
      " 5.06111104e-13 3.26173290e-21 4.99828662e-31]\n",
      "20\n"
     ]
    }
   ],
   "source": [
    "# Inference test\n",
    "predict_result = model.predict(np.array([X_test[0]]))\n",
    "print(np.squeeze(predict_result))\n",
    "print(np.argmax(np.squeeze(predict_result)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "S3U4yNWx9hE4"
   },
   "source": [
    "# Confusion matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 582
    },
    "id": "AP1V6SCk9hE5",
    "outputId": "08e41a80-7a4a-4619-8125-ecc371368d19"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "67/67 [==============================] - 0s 3ms/step\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1000x1000 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Classification Report\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       1.00      1.00      1.00        28\n",
      "           1       1.00      1.00      1.00        37\n",
      "           2       0.97      0.94      0.95        64\n",
      "           3       0.98      1.00      0.99        46\n",
      "           4       1.00      0.98      0.99        47\n",
      "           5       1.00      0.97      0.99        40\n",
      "           6       0.97      1.00      0.99        35\n",
      "           7       1.00      1.00      1.00        13\n",
      "           8       0.97      1.00      0.99        33\n",
      "           9       0.97      1.00      0.98        32\n",
      "          10       0.96      1.00      0.98        23\n",
      "          11       1.00      1.00      1.00        27\n",
      "          12       0.92      1.00      0.96        35\n",
      "          13       1.00      1.00      1.00        33\n",
      "          14       1.00      0.96      0.98        27\n",
      "          15       0.91      1.00      0.95        40\n",
      "          16       1.00      1.00      1.00        35\n",
      "          17       1.00      1.00      1.00        26\n",
      "          18       1.00      1.00      1.00        34\n",
      "          19       0.96      1.00      0.98        22\n",
      "          20       1.00      0.97      0.98        31\n",
      "          21       1.00      0.97      0.98        30\n",
      "          22       1.00      1.00      1.00        27\n",
      "          23       1.00      1.00      1.00        29\n",
      "          24       1.00      1.00      1.00        33\n",
      "          25       1.00      1.00      1.00        31\n",
      "          26       1.00      1.00      1.00        22\n",
      "          27       1.00      1.00      1.00        30\n",
      "          28       1.00      1.00      1.00        15\n",
      "          29       1.00      1.00      1.00        25\n",
      "          30       1.00      1.00      1.00        17\n",
      "          31       1.00      1.00      1.00        28\n",
      "          32       1.00      1.00      1.00        30\n",
      "          33       1.00      0.95      0.97        39\n",
      "          34       1.00      1.00      1.00        27\n",
      "          35       1.00      1.00      1.00        26\n",
      "          36       1.00      1.00      1.00        37\n",
      "          37       1.00      0.93      0.97        30\n",
      "          38       1.00      1.00      1.00        40\n",
      "          39       1.00      1.00      1.00        30\n",
      "          40       1.00      0.94      0.97        32\n",
      "          41       1.00      1.00      1.00        34\n",
      "          42       1.00      1.00      1.00        27\n",
      "          43       1.00      0.96      0.98        23\n",
      "          44       1.00      1.00      1.00        35\n",
      "          45       1.00      1.00      1.00        24\n",
      "          46       1.00      1.00      1.00        24\n",
      "          47       1.00      0.96      0.98        28\n",
      "          48       1.00      1.00      1.00        23\n",
      "          49       1.00      1.00      1.00        32\n",
      "          50       1.00      1.00      1.00        32\n",
      "          51       1.00      0.96      0.98        24\n",
      "          52       1.00      0.94      0.97        32\n",
      "          53       0.95      1.00      0.97        37\n",
      "          54       1.00      1.00      1.00        32\n",
      "          55       1.00      1.00      1.00        21\n",
      "          56       1.00      1.00      1.00        20\n",
      "          57       1.00      1.00      1.00        20\n",
      "          58       1.00      1.00      1.00        21\n",
      "          59       1.00      1.00      1.00        31\n",
      "          60       1.00      1.00      1.00        27\n",
      "          62       1.00      1.00      1.00        36\n",
      "          63       1.00      1.00      1.00        42\n",
      "          64       1.00      1.00      1.00        37\n",
      "          65       0.89      1.00      0.94        25\n",
      "          66       1.00      1.00      1.00        22\n",
      "          67       1.00      0.97      0.99        35\n",
      "          68       1.00      1.00      1.00        25\n",
      "          69       0.97      1.00      0.99        38\n",
      "          70       1.00      1.00      1.00        39\n",
      "\n",
      "    accuracy                           0.99      2132\n",
      "   macro avg       0.99      0.99      0.99      2132\n",
      "weighted avg       0.99      0.99      0.99      2132\n",
      "\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.metrics import confusion_matrix, classification_report\n",
    "\n",
    "def print_confusion_matrix(y_true, y_pred, report=True):\n",
    "    labels = sorted(list(set(y_true)))\n",
    "    cmx_data = confusion_matrix(y_true, y_pred, labels=labels)\n",
    "    \n",
    "    df_cmx = pd.DataFrame(cmx_data, index=labels, columns=labels)\n",
    " \n",
    "    fig, ax = plt.subplots(figsize=(10, 10))\n",
    "    sns.heatmap(df_cmx, annot=True, fmt='g' ,square=False)\n",
    "    ax.set_ylim(len(set(y_true)), 0)\n",
    "    plt.show()\n",
    "    \n",
    "    if report:\n",
    "        print('Classification Report')\n",
    "        print(classification_report(y_test, y_pred))\n",
    "\n",
    "Y_pred = model.predict(X_test)\n",
    "y_pred = np.argmax(Y_pred, axis=1)\n",
    "\n",
    "print_confusion_matrix(y_test, y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "FNP6aqzc9hE5"
   },
   "source": [
    "# Convert to model for Tensorflow-Lite"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "id": "ODjnYyld9hE6"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Sanjay\\AppData\\Local\\Programs\\Python\\Python39\\lib\\site-packages\\keras\\src\\engine\\training.py:3000: UserWarning: You are saving your model as an HDF5 file via `model.save()`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')`.\n",
      "  saving_api.save_model(\n"
     ]
    }
   ],
   "source": [
    "# Save as a model dedicated to inference\n",
    "model.save(model_save_path, include_optimizer=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "zRfuK8Y59hE6",
    "outputId": "a4ca585c-b5d5-4244-8291-8674063209bb"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: C:\\Users\\Sanjay\\AppData\\Local\\Temp\\tmp7syxr2ks\\assets\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: C:\\Users\\Sanjay\\AppData\\Local\\Temp\\tmp7syxr2ks\\assets\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "32144"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Transform model (quantization)\n",
    "\n",
    "converter = tf.lite.TFLiteConverter.from_keras_model(model)\n",
    "converter.optimizations = [tf.lite.Optimize.DEFAULT]\n",
    "tflite_quantized_model = converter.convert()\n",
    "\n",
    "open(tflite_save_path, 'wb').write(tflite_quantized_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "CHBPBXdx9hE6"
   },
   "source": [
    "# Inference test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "id": "mGAzLocO9hE7"
   },
   "outputs": [],
   "source": [
    "interpreter = tf.lite.Interpreter(model_path=tflite_save_path)\n",
    "interpreter.allocate_tensors()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "id": "oQuDK8YS9hE7"
   },
   "outputs": [],
   "source": [
    "# Get I / O tensor\n",
    "input_details = interpreter.get_input_details()\n",
    "output_details = interpreter.get_output_details()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "id": "2_ixAf_l9hE7"
   },
   "outputs": [],
   "source": [
    "interpreter.set_tensor(input_details[0]['index'], np.array([X_test[0]]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "s4FoAnuc9hE7",
    "outputId": "91f18257-8d8b-4ef3-c558-e9b5f94fabbf",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: total: 0 ns\n",
      "Wall time: 1.24 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# Inference implementation\n",
    "interpreter.invoke()\n",
    "tflite_results = interpreter.get_tensor(output_details[0]['index'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "vONjp19J9hE8",
    "outputId": "77205e24-fd00-42c4-f7b6-e06e527c2cba"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[5.17410333e-13 8.16583966e-16 9.80516189e-12 1.34590198e-04\n",
      " 5.03514741e-09 2.23877912e-08 1.93918111e-11 9.81094511e-11\n",
      " 7.25517257e-10 2.95727318e-07 9.03419908e-11 3.24298981e-07\n",
      " 3.79911841e-13 9.40401756e-10 3.41758510e-09 4.88818288e-18\n",
      " 7.12341352e-06 1.08076383e-13 1.73180003e-12 1.52976938e-08\n",
      " 9.99732792e-01 9.92278472e-23 9.87574084e-29 1.01611125e-16\n",
      " 4.05213807e-10 1.34043916e-07 4.02663597e-10 1.15277309e-10\n",
      " 5.75913406e-12 2.22309966e-11 2.04520788e-23 1.06994972e-22\n",
      " 1.93327065e-16 3.24110811e-21 1.29602267e-15 8.54853501e-14\n",
      " 1.14801514e-04 1.36775519e-12 2.08028378e-14 6.47075815e-09\n",
      " 3.78290705e-17 1.14048184e-14 5.82194204e-09 6.12874543e-11\n",
      " 8.63137947e-12 8.48416996e-12 1.74680305e-25 5.17667140e-20\n",
      " 9.09320193e-22 5.28580035e-14 2.61322946e-14 4.83740210e-22\n",
      " 8.05613714e-13 2.86574334e-07 7.83936047e-24 4.55681270e-09\n",
      " 1.73717659e-15 1.47080122e-14 7.99889550e-15 1.03118279e-18\n",
      " 5.02857818e-18 2.39658293e-12 1.41217678e-08 9.03716864e-06\n",
      " 2.51824960e-16 5.07999975e-10 1.53825541e-22 5.13169994e-07\n",
      " 5.43212393e-13 3.23646289e-21 6.68398689e-31]\n",
      "20\n"
     ]
    }
   ],
   "source": [
    "print(np.squeeze(tflite_results))\n",
    "print(np.argmax(np.squeeze(tflite_results)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "keypoint_classification_EN.ipynb",
   "provenance": [],
   "toc_visible": true
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}