Keras-openface-convertion.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import Conv2D, ZeroPadding2D, Activation, Input, concatenate\n",
    "from keras.models import Model\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.layers.pooling import MaxPooling2D, AveragePooling2D\n",
    "from keras.layers.merge import Concatenate\n",
    "from keras.layers.core import Lambda, Flatten, Dense\n",
    "from keras.engine.topology import Layer\n",
    "from keras import backend as K\n",
    "import cv2\n",
    "import os\n",
    "import numpy as np\n",
    "from numpy import genfromtxt\n",
    "import pandas as pd\n",
    "import tensorflow as tf\n",
    "from utils import LRN2D\n",
    "import utils\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "np.set_printoptions(threshold=np.nan)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "myInput = Input(shape=(96, 96, 3))\n",
    "\n",
    "x = ZeroPadding2D(padding=(3, 3), input_shape=(96, 96, 3))(myInput)\n",
    "x = Conv2D(64, (7, 7), strides=(2, 2), name='conv1')(x)\n",
    "x = BatchNormalization(axis=3, epsilon=0.00001, name='bn1')(x)\n",
    "x = Activation('relu')(x)\n",
    "x = ZeroPadding2D(padding=(1, 1))(x)\n",
    "x = MaxPooling2D(pool_size=3, strides=2)(x)\n",
    "x = Lambda(LRN2D, name='lrn_1')(x)\n",
    "x = Conv2D(64, (1, 1), name='conv2')(x)\n",
    "x = BatchNormalization(axis=3, epsilon=0.00001, name='bn2')(x)\n",
    "x = Activation('relu')(x)\n",
    "x = ZeroPadding2D(padding=(1, 1))(x)\n",
    "x = Conv2D(192, (3, 3), name='conv3')(x)\n",
    "x = BatchNormalization(axis=3, epsilon=0.00001, name='bn3')(x)\n",
    "x = Activation('relu')(x)\n",
    "x = Lambda(LRN2D, name='lrn_2')(x)\n",
    "x = ZeroPadding2D(padding=(1, 1))(x)\n",
    "x = MaxPooling2D(pool_size=3, strides=2)(x)\n",
    "\n",
    "# Inception3a\n",
    "inception_3a_3x3 = Conv2D(96, (1, 1), name='inception_3a_3x3_conv1')(x)\n",
    "inception_3a_3x3 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3a_3x3_bn1')(inception_3a_3x3)\n",
    "inception_3a_3x3 = Activation('relu')(inception_3a_3x3)\n",
    "inception_3a_3x3 = ZeroPadding2D(padding=(1, 1))(inception_3a_3x3)\n",
    "inception_3a_3x3 = Conv2D(128, (3, 3), name='inception_3a_3x3_conv2')(inception_3a_3x3)\n",
    "inception_3a_3x3 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3a_3x3_bn2')(inception_3a_3x3)\n",
    "inception_3a_3x3 = Activation('relu')(inception_3a_3x3)\n",
    "\n",
    "inception_3a_5x5 = Conv2D(16, (1, 1), name='inception_3a_5x5_conv1')(x)\n",
    "inception_3a_5x5 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3a_5x5_bn1')(inception_3a_5x5)\n",
    "inception_3a_5x5 = Activation('relu')(inception_3a_5x5)\n",
    "inception_3a_5x5 = ZeroPadding2D(padding=(2, 2))(inception_3a_5x5)\n",
    "inception_3a_5x5 = Conv2D(32, (5, 5), name='inception_3a_5x5_conv2')(inception_3a_5x5)\n",
    "inception_3a_5x5 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3a_5x5_bn2')(inception_3a_5x5)\n",
    "inception_3a_5x5 = Activation('relu')(inception_3a_5x5)\n",
    "\n",
    "inception_3a_pool = MaxPooling2D(pool_size=3, strides=2)(x)\n",
    "inception_3a_pool = Conv2D(32, (1, 1), name='inception_3a_pool_conv')(inception_3a_pool)\n",
    "inception_3a_pool = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3a_pool_bn')(inception_3a_pool)\n",
    "inception_3a_pool = Activation('relu')(inception_3a_pool)\n",
    "inception_3a_pool = ZeroPadding2D(padding=((3, 4), (3, 4)))(inception_3a_pool)\n",
    "\n",
    "inception_3a_1x1 = Conv2D(64, (1, 1), name='inception_3a_1x1_conv')(x)\n",
    "inception_3a_1x1 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3a_1x1_bn')(inception_3a_1x1)\n",
    "inception_3a_1x1 = Activation('relu')(inception_3a_1x1)\n",
    "\n",
    "inception_3a = concatenate([inception_3a_3x3, inception_3a_5x5, inception_3a_pool, inception_3a_1x1], axis=3)\n",
    "\n",
    "# Inception3b\n",
    "inception_3b_3x3 = Conv2D(96, (1, 1), name='inception_3b_3x3_conv1')(inception_3a)\n",
    "inception_3b_3x3 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3b_3x3_bn1')(inception_3b_3x3)\n",
    "inception_3b_3x3 = Activation('relu')(inception_3b_3x3)\n",
    "inception_3b_3x3 = ZeroPadding2D(padding=(1, 1))(inception_3b_3x3)\n",
    "inception_3b_3x3 = Conv2D(128, (3, 3), name='inception_3b_3x3_conv2')(inception_3b_3x3)\n",
    "inception_3b_3x3 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3b_3x3_bn2')(inception_3b_3x3)\n",
    "inception_3b_3x3 = Activation('relu')(inception_3b_3x3)\n",
    "\n",
    "inception_3b_5x5 = Conv2D(32, (1, 1), name='inception_3b_5x5_conv1')(inception_3a)\n",
    "inception_3b_5x5 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3b_5x5_bn1')(inception_3b_5x5)\n",
    "inception_3b_5x5 = Activation('relu')(inception_3b_5x5)\n",
    "inception_3b_5x5 = ZeroPadding2D(padding=(2, 2))(inception_3b_5x5)\n",
    "inception_3b_5x5 = Conv2D(64, (5, 5), name='inception_3b_5x5_conv2')(inception_3b_5x5)\n",
    "inception_3b_5x5 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3b_5x5_bn2')(inception_3b_5x5)\n",
    "inception_3b_5x5 = Activation('relu')(inception_3b_5x5)\n",
    "\n",
    "inception_3b_pool = Lambda(lambda x: x**2, name='power2_3b')(inception_3a)\n",
    "inception_3b_pool = AveragePooling2D(pool_size=(3, 3), strides=(3, 3))(inception_3b_pool)\n",
    "inception_3b_pool = Lambda(lambda x: x*9, name='mult9_3b')(inception_3b_pool)\n",
    "inception_3b_pool = Lambda(lambda x: K.sqrt(x), name='sqrt_3b')(inception_3b_pool)\n",
    "inception_3b_pool = Conv2D(64, (1, 1), name='inception_3b_pool_conv')(inception_3b_pool)\n",
    "inception_3b_pool = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3b_pool_bn')(inception_3b_pool)\n",
    "inception_3b_pool = Activation('relu')(inception_3b_pool)\n",
    "inception_3b_pool = ZeroPadding2D(padding=(4, 4))(inception_3b_pool)\n",
    "\n",
    "inception_3b_1x1 = Conv2D(64, (1, 1), name='inception_3b_1x1_conv')(inception_3a)\n",
    "inception_3b_1x1 = BatchNormalization(axis=3, epsilon=0.00001, name='inception_3b_1x1_bn')(inception_3b_1x1)\n",
    "inception_3b_1x1 = Activation('relu')(inception_3b_1x1)\n",
    "\n",
    "inception_3b = concatenate([inception_3b_3x3, inception_3b_5x5, inception_3b_pool, inception_3b_1x1], axis=3)\n",
    "\n",
    "# Inception3c\n",
    "inception_3c_3x3 = utils.conv2d_bn(inception_3b,\n",
    "                                   layer='inception_3c_3x3',\n",
    "                                   cv1_out=128,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   cv2_out=256,\n",
    "                                   cv2_filter=(3, 3),\n",
    "                                   cv2_strides=(2, 2),\n",
    "                                   padding=(1, 1))\n",
    "\n",
    "inception_3c_5x5 = utils.conv2d_bn(inception_3b,\n",
    "                                   layer='inception_3c_5x5',\n",
    "                                   cv1_out=32,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   cv2_out=64,\n",
    "                                   cv2_filter=(5, 5),\n",
    "                                   cv2_strides=(2, 2),\n",
    "                                   padding=(2, 2))\n",
    "\n",
    "inception_3c_pool = MaxPooling2D(pool_size=3, strides=2)(inception_3b)\n",
    "inception_3c_pool = ZeroPadding2D(padding=((0, 1), (0, 1)))(inception_3c_pool)\n",
    "\n",
    "inception_3c = concatenate([inception_3c_3x3, inception_3c_5x5, inception_3c_pool], axis=3)\n",
    "\n",
    "#inception 4a\n",
    "inception_4a_3x3 = utils.conv2d_bn(inception_3c,\n",
    "                                   layer='inception_4a_3x3',\n",
    "                                   cv1_out=96,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   cv2_out=192,\n",
    "                                   cv2_filter=(3, 3),\n",
    "                                   cv2_strides=(1, 1),\n",
    "                                   padding=(1, 1))\n",
    "inception_4a_5x5 = utils.conv2d_bn(inception_3c,\n",
    "                                   layer='inception_4a_5x5',\n",
    "                                   cv1_out=32,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   cv2_out=64,\n",
    "                                   cv2_filter=(5, 5),\n",
    "                                   cv2_strides=(1, 1),\n",
    "                                   padding=(2, 2))\n",
    "\n",
    "inception_4a_pool = Lambda(lambda x: x**2, name='power2_4a')(inception_3c)\n",
    "inception_4a_pool = AveragePooling2D(pool_size=(3, 3), strides=(3, 3))(inception_4a_pool)\n",
    "inception_4a_pool = Lambda(lambda x: x*9, name='mult9_4a')(inception_4a_pool)\n",
    "inception_4a_pool = Lambda(lambda x: K.sqrt(x), name='sqrt_4a')(inception_4a_pool)\n",
    "inception_4a_pool = utils.conv2d_bn(inception_4a_pool,\n",
    "                                   layer='inception_4a_pool',\n",
    "                                   cv1_out=128,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   padding=(2, 2))\n",
    "inception_4a_1x1 = utils.conv2d_bn(inception_3c,\n",
    "                                   layer='inception_4a_1x1',\n",
    "                                   cv1_out=256,\n",
    "                                   cv1_filter=(1, 1))\n",
    "inception_4a = concatenate([inception_4a_3x3, inception_4a_5x5, inception_4a_pool, inception_4a_1x1], axis=3)\n",
    "\n",
    "#inception4e\n",
    "inception_4e_3x3 = utils.conv2d_bn(inception_4a,\n",
    "                                   layer='inception_4e_3x3',\n",
    "                                   cv1_out=160,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   cv2_out=256,\n",
    "                                   cv2_filter=(3, 3),\n",
    "                                   cv2_strides=(2, 2),\n",
    "                                   padding=(1, 1))\n",
    "inception_4e_5x5 = utils.conv2d_bn(inception_4a,\n",
    "                                   layer='inception_4e_5x5',\n",
    "                                   cv1_out=64,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   cv2_out=128,\n",
    "                                   cv2_filter=(5, 5),\n",
    "                                   cv2_strides=(2, 2),\n",
    "                                   padding=(2, 2))\n",
    "inception_4e_pool = MaxPooling2D(pool_size=3, strides=2)(inception_4a)\n",
    "inception_4e_pool = ZeroPadding2D(padding=((0, 1), (0, 1)))(inception_4e_pool)\n",
    "\n",
    "inception_4e = concatenate([inception_4e_3x3, inception_4e_5x5, inception_4e_pool], axis=3)\n",
    "\n",
    "#inception5a\n",
    "inception_5a_3x3 = utils.conv2d_bn(inception_4e,\n",
    "                                   layer='inception_5a_3x3',\n",
    "                                   cv1_out=96,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   cv2_out=384,\n",
    "                                   cv2_filter=(3, 3),\n",
    "                                   cv2_strides=(1, 1),\n",
    "                                   padding=(1, 1))\n",
    "\n",
    "inception_5a_pool = Lambda(lambda x: x**2, name='power2_5a')(inception_4e)\n",
    "inception_5a_pool = AveragePooling2D(pool_size=(3, 3), strides=(3, 3))(inception_5a_pool)\n",
    "inception_5a_pool = Lambda(lambda x: x*9, name='mult9_5a')(inception_5a_pool)\n",
    "inception_5a_pool = Lambda(lambda x: K.sqrt(x), name='sqrt_5a')(inception_5a_pool)\n",
    "inception_5a_pool = utils.conv2d_bn(inception_5a_pool,\n",
    "                                   layer='inception_5a_pool',\n",
    "                                   cv1_out=96,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   padding=(1, 1))\n",
    "inception_5a_1x1 = utils.conv2d_bn(inception_4e,\n",
    "                                   layer='inception_5a_1x1',\n",
    "                                   cv1_out=256,\n",
    "                                   cv1_filter=(1, 1))\n",
    "\n",
    "inception_5a = concatenate([inception_5a_3x3, inception_5a_pool, inception_5a_1x1], axis=3)\n",
    "\n",
    "#inception_5b\n",
    "inception_5b_3x3 = utils.conv2d_bn(inception_5a,\n",
    "                                   layer='inception_5b_3x3',\n",
    "                                   cv1_out=96,\n",
    "                                   cv1_filter=(1, 1),\n",
    "                                   cv2_out=384,\n",
    "                                   cv2_filter=(3, 3),\n",
    "                                   cv2_strides=(1, 1),\n",
    "                                   padding=(1, 1))\n",
    "inception_5b_pool = MaxPooling2D(pool_size=3, strides=2)(inception_5a)\n",
    "inception_5b_pool = utils.conv2d_bn(inception_5b_pool,\n",
    "                                   layer='inception_5b_pool',\n",
    "                                   cv1_out=96,\n",
    "                                   cv1_filter=(1, 1))\n",
    "inception_5b_pool = ZeroPadding2D(padding=(1, 1))(inception_5b_pool)\n",
    "\n",
    "inception_5b_1x1 = utils.conv2d_bn(inception_5a,\n",
    "                                   layer='inception_5b_1x1',\n",
    "                                   cv1_out=256,\n",
    "                                   cv1_filter=(1, 1))\n",
    "inception_5b = concatenate([inception_5b_3x3, inception_5b_pool, inception_5b_1x1], axis=3)\n",
    "\n",
    "av_pool = AveragePooling2D(pool_size=(3, 3), strides=(1, 1))(inception_5b)\n",
    "reshape_layer = Flatten()(av_pool)\n",
    "dense_layer = Dense(128, name='dense_layer')(reshape_layer)\n",
    "norm_layer = Lambda(lambda  x: K.l2_normalize(x, axis=1), name='norm_layer')(dense_layer)\n",
    "\n",
    "\n",
    "# Final Model\n",
    "model = Model(inputs=[myInput], outputs=norm_layer)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Contruct the model structure"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load weights, Set weights and Save the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# Load weights from csv files (which was exported from Openface torch model)\n",
    "weights = utils.weights\n",
    "weights_dict = utils.load_weights()\n",
    "\n",
    "# Set layer weights of the model\n",
    "for name in weights:\n",
    "  if model.get_layer(name) != None:\n",
    "    model.get_layer(name).set_weights(weights_dict[name])\n",
    "  elif model.get_layer(name) != None:\n",
    "    model.get_layer(name).set_weights(weights_dict[name])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "model.save('./model/nn4.small2.lrn.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "!open ."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": false
   },
   "source": [
    "## Generate the embeddings "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/victor_sy_wang/anaconda/lib/python2.7/site-packages/keras/models.py:245: UserWarning: No training configuration found in save file: the model was *not* compiled. Compile it manually.\n",
      "  warnings.warn('No training configuration found in save file: '\n"
     ]
    },
    {
     "ename": "NameError",
     "evalue": "name 'cv2' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-2-d93c051831bc>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      7\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      8\u001b[0m \u001b[0;31m# Read a sample image as input to test the model\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 9\u001b[0;31m \u001b[0mimg\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcv2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mimread\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m'./data/dlib-affine-sz/Aaron_Eckhart/Aaron_Eckhart_0001.png'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     10\u001b[0m \u001b[0mimg\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mimg\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     11\u001b[0m \u001b[0mimg\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0maround\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtranspose\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mimg\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m/\u001b[0m\u001b[0;36m255.0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mdecimals\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m12\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'cv2' is not defined"
     ]
    }
   ],
   "source": [
    "from keras.models import load_model\n",
    "# model = load_model('./model/nn4.small2.lrn.h5')\n",
    "from keras.utils import CustomObjectScope\n",
    "import tensorflow as tf\n",
    "with CustomObjectScope({'tf': tf}):\n",
    "    model = load_model('./model/nn4.small2.lrn.h5')\n",
    "\n",
    "# Read a sample image as input to test the model\n",
    "img = cv2.imread('./data/dlib-affine-sz/Aaron_Eckhart/Aaron_Eckhart_0001.png', 1)\n",
    "img = img[...,::-1]\n",
    "img = np.around(np.transpose(img, (0,1,2))/255.0, decimals=12)\n",
    "print img.shape\n",
    "x_train = np.array([img])\n",
    "y = model.predict_on_batch(x_train)\n",
    "print y"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Validate on LFW"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Skipped 45 image pairs\n",
      "('batch: ', 0, ' time: ', 4.964755058288574)\n",
      "('batch: ', 1, ' time: ', 4.126990079879761)\n",
      "('batch: ', 2, ' time: ', 3.9385411739349365)\n",
      "('batch: ', 3, ' time: ', 3.8577771186828613)\n",
      "('batch: ', 4, ' time: ', 3.669827938079834)\n",
      "('batch: ', 5, ' time: ', 4.744009971618652)\n",
      "('batch: ', 6, ' time: ', 6.763217926025391)\n",
      "('batch: ', 7, ' time: ', 6.852792024612427)\n",
      "('batch: ', 8, ' time: ', 7.72891902923584)\n",
      "('batch: ', 9, ' time: ', 6.260838985443115)\n",
      "('batch: ', 10, ' time: ', 5.011318922042847)\n",
      "('batch: ', 11, ' time: ', 5.252028942108154)\n",
      "('batch: ', 12, ' time: ', 7.59849214553833)\n",
      "('batch: ', 13, ' time: ', 4.279605865478516)\n",
      "('batch: ', 14, ' time: ', 5.387643098831177)\n",
      "('batch: ', 15, ' time: ', 4.681977987289429)\n",
      "('batch: ', 16, ' time: ', 4.169487953186035)\n",
      "('batch: ', 17, ' time: ', 4.885305881500244)\n",
      "('batch: ', 18, ' time: ', 5.039474010467529)\n",
      "('batch: ', 19, ' time: ', 5.655552864074707)\n",
      "('batch: ', 20, ' time: ', 4.816269159317017)\n",
      "('batch: ', 21, ' time: ', 5.589965105056763)\n",
      "('batch: ', 22, ' time: ', 4.919957160949707)\n",
      "('batch: ', 23, ' time: ', 4.202291965484619)\n",
      "('batch: ', 24, ' time: ', 3.9112789630889893)\n",
      "('batch: ', 25, ' time: ', 4.191163778305054)\n",
      "('batch: ', 26, ' time: ', 5.246427059173584)\n",
      "('batch: ', 27, ' time: ', 3.8255679607391357)\n",
      "('batch: ', 28, ' time: ', 4.066335916519165)\n",
      "('batch: ', 29, ' time: ', 4.385252952575684)\n",
      "('batch: ', 30, ' time: ', 5.004914999008179)\n",
      "('batch: ', 31, ' time: ', 4.990630865097046)\n",
      "('batch: ', 32, ' time: ', 3.998673915863037)\n",
      "('batch: ', 33, ' time: ', 4.257556915283203)\n",
      "('batch: ', 34, ' time: ', 5.200344085693359)\n",
      "('batch: ', 35, ' time: ', 4.053833961486816)\n",
      "('batch: ', 36, ' time: ', 3.932110071182251)\n",
      "('batch: ', 37, ' time: ', 3.787412166595459)\n",
      "('batch: ', 38, ' time: ', 4.025928974151611)\n",
      "('batch: ', 39, ' time: ', 4.52328896522522)\n",
      "('batch: ', 40, ' time: ', 4.194962024688721)\n",
      "('batch: ', 41, ' time: ', 3.8770461082458496)\n",
      "('batch: ', 42, ' time: ', 3.722640037536621)\n",
      "('batch: ', 43, ' time: ', 4.133557081222534)\n",
      "('batch: ', 44, ' time: ', 3.732741117477417)\n",
      "('batch: ', 45, ' time: ', 5.193419933319092)\n",
      "('batch: ', 46, ' time: ', 6.77160906791687)\n",
      "('batch: ', 47, ' time: ', 4.887854099273682)\n",
      "('batch: ', 48, ' time: ', 7.133107900619507)\n",
      "('batch: ', 49, ' time: ', 5.716022968292236)\n",
      "('batch: ', 50, ' time: ', 7.009087085723877)\n",
      "('batch: ', 51, ' time: ', 5.228417873382568)\n",
      "('batch: ', 52, ' time: ', 6.6397340297698975)\n",
      "('batch: ', 53, ' time: ', 4.809391975402832)\n",
      "('batch: ', 54, ' time: ', 4.325794219970703)\n",
      "('batch: ', 55, ' time: ', 3.707091808319092)\n",
      "('batch: ', 56, ' time: ', 5.261268138885498)\n",
      "('batch: ', 57, ' time: ', 6.722397089004517)\n",
      "('batch: ', 58, ' time: ', 5.574383974075317)\n",
      "('batch: ', 59, ' time: ', 4.9825379848480225)\n",
      "('batch: ', 60, ' time: ', 4.662885904312134)\n",
      "('batch: ', 61, ' time: ', 4.378371000289917)\n",
      "('batch: ', 62, ' time: ', 4.067836046218872)\n",
      "('batch: ', 63, ' time: ', 4.711036920547485)\n",
      "('batch: ', 64, ' time: ', 4.616883993148804)\n",
      "('batch: ', 65, ' time: ', 5.247098922729492)\n",
      "('batch: ', 66, ' time: ', 4.415115118026733)\n",
      "('batch: ', 67, ' time: ', 4.392107009887695)\n",
      "('batch: ', 68, ' time: ', 5.248409032821655)\n",
      "('batch: ', 69, ' time: ', 5.991009950637817)\n",
      "('batch: ', 70, ' time: ', 4.762402057647705)\n",
      "('batch: ', 71, ' time: ', 5.578366994857788)\n",
      "('batch: ', 72, ' time: ', 5.143221855163574)\n",
      "('batch: ', 73, ' time: ', 5.022511005401611)\n",
      "('batch: ', 74, ' time: ', 4.705243110656738)\n",
      "('batch: ', 75, ' time: ', 4.030544996261597)\n",
      "('batch: ', 76, ' time: ', 3.876678943634033)\n",
      "('batch: ', 77, ' time: ', 3.7456068992614746)\n",
      "('batch: ', 78, ' time: ', 3.9784669876098633)\n",
      "('batch: ', 79, ' time: ', 3.6452879905700684)\n",
      "('batch: ', 80, ' time: ', 3.447438955307007)\n",
      "('batch: ', 81, ' time: ', 3.838253974914551)\n",
      "('batch: ', 82, ' time: ', 4.379256010055542)\n",
      "('batch: ', 83, ' time: ', 3.9987690448760986)\n",
      "('batch: ', 84, ' time: ', 3.551339864730835)\n",
      "('batch: ', 85, ' time: ', 3.5833840370178223)\n",
      "('batch: ', 86, ' time: ', 3.4598889350891113)\n",
      "('batch: ', 87, ' time: ', 3.6061689853668213)\n",
      "('batch: ', 88, ' time: ', 4.1108739376068115)\n",
      "('batch: ', 89, ' time: ', 3.9372000694274902)\n",
      "('batch: ', 90, ' time: ', 4.048888921737671)\n",
      "('batch: ', 91, ' time: ', 4.505039930343628)\n",
      "('batch: ', 92, ' time: ', 3.515172004699707)\n",
      "('batch: ', 93, ' time: ', 3.5279951095581055)\n",
      "('batch: ', 94, ' time: ', 3.599358081817627)\n",
      "('batch: ', 95, ' time: ', 4.611531019210815)\n",
      "('batch: ', 96, ' time: ', 3.821118116378784)\n",
      "('batch: ', 97, ' time: ', 3.67684006690979)\n",
      "('batch: ', 98, ' time: ', 4.127606153488159)\n",
      "('batch: ', 99, ' time: ', 6.083022117614746)\n",
      "('batch: ', 100, ' time: ', 3.9671638011932373)\n",
      "('batch: ', 101, ' time: ', 3.8468191623687744)\n",
      "('batch: ', 102, ' time: ', 4.213953018188477)\n",
      "('batch: ', 103, ' time: ', 4.644377946853638)\n",
      "('batch: ', 104, ' time: ', 5.670758008956909)\n",
      "('batch: ', 105, ' time: ', 4.525309085845947)\n",
      "('batch: ', 106, ' time: ', 4.213490009307861)\n",
      "('batch: ', 107, ' time: ', 5.545454025268555)\n",
      "('batch: ', 108, ' time: ', 5.678619146347046)\n",
      "('batch: ', 109, ' time: ', 4.480992078781128)\n",
      "('batch: ', 110, ' time: ', 4.185038089752197)\n",
      "('batch: ', 111, ' time: ', 3.5559921264648438)\n",
      "('batch: ', 112, ' time: ', 4.00173807144165)\n",
      "('batch: ', 113, ' time: ', 4.708425998687744)\n",
      "('batch: ', 114, ' time: ', 5.451466083526611)\n",
      "('batch: ', 115, ' time: ', 3.616629123687744)\n",
      "('batch: ', 116, ' time: ', 3.7177460193634033)\n",
      "('batch: ', 117, ' time: ', 4.1997230052948)\n",
      "('batch: ', 118, ' time: ', 6.3298180103302)\n",
      "('batch: ', 119, ' time: ', 0.5797300338745117)\n",
      "Accuracy: 0.938+-0.013\n",
      "Validation rate: 0.47469+-0.04227 @ FAR=0.00134\n",
      "Area Under Curve (AUC): 0.979\n",
      "Equal Error Rate (EER): 0.062\n"
     ]
    }
   ],
   "source": [
    "# LFW TEST\n",
    "import lfw\n",
    "import os\n",
    "import numpy as np\n",
    "import math\n",
    "import facenet\n",
    "import time\n",
    "import tensorflow as tf\n",
    "\n",
    "lfw_pairs='data/pairs.txt'\n",
    "lfw_dir='data/dlib-affine-sz'\n",
    "lfw_file_ext='png'\n",
    "lfw_nrof_folds=10\n",
    "image_size=96\n",
    "batch_size=100\n",
    "\n",
    "# Read the file containing the pairs used for testing\n",
    "pairs = lfw.read_pairs(os.path.expanduser(lfw_pairs))\n",
    "\n",
    "# Get the paths for the corresponding images\n",
    "paths, actual_issame = lfw.get_paths(os.path.expanduser(lfw_dir), pairs, lfw_file_ext)\n",
    "\n",
    "embedding_size=128\n",
    "nrof_images = len(paths)\n",
    "nrof_batches = int(math.ceil(1.0*nrof_images / batch_size))\n",
    "emb_array = np.zeros((nrof_images, embedding_size))\n",
    "\n",
    "# print paths\n",
    "\n",
    "for i in range(nrof_batches):\n",
    "  start_index = i*batch_size\n",
    "  end_index = min((i+1)*batch_size, nrof_images)\n",
    "  paths_batch = paths[start_index:end_index]\n",
    "  images = facenet.load_data(paths_batch, False, False, image_size)\n",
    "#   images = np.transpose(images, (0,3,1,2))\n",
    "  \n",
    "  t0 = time.time()\n",
    "  y = model.predict_on_batch(images)\n",
    "  emb_array[start_index:end_index,:] = y\n",
    "#   print('y', y)\n",
    "#   print('emb', emb_array[start_index:end_index,:])\n",
    "  t1 = time.time()\n",
    "  \n",
    "  print('batch: ', i, ' time: ', t1-t0)\n",
    "\n",
    "from sklearn import metrics\n",
    "from scipy.optimize import brentq\n",
    "from scipy import interpolate\n",
    "\n",
    "tpr, fpr, accuracy, val, val_std, far = lfw.evaluate(emb_array, \n",
    "                actual_issame, nrof_folds=lfw_nrof_folds)\n",
    "\n",
    "print('Accuracy: %1.3f+-%1.3f' % (np.mean(accuracy), np.std(accuracy)))\n",
    "print('Validation rate: %2.5f+-%2.5f @ FAR=%2.5f' % (val, val_std, far))\n",
    "auc = metrics.auc(fpr, tpr)\n",
    "print('Area Under Curve (AUC): %1.3f' % auc)\n",
    "\n",
    "eer = brentq(lambda x: 1. - x - interpolate.interp1d(fpr, tpr)(x), 0., 1.)\n",
    "print('Equal Error Rate (EER): %1.3f' % eer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}