Project3_Updated.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "path =  /Users/nitin/Documents/TUM/TUM/Sem_3/Tracking and Detection/Exercise/Python/Object_Updated\n",
      "['cat', 'ape', 'duck', 'benchvise', 'cam']\n",
      "(5055, 64, 64, 3) (5055, 4) (5055,)\n",
      "['cat', 'ape', 'duck', 'benchvise', 'cam']\n",
      "(1335, 64, 64, 3) (1335, 4) (1335,)\n",
      "['cat', 'ape', 'duck', 'benchvise', 'cam']\n",
      "(5890, 64, 64, 3) (5890, 4) (5890,)\n",
      "building train set\n",
      "(7410, 64, 64, 3) (7410, 4) (7410,)\n",
      "building test set\n",
      "(3535, 64, 64, 3) (3535, 4) (3535,)\n",
      "building triplet list\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:44: RuntimeWarning: invalid value encountered in arccos\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "number of triplets =  7410\n",
      "Db , train, test shapes \n",
      "\n",
      "(1335, 64, 64, 3)\n",
      "(7410, 64, 64, 3)\n",
      "(3535, 64, 64, 3)\n",
      "Db , train, test class annotations \n",
      "\n",
      "[0 0 0 ... 4 4 4]\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import util as util\n",
    "import cv2\n",
    "import random\n",
    "\n",
    "# Steps for data generation\n",
    "\n",
    "# load the image from the path\n",
    "# function to read corresponding pose from the file (pose are stored in quaterions)\n",
    "# create a class or struct for the samples\n",
    "\n",
    "# s_train =  real + fine\n",
    "# s_db = coarse folder\n",
    "# s_test = real\n",
    "\n",
    "# batch generator function -> triplet generation\n",
    "# anchor = randomly chosen from the training set\n",
    "# Puller = most similar to anchor from the db\n",
    "# pusher = same object different pose and different object pusher\n",
    "\n",
    "def visualizer(anchor,puller,s_pusher):\n",
    "    cv2.imshow(\"anchor \", anchor)\n",
    "    cv2.imshow(\"puller\", puller)\n",
    "    cv2.imshow(\"same class pusher\", s_pusher)\n",
    "    cv2.waitKey(1)\n",
    "\n",
    "def get_puller(anchor,anc_class,s_db):\n",
    "    thresh = 2\n",
    "    thetha_list = []\n",
    "    puller = -1\n",
    "\n",
    "    #select only sample from the class\n",
    "    indx = [i for i in range(s_db['pose'].shape[0]) if s_db['label'][i] ==  anc_class]\n",
    "\n",
    "    for idx in indx:\n",
    "        test_pusher = s_db['pose'][idx, :]\n",
    "        anchor = list(map(float, anchor))\n",
    "        test_pusher = list(map(float, test_pusher))\n",
    "\n",
    "        x = np.absolute(np.dot(test_pusher, anchor))\n",
    "        theth_var = 2 * np.arccos(x)\n",
    "        thetha_list.append(theth_var)\n",
    "        \n",
    "\n",
    "    puller = np.argmin(thetha_list)\n",
    "\n",
    "    #same class pusher  - code enforces\n",
    "    a = thetha_list\n",
    "    s_pusher_value = np.partition(a, thresh)[thresh]\n",
    "    s_pusher = list(thetha_list).index(s_pusher_value)\n",
    "    \n",
    "    #need to change it\n",
    "    return puller + indx[0], s_pusher + indx[0]\n",
    "\n",
    "\n",
    "def gen_triplet_list(s_train,s_db):\n",
    "    print('building triplet list')\n",
    "    triplet_list = []\n",
    "\n",
    "    for i in range(s_train['pose'].shape[0]):\n",
    "        test_anchor = s_train['pose'][i, :]\n",
    "        test_anchor_class = s_train['label'][i]\n",
    "        puller, s_pusher = get_puller(test_anchor, test_anchor_class, s_db)\n",
    "        triplet_list.append([i,puller,s_pusher])\n",
    "    print('number of triplets = ', len(triplet_list))\n",
    "    return triplet_list\n",
    "\n",
    "\n",
    "def  get_batch(s_train,s_db,triplet_list,batch_size):\n",
    "    batch = []\n",
    "    idx = random.sample(range(0, len(triplet_list)),batch_size)\n",
    "    for i in idx:\n",
    "        t = triplet_list[i]\n",
    "        batch.append(s_train['img'][t[0],:,:,:])\n",
    "        batch.append(s_db['img'][t[1],:,:,:])\n",
    "        batch.append(s_db['img'][t[2],:,:,:])\n",
    "\n",
    "    return np.asarray(batch)\n",
    "\n",
    "if '__main__' == __name__:\n",
    "\n",
    "    base_path = os.getcwd()\n",
    "    print('path = ', base_path)\n",
    "\n",
    "    fine_set = util.loadDataset(os.path.join(base_path, \"dataset/fine\"), 'poses.txt')\n",
    "    s_db = util.loadDataset(os.path.join(base_path, \"dataset/coarse\"), 'poses.txt')\n",
    "\n",
    "    S_db_img=s_db['img']\n",
    "    S_db_class=s_db['label']\n",
    "    S_db_quat=s_db['pose']\n",
    "    \n",
    "    \n",
    "    \n",
    "    real_set = util.loadDataset(os.path.join(base_path, \"dataset/real\"), 'poses.txt')\n",
    "    train_idx, test_idx = util.read_indicies(os.path.join(base_path, \"dataset/real\"), 'training_split.txt')\n",
    "    s_train = util.build_train_set(real_set,fine_set,train_idx)\n",
    "    S_train_img=s_train['img']\n",
    "    S_train_class=s_train['label']\n",
    "    S_train_quat=s_train['pose']\n",
    "    \n",
    "\n",
    "    s_test = util.build_test_set(real_set,test_idx) \n",
    "    S_test_img=s_test['img']\n",
    "    S_test_class=s_test['label']\n",
    "    S_test_quat=s_test['pose']\n",
    "    \n",
    "    \n",
    "    triplet_list = gen_triplet_list(s_train,s_db)\n",
    "\n",
    "    print(\"Db , train, test shapes \\n\")\n",
    "    print(S_db_img.shape)\n",
    "    print(S_train_img.shape)\n",
    "    print(S_test_img.shape)\n",
    "    \n",
    "    print(\"Db , train, test class annotations \\n\")\n",
    "    print(S_train_class)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([-0.19322862, -0.6801931 ,  0.6801931 , -0.19322862], dtype=float32)"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "S_db_quat[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import seaborn as sn\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from util import *\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Displaying a batch of anchor, puller, and pusher\n",
    "batch_size = 3\n",
    "batch_img, batch_class, batch_quat = batch_generator(S_train_img, S_train_class, S_train_quat,\n",
    "                                                     S_db_img, S_db_class, S_db_quat, batch_size)\n",
    "plot_images(batch_img)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "##Normalising the values between (0,1)\n",
    "\n",
    "S_db_img = np.array(S_db_img, dtype=\"float32\", copy = True)\n",
    "S_train_img = np.array(S_train_img, dtype=\"float32\", copy = True)\n",
    "S_test_img = np.array(S_test_img, dtype=\"float32\", copy = True)\n",
    "\n",
    "for i in range(0, len(S_db_img)):\n",
    "    S_db_img[i] = normalize_img(S_db_img[i])\n",
    "\n",
    "for i in range(0, len(S_train_img)):\n",
    "    S_train_img[i] = normalize_img(S_train_img[i])\n",
    "\n",
    "for i in range(0, len(S_test_img)):\n",
    "    S_test_img[i] = normalize_img(S_test_img[i])\n",
    "\n",
    "    \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"x:0\", shape=(?, 64, 64, 3), dtype=float32)\n",
      "Tensor(\"MaxPool:0\", shape=(?, 28, 28, 16), dtype=float32)\n",
      "Tensor(\"MaxPool_1:0\", shape=(?, 12, 12, 7), dtype=float32)\n",
      "Tensor(\"Relu_2:0\", shape=(?, 256), dtype=float32)\n",
      "Tensor(\"add_3:0\", shape=(?, 16), dtype=float32)\n"
     ]
    }
   ],
   "source": [
    "x = tf.placeholder(tf.float32, shape=[None, 64, 64, 3], name='x')\n",
    "\n",
    "#LAYERS\n",
    "layer_conv1 = new_conv_layer(input=x, num_input_channels=3,\n",
    "                   filter_size=8, num_filters=16, pooling=True, name=\"Conv1\")\n",
    "\n",
    "layer_conv2 = new_conv_layer(input=layer_conv1, num_input_channels=16,\n",
    "                   filter_size=5, num_filters=7, pooling=True, name=\"Conv2\")\n",
    "\n",
    "layer_flat, num_features = flatten_layer(layer_conv2, name=\"Flatten\")\n",
    "layer_fc1 = new_fc_layer(input=layer_flat, num_inputs=num_features, num_outputs=256, relu=True, name=\"FC1\")\n",
    "layer_fc2 = new_fc_layer(input=layer_fc1, num_inputs=256, num_outputs=16, relu=False, name=\"FC2\")\n",
    "\n",
    "#LOSS\n",
    "loss = total_loss(layer_fc2)\n",
    "\n",
    "#TYPE OF OPTIMIZER\n",
    "optimizer = tf.train.AdamOptimizer(learning_rate = 1e-4).minimize(loss)\n",
    "\n",
    "#SIZES\n",
    "print(x)\n",
    "print(layer_conv1)\n",
    "print(layer_conv2)\n",
    "print(layer_fc1)\n",
    "print(layer_fc2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "##Inserting the parameters into the model\n",
    "\n",
    "batch_size = 32\n",
    "epochs = 250\n",
    "iterations_per_epoch = max(len(S_train_img) // batch_size, 1)\n",
    "num_iterations = epochs * iterations_per_epoch\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loss value at iteration 0/57749 : 8.130266\n",
      "INFO:tensorflow:./checkpoint/model0.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model0.ckpt\n",
      "[ 0.14144272  1.10325318  4.3281471  28.71287129]\n",
      "\n",
      "Loss value at iteration 1000/57749 : 1.1443081\n",
      "INFO:tensorflow:./checkpoint/model1000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model1000.ckpt\n",
      "[ 3.73408769 17.93493635 47.35502122 93.91796322]\n",
      "\n",
      "Loss value at iteration 2000/57749 : 0.39949083\n",
      "INFO:tensorflow:./checkpoint/model2000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model2000.ckpt\n",
      "[ 6.13861386 27.86421499 60.70721358 96.03960396]\n",
      "\n",
      "Loss value at iteration 3000/57749 : 0.27424404\n",
      "INFO:tensorflow:./checkpoint/model3000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model3000.ckpt\n",
      "[ 9.87270156 38.84016973 74.14427157 98.81188119]\n",
      "\n",
      "Loss value at iteration 4000/57749 : 0.23851103\n",
      "INFO:tensorflow:./checkpoint/model4000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model4000.ckpt\n",
      "[ 9.73125884 38.47241867 74.59688826 99.00990099]\n",
      "\n",
      "Loss value at iteration 5000/57749 : 0.19554734\n",
      "INFO:tensorflow:./checkpoint/model5000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model5000.ckpt\n",
      "[11.06082037 42.54596888 78.01980198 99.34936351]\n",
      "\n",
      "Loss value at iteration 6000/57749 : 0.19847235\n",
      "INFO:tensorflow:./checkpoint/model6000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model6000.ckpt\n",
      "[12.70155587 46.42149929 83.11173975 99.51909477]\n",
      "\n",
      "Loss value at iteration 7000/57749 : 0.16553336\n",
      "INFO:tensorflow:./checkpoint/model7000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model7000.ckpt\n",
      "[14.6251768  50.32531825 87.43988685 99.91513437]\n",
      "\n",
      "Loss value at iteration 8000/57749 : 0.16498412\n",
      "INFO:tensorflow:./checkpoint/model8000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model8000.ckpt\n",
      "[16.0678925  52.70155587 86.76096181 99.83026874]\n",
      "\n",
      "Loss value at iteration 9000/57749 : 0.16202113\n",
      "INFO:tensorflow:./checkpoint/model9000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model9000.ckpt\n",
      "[15.30410184 52.89957567 88.14710042 99.94342291]\n",
      "\n",
      "Loss value at iteration 10000/57749 : 0.15470827\n",
      "INFO:tensorflow:./checkpoint/model10000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model10000.ckpt\n",
      "[11.06082037 45.31824611 84.10183876 99.51909477]\n",
      "\n",
      "Loss value at iteration 11000/57749 : 0.34502944\n",
      "INFO:tensorflow:./checkpoint/model11000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model11000.ckpt\n",
      "[15.41725601 53.97454031 88.59971711 99.83026874]\n",
      "\n",
      "Loss value at iteration 12000/57749 : 0.12410284\n",
      "INFO:tensorflow:./checkpoint/model12000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model12000.ckpt\n",
      "[15.24752475 52.36209335 87.83592645 99.71711457]\n",
      "\n",
      "Loss value at iteration 13000/57749 : 0.14909619\n",
      "INFO:tensorflow:./checkpoint/model13000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model13000.ckpt\n",
      "[12.70155587 47.4115983  84.80905233 99.66053748]\n",
      "\n",
      "Loss value at iteration 14000/57749 : 0.10688153\n",
      "INFO:tensorflow:./checkpoint/model14000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model14000.ckpt\n",
      "[ 16.74681754  56.94483734  89.22206506 100.        ]\n",
      "\n",
      "Loss value at iteration 15000/57749 : 0.11992465\n",
      "INFO:tensorflow:./checkpoint/model15000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model15000.ckpt\n",
      "[16.66195191 55.95473833 89.30693069 99.97171146]\n",
      "\n",
      "Loss value at iteration 16000/57749 : 0.11460137\n",
      "INFO:tensorflow:./checkpoint/model16000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model16000.ckpt\n",
      "[16.85997171 57.70862801 90.74964639 99.8019802 ]\n",
      "\n",
      "Loss value at iteration 17000/57749 : 0.10283168\n",
      "INFO:tensorflow:./checkpoint/model17000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model17000.ckpt\n",
      "[ 16.66195191  57.42574257  91.96605375 100.        ]\n",
      "\n",
      "Loss value at iteration 18000/57749 : 0.09512212\n",
      "INFO:tensorflow:./checkpoint/model18000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model18000.ckpt\n",
      "[16.23762376 57.68033946 92.30551627 99.94342291]\n",
      "\n",
      "Loss value at iteration 19000/57749 : 0.0883124\n",
      "INFO:tensorflow:./checkpoint/model19000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model19000.ckpt\n",
      "[17.02970297 59.77369165 92.36209335 99.94342291]\n",
      "\n",
      "Loss value at iteration 20000/57749 : 0.08767728\n",
      "INFO:tensorflow:./checkpoint/model20000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model20000.ckpt\n",
      "[15.24752475 54.39886846 90.35360679 99.88684583]\n",
      "\n",
      "Loss value at iteration 21000/57749 : 0.063710675\n",
      "INFO:tensorflow:./checkpoint/model21000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model21000.ckpt\n",
      "[18.61386139 58.27439887 91.40028289 99.94342291]\n",
      "\n",
      "Loss value at iteration 22000/57749 : 0.09147061\n",
      "INFO:tensorflow:./checkpoint/model22000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model22000.ckpt\n",
      "[17.28429986 57.5106082  91.96605375 99.94342291]\n",
      "\n",
      "Loss value at iteration 23000/57749 : 0.06564817\n",
      "INFO:tensorflow:./checkpoint/model23000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model23000.ckpt\n",
      "[18.7553041  60.67892504 93.06930693 99.97171146]\n",
      "\n",
      "Loss value at iteration 24000/57749 : 0.08724272\n",
      "INFO:tensorflow:./checkpoint/model24000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model24000.ckpt\n",
      "[17.53889675 56.97312588 91.85289958 99.94342291]\n",
      "\n",
      "Loss value at iteration 25000/57749 : 0.073024176\n",
      "INFO:tensorflow:./checkpoint/model25000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model25000.ckpt\n",
      "[16.01131542 56.74681754 91.17397454 99.91513437]\n",
      "\n",
      "Loss value at iteration 26000/57749 : 0.07317094\n",
      "INFO:tensorflow:./checkpoint/model26000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model26000.ckpt\n",
      "[18.98161245 61.98019802 93.63507779 99.97171146]\n",
      "\n",
      "Loss value at iteration 27000/57749 : 0.06616758\n",
      "INFO:tensorflow:./checkpoint/model27000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model27000.ckpt\n",
      "[14.85148515 54.45544554 90.0990099  99.85855728]\n",
      "\n",
      "Loss value at iteration 28000/57749 : 0.06613603\n",
      "INFO:tensorflow:./checkpoint/model28000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model28000.ckpt\n",
      "[17.79349364 59.97171146 93.12588402 99.97171146]\n",
      "\n",
      "Loss value at iteration 29000/57749 : 0.07581375\n",
      "INFO:tensorflow:./checkpoint/model29000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model29000.ckpt\n",
      "[15.81329562 56.54879774 91.76803395 99.97171146]\n",
      "\n",
      "Loss value at iteration 30000/57749 : 0.05963672\n",
      "INFO:tensorflow:./checkpoint/model30000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model30000.ckpt\n",
      "[19.00990099 60.56577086 93.63507779 99.88684583]\n",
      "\n",
      "Loss value at iteration 31000/57749 : 0.054163404\n",
      "INFO:tensorflow:./checkpoint/model31000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model31000.ckpt\n",
      "[17.62376238 60.45261669 93.49363508 99.94342291]\n",
      "\n",
      "Loss value at iteration 32000/57749 : 0.28577846\n",
      "INFO:tensorflow:./checkpoint/model32000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model32000.ckpt\n",
      "[19.40594059 60.59405941 93.40876945 99.94342291]\n",
      "\n",
      "Loss value at iteration 33000/57749 : 0.053292725\n",
      "INFO:tensorflow:./checkpoint/model33000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model33000.ckpt\n",
      "[19.43422914 60.50919378 93.43705799 99.97171146]\n",
      "\n",
      "Loss value at iteration 34000/57749 : 0.058332145\n",
      "INFO:tensorflow:./checkpoint/model34000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model34000.ckpt\n",
      "[18.64214993 61.52758133 93.52192362 99.97171146]\n",
      "\n",
      "Loss value at iteration 35000/57749 : 0.042943917\n",
      "INFO:tensorflow:./checkpoint/model35000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model35000.ckpt\n",
      "[20.33946252 64.46958982 95.13437058 99.94342291]\n",
      "\n",
      "Loss value at iteration 36000/57749 : 0.039598145\n",
      "INFO:tensorflow:./checkpoint/model36000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model36000.ckpt\n",
      "[20.         63.84724187 94.39886846 99.94342291]\n",
      "\n",
      "Loss value at iteration 37000/57749 : 0.045738112\n",
      "INFO:tensorflow:./checkpoint/model37000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model37000.ckpt\n",
      "[21.35785007 63.62093352 93.38048091 99.94342291]\n",
      "\n",
      "Loss value at iteration 38000/57749 : 0.037493438\n",
      "INFO:tensorflow:./checkpoint/model38000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model38000.ckpt\n",
      "[18.5572843  60.79207921 93.60678925 99.91513437]\n",
      "\n",
      "Loss value at iteration 39000/57749 : 0.046317823\n",
      "INFO:tensorflow:./checkpoint/model39000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model39000.ckpt\n",
      "[20.79207921 62.6874116  94.11598303 99.94342291]\n",
      "\n",
      "Loss value at iteration 40000/57749 : 0.039422754\n",
      "INFO:tensorflow:./checkpoint/model40000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model40000.ckpt\n",
      "[21.69731259 64.29985856 94.54031117 99.94342291]\n",
      "\n",
      "Loss value at iteration 41000/57749 : 0.060229592\n",
      "INFO:tensorflow:./checkpoint/model41000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model41000.ckpt\n",
      "[15.98302687 55.75671853 92.24893918 99.94342291]\n",
      "\n",
      "Loss value at iteration 42000/57749 : 0.035938412\n",
      "INFO:tensorflow:./checkpoint/model42000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model42000.ckpt\n",
      "[20.67892504 63.30975955 94.39886846 99.94342291]\n",
      "\n",
      "Loss value at iteration 43000/57749 : 0.0314852\n",
      "INFO:tensorflow:./checkpoint/model43000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model43000.ckpt\n",
      "[19.83026874 60.93352192 93.38048091 99.94342291]\n",
      "\n",
      "Loss value at iteration 44000/57749 : 0.04255145\n",
      "INFO:tensorflow:./checkpoint/model44000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model44000.ckpt\n",
      "[21.78217822 64.7241867  94.20084866 99.94342291]\n",
      "\n",
      "Loss value at iteration 45000/57749 : 0.03338835\n",
      "INFO:tensorflow:./checkpoint/model45000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model45000.ckpt\n",
      "[22.06506365 63.79066478 93.46534653 99.97171146]\n",
      "\n",
      "Loss value at iteration 46000/57749 : 0.033182822\n",
      "INFO:tensorflow:./checkpoint/model46000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model46000.ckpt\n",
      "[20.93352192 63.11173975 93.71994342 99.94342291]\n",
      "\n",
      "Loss value at iteration 47000/57749 : 0.044539962\n",
      "INFO:tensorflow:./checkpoint/model47000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model47000.ckpt\n",
      "[20.67892504 62.91371994 94.2291372  99.91513437]\n",
      "\n",
      "Loss value at iteration 48000/57749 : 0.046822235\n",
      "INFO:tensorflow:./checkpoint/model48000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model48000.ckpt\n",
      "[ 20.          60.99009901  93.74823197 100.        ]\n",
      "\n",
      "Loss value at iteration 49000/57749 : 0.14138122\n",
      "INFO:tensorflow:./checkpoint/model49000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model49000.ckpt\n",
      "[19.06647808 60.79207921 92.02263083 99.97171146]\n",
      "\n",
      "Loss value at iteration 50000/57749 : 0.02340959\n",
      "INFO:tensorflow:./checkpoint/model50000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model50000.ckpt\n",
      "[24.55445545 67.55304102 94.76661952 99.97171146]\n",
      "\n",
      "Loss value at iteration 51000/57749 : 0.03454431\n",
      "INFO:tensorflow:./checkpoint/model51000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model51000.ckpt\n",
      "[18.18953324 59.34936351 92.72984441 99.97171146]\n",
      "\n",
      "Loss value at iteration 52000/57749 : 0.046355825\n",
      "INFO:tensorflow:./checkpoint/model52000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model52000.ckpt\n",
      "[19.51909477 59.6039604  92.41867044 99.97171146]\n",
      "\n",
      "Loss value at iteration 53000/57749 : 0.04237319\n",
      "INFO:tensorflow:./checkpoint/model53000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model53000.ckpt\n",
      "[18.7553041  58.64214993 92.56011315 99.97171146]\n",
      "\n",
      "Loss value at iteration 54000/57749 : 0.020649377\n",
      "INFO:tensorflow:./checkpoint/model54000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model54000.ckpt\n",
      "[22.65912306 65.09193777 94.71004243 99.97171146]\n",
      "\n",
      "Loss value at iteration 55000/57749 : 0.037520118\n",
      "INFO:tensorflow:./checkpoint/model55000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model55000.ckpt\n",
      "[19.66053748 60.3960396  94.17256011 99.97171146]\n",
      "\n",
      "Loss value at iteration 56000/57749 : 0.026194388\n",
      "INFO:tensorflow:./checkpoint/model56000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model56000.ckpt\n",
      "[ 22.97029703  64.29985856  93.97454031 100.        ]\n",
      "\n",
      "Loss value at iteration 57000/57749 : 0.025051683\n",
      "INFO:tensorflow:./checkpoint/model57000.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n",
      "INFO:tensorflow:Restoring parameters from ./checkpoint/model57000.ckpt\n",
      "[ 22.99858557  65.71428571  94.25742574 100.        ]\n",
      "\n",
      "INFO:tensorflow:./checkpoint/model.ckpt is not in all_model_checkpoint_paths. Manually adding it.\n"
     ]
    }
   ],
   "source": [
    "batch_size = 32\n",
    "epochs = 250\n",
    "iterations_per_epoch = max(len(S_train_img) // batch_size, 1)\n",
    "num_iterations = epochs * iterations_per_epoch\n",
    "\n",
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver(max_to_keep=None)\n",
    "\n",
    "logs_path = './logs'\n",
    "tf.summary.scalar(\"loss\", loss)\n",
    "merged_summary = tf.summary.merge_all()\n",
    "\n",
    "with tf.Session() as sess:\n",
    "\n",
    "    sess.run(init)\n",
    "    summary_writer = tf.summary.FileWriter(logs_path, sess.graph)\n",
    "    hist_angles = []\n",
    "    \n",
    "    for i in range(0, num_iterations):\n",
    "        batch_img, batch_class, batch_quat = batch_generator(S_train_img, S_train_class, S_train_quat,\n",
    "                                                     S_test_img, S_test_class, S_test_quat, batch_size)\n",
    "        \n",
    "        _, loss_val, summary = sess.run([optimizer, loss, merged_summary], feed_dict={x: batch_img})\n",
    "\n",
    "        if i % 10 == 0:\n",
    "            summary_writer.add_summary(summary, i)\n",
    "            \n",
    "        if i % 1000 == 0:\n",
    "            print(\"Loss value at iteration \" + str(i) + \"/\" + str(num_iterations-1) + \" : \" + str(loss_val))\n",
    "            save_path = saver.save(sess, \"./checkpoint/model\" + str(i) + \".ckpt\")\n",
    "\n",
    "            db_features, test_features = output_features(S_db_img, S_test_img, layer_fc2, loss, x, i)\n",
    "            hist, indeces, _ = matching_feature_map(db_features, test_features, S_db_class, S_test_class, S_db_quat, S_test_quat)\n",
    "            hist_angles.append(hist)\n",
    "            save_histogram(hist, i)\n",
    "            print(hist), print()\n",
    "            \n",
    "    save_path = saver.save(sess, \"./checkpoint/model.ckpt\")\n",
    "    \n",
    "save_plot_angles(hist_angles)\n",
    "summary_writer.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "#index_saved_model = 55000\n",
    "#db_features, test_features = output_features(S_db_img, S_test_img, layer_fc2, loss, x, index_saved_model)\n",
    "#hist, indeces, conf_matrix = matching_feature_map(db_features, test_features, S_db_class, S_test_class, S_db_quat, S_test_quat)\n",
    "#print(hist)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.contrib.tensorboard.plugins import projector"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## PCA for feature map (16D → 3D)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "LOG_DIR = './logs'\n",
    "names = [\"ape\", \"benchvise\", \"cam\", \"cat\", \"duck\"]\n",
    "\n",
    "#Classification of test data in PCA 3D\n",
    "metadata_file = open(os.path.join(LOG_DIR, 'metadata.tsv'), 'w')\n",
    "metadata_file.write('Name\\tClass\\n')\n",
    "for i in range(len(S_test_class)):\n",
    "    metadata_file.write('%06d\\t%s\\n' % (i, names[S_test_class[i]]))\n",
    "metadata_file.close() \n",
    "\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    \n",
    "    embedding_var = tf.Variable(test_features,  name='feature_map')\n",
    "    sess.run(embedding_var.initializer)\n",
    "    \n",
    "    summary_writer = tf.summary.FileWriter(LOG_DIR)\n",
    "    config = projector.ProjectorConfig()\n",
    "    embedding = config.embeddings.add()\n",
    "    embedding.tensor_name = embedding_var.name\n",
    "\n",
    "    #metadata\n",
    "    embedding.metadata_path = os.path.join('metadata.tsv')\n",
    "\n",
    "    projector.visualize_embeddings(summary_writer, config)\n",
    "    saver = tf.train.Saver([embedding_var])\n",
    "    saver.save(sess, os.path.join(LOG_DIR, 'model2.ckpt'), 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Confusion Matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x504 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "names = ['ape', 'benchvise', 'cam', 'cat', 'duck']\n",
    "conf_matrix = np.array(conf_matrix, copy = True, dtype = \"int16\")\n",
    "\n",
    "df = pd.DataFrame(conf_matrix, index = names, columns = names)\n",
    "\n",
    "plt.figure(figsize = (10,7))\n",
    "ax = plt.axes()\n",
    "sn.heatmap(df, annot=True, annot_kws={\"size\": 12}, fmt=\"d\")\n",
    "ax.set_title('Confusion matrix')\n",
    "plt.xlabel('Predicted')\n",
    "plt.ylabel('Actual')\n",
    "plt.show()\n",
    "\n",
    "plt.clf()\n",
    "plt.cla()\n",
    "plt.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}