Singleモデル_埋め込み_topK計算用.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Singleモデルにおける画像の埋め込みとtopK計算用"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pickle\n",
    "from PIL import Image\n",
    "import matplotlib.pyplot as plt\n",
    "from tqdm import tqdm\n",
    "from keras.models import load_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "config = tf.ConfigProto()\n",
    "config.gpu_options.allow_growth = True\n",
    "tf.keras.backend.set_session(tf.Session(config=config))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "BASE_PATH = './dataset/Pants_all/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_epoch = 99\n",
    "ALPHA = 0.1\n",
    "BETA = 0.05\n",
    "vec_length = 512\n",
    "category = 'Pants'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model_dir = './model/{}/Single_improved/a{}b{}/{}'.format(category,ALPHA,BETA,vec_length)\n",
    "# model_dir = './model/{}/Single_normal/a{}/{}'.format(category,ALPHA,vec_length)\n",
    "model_dir = './model/{}/Single_improved/a{}b{}/{}'.format(category,ALPHA,BETA,vec_length)\n",
    "# os.listdir(model_dir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "con_embNet = load_model(model_dir+'/con_e{}.h5'.format(model_epoch), compile=False)\n",
    "shop_embNet = load_model(model_dir+'/shop_e{}.h5'.format(model_epoch), compile=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# con_embNet = load_model(model_dir+'/con_emb_e{}.h5'.format(model_epoch), compile=False)\n",
    "# shop_embNet = load_model(model_dir+'/shop_emb_e{}.h5'.format(model_epoch), compile=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "生成済みのpickleを利用するため`datagen.py`は使わない"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "f = open(\"./pickle/{}/test_pairs.pickle\".format(category),\"rb\")\n",
    "test_pairs = pickle.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[20455,\n",
       " ['./dataset/Pants_all/id_00020455/comsumer_01.jpg',\n",
       "  './dataset/Pants_all/id_00020455/shop_01.jpg']]"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_pairs[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "gallery = []\n",
    "for tp in test_pairs:\n",
    "    gallery.append([tp[0],tp[1][1]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[20455, './dataset/Pants_all/id_00020455/shop_01.jpg']\n",
      "gallery length = 894\n"
     ]
    }
   ],
   "source": [
    "print(gallery[0])\n",
    "print('gallery length = {}'.format(len(gallery)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(894, 512)\n",
      "(894,)\n"
     ]
    }
   ],
   "source": [
    "emb_vecs = np.zeros((len(gallery),vec_length))\n",
    "ans_ids = np.zeros(len(gallery))\n",
    "print(emb_vecs.shape)\n",
    "print(ans_ids.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 894/894 [00:05<00:00, 163.91it/s]\n"
     ]
    }
   ],
   "source": [
    "for i,g in enumerate(tqdm(gallery)):\n",
    "    img =np.array(Image.open(g[1]).resize((128,128)).convert('RGB'))/255.\n",
    "    ans_ids[i] = g[0]\n",
    "    emb_vecs[i] = shop_embNet.predict(np.expand_dims(img,axis=0))[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "f = open(model_dir+'/gallery{}.txt'.format(model_epoch),'wb')\n",
    "pickle.dump(emb_vecs,f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "def euclid_dist(emb1, emb2):\n",
    "    return np.sum(np.square(emb1 - emb2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(894,)\n"
     ]
    }
   ],
   "source": [
    "query_ids = np.zeros(len(test_pairs))\n",
    "print(query_ids.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([20455., 22017., 10211., 23875.,  6604.])"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "for i,tp in enumerate(test_pairs):\n",
    "    query_ids[i] = tp[0]\n",
    "query_ids[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "N = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 894/894 [00:12<00:00, 73.61it/s]\n"
     ]
    }
   ],
   "source": [
    "topN_id_list = []\n",
    "query_emb_vecs = np.zeros((len(gallery),vec_length))\n",
    "for i,tp in enumerate(tqdm(test_pairs)):\n",
    "    img = np.array(Image.open(tp[1][0]).resize((128,128)).convert('RGB'))/255.\n",
    "    query_emb = con_embNet.predict([np.expand_dims(img.astype(np.float32),axis=0)])[0]\n",
    "    query_emb_vecs[i] = query_emb\n",
    "    # calculation\n",
    "    similarities = np.zeros(len(gallery))\n",
    "    for i,emb_vec in enumerate(emb_vecs):\n",
    "        similarities[i] = euclid_dist(query_emb,emb_vec)\n",
    "    # argsort\n",
    "    idxs = similarities.argsort()\n",
    "    topN_id_list.append(ans_ids[idxs[:N]])  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([31435., 10970., 27441.,  5018., 31873., 33333., 29795.,  5034.,\n",
       "       30218.,  8627.,  7039., 12553., 26290., 10761., 29020.,  7975.,\n",
       "       31718.,  6924., 32873., 19478., 18523., 29206., 26155.,  1453.,\n",
       "       30119.,  5753.,  2449., 27920.,  2839.,   355.,  2163.,  1505.,\n",
       "        5424., 13066.,  4147.,  3402., 27663.,  8174., 15638., 30291.,\n",
       "       13349.,  6178., 33696.,  7857.,  5499., 17414.,  2476.,  3435.,\n",
       "       27649., 14243.])"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "topN_id_list[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "f = open(model_dir+'/topN_id_list{}.txt'.format(model_epoch),'wb')\n",
    "pickle.dump(topN_id_list,f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "f = open(model_dir+'/query{}.txt'.format(model_epoch),'wb')\n",
    "pickle.dump(query_emb_vecs,f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_topk_vals(k_values,ans_ids,topN_id_list):\n",
    "    topk_vals = []\n",
    "    for k in k_values:\n",
    "        cnt = 0\n",
    "        for i,(ans_id,topN_id) in enumerate(zip(ans_ids,topN_id_list)):\n",
    "            if ans_id in topN_id[:k]:\n",
    "                cnt+=1\n",
    "        topk_vals.append(cnt/len(ans_ids))\n",
    "    return topk_vals"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "x = list(range(1,51))\n",
    "topk_vals = calc_topk_vals(x,ans_ids,topN_id_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "f = open(model_dir+'/topk_vals.txt'.format(model_epoch), 'wb')\n",
    "pickle.dump(topk_vals, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.011185682326621925,\n",
       " 0.021252796420581657,\n",
       " 0.03131991051454139,\n",
       " 0.04138702460850112,\n",
       " 0.05145413870246085,\n",
       " 0.05704697986577181,\n",
       " 0.06152125279642058,\n",
       " 0.07270693512304251,\n",
       " 0.07606263982102908,\n",
       " 0.08053691275167785,\n",
       " 0.09284116331096197,\n",
       " 0.09619686800894854,\n",
       " 0.1017897091722595,\n",
       " 0.10626398210290827,\n",
       " 0.11185682326621924,\n",
       " 0.11297539149888143,\n",
       " 0.11521252796420582,\n",
       " 0.11968680089485459,\n",
       " 0.12639821029082773,\n",
       " 0.13087248322147652,\n",
       " 0.1331096196868009,\n",
       " 0.13646532438478748,\n",
       " 0.1476510067114094,\n",
       " 0.15436241610738255,\n",
       " 0.15548098434004473,\n",
       " 0.16331096196868009,\n",
       " 0.17225950782997762,\n",
       " 0.18120805369127516,\n",
       " 0.18680089485458612,\n",
       " 0.19239373601789708,\n",
       " 0.19463087248322147,\n",
       " 0.20022371364653244,\n",
       " 0.20134228187919462,\n",
       " 0.2058165548098434,\n",
       " 0.21252796420581654,\n",
       " 0.21923937360178972,\n",
       " 0.22371364653243847,\n",
       " 0.22930648769574943,\n",
       " 0.232662192393736,\n",
       " 0.2371364653243848,\n",
       " 0.23937360178970918,\n",
       " 0.24608501118568232,\n",
       " 0.2483221476510067,\n",
       " 0.2550335570469799,\n",
       " 0.2606263982102908,\n",
       " 0.2639821029082774,\n",
       " 0.2662192393736018,\n",
       " 0.26733780760626397,\n",
       " 0.27069351230425054,\n",
       " 0.27293064876957496]"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "topk_vals"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/pip/koki_ishizuka/.conda/envs/py35-zukapy/lib/python3.5/site-packages/matplotlib/cbook/deprecation.py:107: MatplotlibDeprecationWarning: Passing one of 'on', 'true', 'off', 'false' as a boolean is deprecated; use an actual boolean (True/False) instead.\n",
      "  warnings.warn(message, mplDeprecation, stacklevel=1)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = list(range(1,51))\n",
    "chancelv=np.array(x)/len(test_pairs)\n",
    "\n",
    "topk_vals = calc_topk_vals(x,ans_ids,topN_id_list)\n",
    "plt.plot(x,topk_vals,c='r',label='TripletLoss({})'.format(round(topk_vals[19],3)))\n",
    "# plt.plot(x,topk_vals,c='r',marker='o',label='improvedTripletLoss({})'.format(round(topk_vals[4],3)))\n",
    "plt.plot(x,chancelv,linestyle='dashed',label='chance level({})'.format(round(chancelv[19],3)))\n",
    "\n",
    "# graph settings\n",
    "plt.legend(loc='best')#,bbox_to_anchor=(1, 0.2))\n",
    "plt.grid('True')\n",
    "plt.title(\"top-k accuracy\")\n",
    "plt.xlabel(\"k-value\")\n",
    "plt.ylabel(\"top-k acc\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "py35-zukapy",
   "language": "python",
   "name": "py35-zukapy"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}