siamese.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "JOlX8afCVlcu"
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import os\n",
    "import shutil\n",
    "from glob import glob\n",
    "from tqdm import tqdm\n",
    "from cv2 import imread, resize\n",
    "\n",
    "import tensorflow as tf\n",
    "import keras\n",
    "from keras.layers import Input, Conv2D, MaxPooling2D, Lambda, Dense, Flatten\n",
    "from keras.models import Model, Sequential\n",
    "from keras.regularizers import l2\n",
    "from keras import backend as K\n",
    "from keras.optimizers import Adam\n",
    "from keras.losses import binary_crossentropy\n",
    "from keras.utils import plot_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "Jwgaam-pocBs"
   },
   "outputs": [],
   "source": [
    "n_classes = 28\n",
    "train_size = 0.6\n",
    "class_size = 20\n",
    "seed = 47\n",
    "input_shape = (100, 100, 1)\n",
    "w, h, c = input_shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "u3GKlFsUZ9BR"
   },
   "outputs": [],
   "source": [
    "def load_data(path):\n",
    "\n",
    "    X = []\n",
    "    for char in range(1, 29):\n",
    "        \n",
    "        class_list = []\n",
    "        files = glob(os.path.join(path, str(char))+'/*.png')\n",
    "        np.random.seed(seed=seed)\n",
    "        files = np.random.choice(files, size=class_size, replace=False)\n",
    "        for f in files[:class_size]:\n",
    "\n",
    "            img = imread(f, 0)\n",
    "            class_list.append(resize(img, (w,h)))\n",
    "\n",
    "        X.append(np.array(class_list))\n",
    "    X = np.array(X)[..., None] / 255.0\n",
    "    \n",
    "    return X"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "k3O_qUkeZ-c_"
   },
   "outputs": [],
   "source": [
    "data = load_data('/content/drive/My Drive/dataset')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "RW95iIcqleTp"
   },
   "outputs": [],
   "source": [
    "def get_batch(batch_size=64):\n",
    "\n",
    "    targets = np.zeros(batch_size)\n",
    "    targets[batch_size//2:] = 1\n",
    "    np.random.shuffle(targets)\n",
    "\n",
    "    pairs = [np.zeros((batch_size, w, h, c)) for i in range(2)]\n",
    "\n",
    "    train_classes = int(train_size*n_classes)\n",
    "    class_one = np.random.choice(train_classes, size=batch_size, replace=False)\n",
    "    idx1 = np.random.randint(0, class_size, batch_size)\n",
    "    idx2 = np.random.randint(0, class_size, batch_size)\n",
    "\n",
    "    for i in range(batch_size):\n",
    "\n",
    "        pairs[0][i] = data[class_one[i], idx1[i]]\n",
    "        \n",
    "        class_two = None\n",
    "        if targets[i] == 1:\n",
    "            class_two = class_one[i]\n",
    "        else:\n",
    "            class_two = (class_one[i] + np.random.randint(1, train_classes)) % train_classes\n",
    "            assert class_two != class_one[i], \"Error, same class\"\n",
    "\n",
    "        pairs[1][i] = data[class_two, idx2[i]]\n",
    "        \n",
    "    return pairs, targets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "hyKMzYisWHvs"
   },
   "outputs": [],
   "source": [
    "def model():\n",
    "    \n",
    "    W_init = keras.initializers.RandomNormal(mean = 0.0, stddev = 0.01)\n",
    "    FC_W_init = keras.initializers.RandomNormal(mean = 0.0, stddev = 0.2)\n",
    "    b_init = keras.initializers.RandomNormal(mean = 0.5, stddev = 0.01)\n",
    "    \n",
    "    input_img1 = Input(input_shape)\n",
    "    input_img2 = Input(input_shape)\n",
    "\n",
    "    model = Sequential()\n",
    "    model.add(Conv2D(64, (10, 10), activation='relu', input_shape=input_shape,\n",
    "                    kernel_initializer=W_init, bias_initializer=b_init, kernel_regularizer=l2(2e-4)))\n",
    "    model.add(MaxPooling2D())\n",
    "\n",
    "    model.add(Conv2D(128, (7, 7), activation='relu',\n",
    "                    kernel_initializer=W_init, bias_initializer=b_init, kernel_regularizer=l2(2e-4)))\n",
    "    model.add(MaxPooling2D())\n",
    "\n",
    "    model.add(Conv2D(128, (4,4), activation='relu',\n",
    "                    kernel_initializer=W_init, bias_initializer=b_init, kernel_regularizer=l2(2e-4)))\n",
    "    model.add(MaxPooling2D())\n",
    "    \n",
    "    model.add(Conv2D(256, (4,4), activation='relu',\n",
    "                    kernel_initializer=W_init, bias_initializer=b_init, kernel_regularizer=l2(2e-4)))\n",
    "    model.add(Flatten())\n",
    "    \n",
    "    model.add(Dense(4096, activation='sigmoid',\n",
    "                   kernel_initializer=FC_W_init, bias_initializer=b_init, kernel_regularizer=l2(1e-3)))\n",
    "    \n",
    "    encoding1 = model(input_img1)\n",
    "    encoding2 = model(input_img2)\n",
    "\n",
    "    distance = Lambda(lambda embed: K.abs(embed[0] - embed[1])) ([encoding1, encoding2])\n",
    "    prediction = Dense(1, activation='sigmoid', bias_initializer=b_init)(distance)\n",
    "\n",
    "    siamese_net = Model(inputs=[input_img1, input_img2], outputs=prediction)\n",
    "\n",
    "    return siamese_net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 762
    },
    "colab_type": "code",
    "id": "cc7oMBW8bSN8",
    "outputId": "0291e786-a5d5-49e2-ffa6-55c14f303b28"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"functional_1\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            [(None, 100, 100, 1) 0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_2 (InputLayer)            [(None, 100, 100, 1) 0                                            \n",
      "__________________________________________________________________________________________________\n",
      "sequential (Sequential)         (None, 4096)         27413312    input_1[0][0]                    \n",
      "                                                                 input_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "lambda (Lambda)                 (None, 4096)         0           sequential[0][0]                 \n",
      "                                                                 sequential[1][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 1)            4097        lambda[0][0]                     \n",
      "==================================================================================================\n",
      "Total params: 27,417,409\n",
      "Trainable params: 27,417,409\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 7,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "siamese_net = model()\n",
    "optimizer= Adam(learning_rate=0.0006)\n",
    "siamese_net.compile(loss='binary_crossentropy', optimizer=optimizer)\n",
    "\n",
    "siamese_net.summary()\n",
    "plot_model(siamese_net, show_shapes=True, show_layer_names=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "QLp3RH6kZc7x"
   },
   "outputs": [],
   "source": [
    "def nway_one_shot(n_way, split='val'):\n",
    "\n",
    "    train_classes = int(train_size*n_classes)\n",
    "    val_classes = [x for x in range(train_classes, n_classes)] if split == 'val' else [x for x in range(0, train_classes)]\n",
    "\n",
    "    classes = np.random.choice(val_classes, size=n_way, replace=False)\n",
    "    pairs = [np.zeros((n_way, w, h, c)) for k in range(2)]\n",
    "    true_class = classes[0]\n",
    "    indices = np.random.randint(0, class_size, size=n_way)\n",
    "    idx1, idx2 = np.random.choice(class_size, size=2, replace=False)\n",
    "\n",
    "    for k in range(n_way):\n",
    "        \n",
    "        pairs[0][k] = data[true_class][idx1]\n",
    "\n",
    "        if k == 0:\n",
    "            pairs[1][k] = data[true_class][idx2]\n",
    "        else:\n",
    "            pairs[1][k] = data[classes[k]][indices[k]]\n",
    "\n",
    "    return pairs        \n",
    "\n",
    "def one_shot_task(n_val, n_way, split='val'):\n",
    "\n",
    "    n_correct = 0\n",
    "    for i in range(n_val):\n",
    "\n",
    "        pairs = nway_one_shot(n_way, split)\n",
    "        probs = siamese_net.predict(pairs)\n",
    "        if np.argmax(probs) == 0:\n",
    "            n_correct += 1\n",
    "        \n",
    "    accuracy = 100 * n_correct / n_val\n",
    "    # print(f'Accuracy = {accuracy}')\n",
    "    return accuracy\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "8tfzoEUvZd7M"
   },
   "outputs": [],
   "source": [
    "epochs = 200*1000\n",
    "every = 7000\n",
    "batch_size = 16\n",
    "n_val = 200\n",
    "n_way = 12\n",
    "best = 80\n",
    "model_path = '/content/drive/My Drive/Colab Notebooks'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "colab_type": "code",
    "id": "vxXKAhpPZq65",
    "outputId": "3ffc139d-0efe-414b-d059-56c24d16b2b4",
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "for i in range(epochs):\n",
    "\n",
    "    pairs, targets = get_batch(batch_size=batch_size)\n",
    "    loss = siamese_net.train_on_batch(pairs, targets)\n",
    "    if i % every == 0:\n",
    "        print(\"\\n ------------- \\n\")\n",
    "        print(\"Train Loss: {0}\".format(loss)) \n",
    "        val_acc = one_shot_task(n_val, n_way)\n",
    "        if val_acc >= best:\n",
    "            siamese_net.save_weights(os.path.join(model_path, f'weights.{seed}.{i}.h5'))\n",
    "            print(f'\\nCurrent best: {val_acc}, previous best: {best}')\n",
    "            best = val_acc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "9o3fQ-wQbRc3"
   },
   "outputs": [],
   "source": [
    "def nn_model(nway):\n",
    "\n",
    "    pairs = nway_one_shot(nway)\n",
    "    distance = np.zeros(nway)\n",
    "    for i in range(nway):\n",
    "\n",
    "        distance[i] = np.sum((pairs[0][i] - pairs[1][i])**2)\n",
    "    \n",
    "    return np.argmin(distance) == 0\n",
    "\n",
    "def test_nn(trials, nway):\n",
    "\n",
    "    correct = 0\n",
    "    for i in range(trials):\n",
    "        correct += nn_model(nway)\n",
    "\n",
    "    return 100 * correct / trials"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "ys7QZ_TdBCJf"
   },
   "source": [
    "# Testing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "PnNdefxWtzo7"
   },
   "outputs": [],
   "source": [
    "siamese_net.load_weights('/content/drive/My Drive/Colab Notebooks/weights.47.14000.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "colab_type": "code",
    "id": "AKbMKXiqBBWQ",
    "outputId": "48095afd-b53f-4fb9-c867-c54b601df294"
   },
   "outputs": [],
   "source": [
    "train_acc, val_acc, nn_acc = [], [], []\n",
    "trials = 500\n",
    "ways = [x for x in range(2, 13)]\n",
    "\n",
    "for nway in tqdm(ways):\n",
    "\n",
    "    train_acc.append(one_shot_task(trials, nway, split='train'))\n",
    "    val_acc.append(one_shot_task(trials, nway, split='val'))\n",
    "    nn_acc.append(test_nn(trials, nway))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "bAFvfHEsv3x-",
    "outputId": "11861f32-46e2-40d4-8519-d9fcc39b7976"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[99.8, 99.6, 99.0, 98.6, 98.8, 97.4, 97.2, 96.8, 96.6, 97.0, 96.0]"
      ]
     },
     "execution_count": 58,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_acc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "0NHALHd4wiPv",
    "outputId": "a0c8acd0-b955-4409-ac81-d1da82077828"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[97.8, 97.2, 98.2, 98.4, 96.6, 95.4, 95.0, 95.6, 94.0, 96.0, 92.8]"
      ]
     },
     "execution_count": 59,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "val_acc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "OG7GXdLlwlsO",
    "outputId": "517d24bf-dff9-4e3f-fa14-0a030c747b5e"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[57.8, 45.8, 37.4, 31.4, 31.0, 26.2, 23.2, 19.8, 19.6, 22.8, 15.2]"
      ]
     },
     "execution_count": 60,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "nn_acc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 295
    },
    "colab_type": "code",
    "id": "RvZEaekc5-57",
    "outputId": "65d5d087-5c4b-4375-b710-abba26b16b2f"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib. pyplot as plt\n",
    "fig,ax = plt.subplots(1)\n",
    "ax.plot(ways, train_acc, \"r\", label=\"Siamese(train set)\")\n",
    "ax.plot(ways, val_acc, \"b\", label=\"Siamese(val set)\")\n",
    "plt.plot(ways, nn_acc, \"g\", label=\"Nearest neighbour\")\n",
    "\n",
    "plt.xlabel(\"Number of possible classes in one-shot tasks\")\n",
    "plt.ylabel(\"% Accuracy\")\n",
    "plt.title(\"Arabic One-Shot Learning Performance of a Siamese Network\")\n",
    "ax.legend()\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "siamese",
   "provenance": [],
   "toc_visible": true
  },
  "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}