gan_attack.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib\n",
    "import matplotlib.pylab as plt\n",
    "import IPython.display as ipd\n",
    "from sklearn.metrics import confusion_matrix\n",
    "import os\n",
    "import numpy as np\n",
    "import torch\n",
    "from models import Generator, SpeakerRecognitionModel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<torch.autograd.grad_mode.no_grad at 0x7f50836a91d0>"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.no_grad()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load GAN generator\n",
    "gan_path = 'checkpoint_dcgan_speaker0_beta1.0_clr0.0001_grl0.0001_30000'\n",
    "G_net = Generator(64)\n",
    "G_net.load_state_dict(torch.load(gan_path, map_location='cpu')['G_net'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load Speaker Recognition\n",
    "sr_path = 'sr_model.pt'\n",
    "N_CLASSES = 102\n",
    "SR_model = SpeakerRecognitionModel(N_CLASSES)\n",
    "SR_model.load_state_dict(torch.load(sr_path, map_location='cpu')['model'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Generator(\n",
       "  (preprocess): Linear(in_features=128, out_features=8192, bias=True)\n",
       "  (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (relu): ReLU(inplace)\n",
       "  (block1): Sequential(\n",
       "    (0): ConvTranspose2d(512, 256, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))\n",
       "    (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU(inplace)\n",
       "  )\n",
       "  (block2): Sequential(\n",
       "    (0): ConvTranspose2d(256, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))\n",
       "    (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU(inplace)\n",
       "  )\n",
       "  (block3): Sequential(\n",
       "    (0): ConvTranspose2d(128, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))\n",
       "    (1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): ReLU(inplace)\n",
       "  )\n",
       "  (deconv_out): ConvTranspose2d(64, 1, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1))\n",
       "  (tanh): Tanh()\n",
       ")"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "G_net.cuda()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SpeakerRecognitionModel(\n",
       "  (convs): Sequential(\n",
       "    (0): Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (1): ReLU()\n",
       "    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "    (3): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
       "    (4): ReLU()\n",
       "    (5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
       "  )\n",
       "  (linears): Sequential(\n",
       "    (0): Linear(in_features=16384, out_features=256, bias=True)\n",
       "    (1): Dropout(p=0.5)\n",
       "    (2): Linear(in_features=256, out_features=102, bias=True)\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "SR_model.cuda()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "SPEAKER_ID = 0\n",
    "N_ATTACKS = 100\n",
    "SAMPLES_PER_ATTACK = 256\n",
    "predictions = []\n",
    "accuracy = 0.0\n",
    "for i in range(0, N_ATTACKS*SAMPLES_PER_ATTACK, SAMPLES_PER_ATTACK):\n",
    "    noise = torch.autograd.Variable(torch.randn(SAMPLES_PER_ATTACK, 128)).cuda()\n",
    "    samples = torch.autograd.Variable(G_net(noise)).unsqueeze(1).cuda()\n",
    "\n",
    "    y_hat = SR_model(samples)                                                                 \n",
    "    _, predicted = torch.max(y_hat, 1)\n",
    "    predictions.append(predicted.data.cpu().numpy())\n",
    "    if SPEAKER_ID is not None:\n",
    "        y = torch.ones(SAMPLES_PER_ATTACK).long().cuda() * SPEAKER_ID\n",
    "        accuracy += float((predicted == y).squeeze().sum())\n",
    "predictions = np.array(predictions).flatten()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Attack accuracy 97.54296875%\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x1152 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "if SPEAKER_ID is not None:        \n",
    "    print(\"Attack accuracy {}%\".format(100*accuracy / (N_ATTACKS * SAMPLES_PER_ATTACK)))\n",
    "    cm = confusion_matrix(y_true=[SPEAKER_ID]*len(predictions), y_pred=predictions, labels=np.arange(0, N_CLASSES))\n",
    "    fig, ax = plt.subplots(1, 1, figsize=(16, 16))\n",
    "    ax.imshow(cm, interpolation='nearest', cmap='gray')\n",
    "    tick_marks = np.arange(N_CLASSES)\n",
    "    plt.tight_layout()\n",
    "    ax.set_ylabel('True label')\n",
    "    ax.set_xlabel('Predicted label')\n",
    "else:\n",
    "    predictions = np.array(predictions).flatten()\n",
    "    _ = plt.hist(predictions, bins=102)    \n",
    "    print(\"Attack accuracy {}%\".format(100*np.sum(predictions < 50) / float(len(predictions))))"
   ]
  }
 ],
 "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": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}