{
"cells": [
{
"cell_type": "markdown",
"id": "7f959254",
"metadata": {},
"source": [
"# Feature Adversaries Attack in PyTorch\n",
"\n",
"Before diving into the attack, let's first prepare a classification model. We utilize a script from the `examples` folders."
]
},
{
"cell_type": "code",
"execution_count": 1,
"id": "d4de108e",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Accuracy on benign test examples: 96.61999999999999%\n"
]
}
],
"source": [
"import torch\n",
"import torch.nn as nn\n",
"import torch.nn.functional as F\n",
"import torch.optim as optim\n",
"import matplotlib.pyplot as plt\n",
"import numpy as np\n",
"\n",
"from art.attacks.evasion import FeatureAdversariesPyTorch\n",
"from art.estimators.classification import PyTorchClassifier\n",
"from art.utils import load_mnist\n",
"\n",
"\n",
"# Step 1: Load the MNIST dataset\n",
"\n",
"(x_train, y_train), (x_test, y_test), min_pixel_value, max_pixel_value = load_mnist()\n",
"\n",
"# Step 1a: Swap axes to PyTorch's NCHW format\n",
"\n",
"x_train = np.transpose(x_train, (0, 3, 1, 2)).astype(np.float32)\n",
"x_test = np.transpose(x_test, (0, 3, 1, 2)).astype(np.float32)\n",
"\n",
"# Step 2: Create the model\n",
"\n",
"model = nn.Sequential(\n",
" nn.Conv2d(1, 4, 5), nn.ReLU(), nn.MaxPool2d(2, 2), \n",
" nn.Conv2d(4, 10, 5), nn.ReLU(), nn.MaxPool2d(2, 2),\n",
" nn.Flatten(), \n",
" nn.Linear(4*4*10, 100), \n",
" nn.Linear(100, 10)\n",
")\n",
"# Step 2a: Define the loss function and the optimizer\n",
"\n",
"criterion = nn.CrossEntropyLoss()\n",
"optimizer = optim.Adam(model.parameters(), lr=0.01)\n",
"\n",
"# Step 3: Create the ART classifier\n",
"\n",
"classifier = PyTorchClassifier(\n",
" model=model,\n",
" clip_values=(min_pixel_value, max_pixel_value),\n",
" loss=criterion,\n",
" optimizer=optimizer,\n",
" input_shape=(1, 28, 28),\n",
" nb_classes=10,\n",
")\n",
"\n",
"# Step 4: Train the ART classifier\n",
"\n",
"classifier.fit(x_train, y_train, batch_size=64, nb_epochs=3)\n",
"\n",
"# Step 5: Evaluate the ART classifier on benign test examples\n",
"\n",
"predictions = classifier.predict(x_test)\n",
"accuracy = np.sum(np.argmax(predictions, axis=1) == np.argmax(y_test, axis=1)) / len(y_test)\n",
"print(\"Accuracy on benign test examples: {}%\".format(accuracy * 100))"
]
},
{
"cell_type": "markdown",
"id": "8a174860",
"metadata": {},
"source": [
"---\n",
"\n",
"## PGD variant of Feature Adversaries\n"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "2bbdefd4",
"metadata": {},
"outputs": [],
"source": [
"# Step 5: prepare a batch of source and guide images\n",
"valid = np.argmax(y_test, axis=1)[:100] != np.argmax(y_test, axis=1)[100:200]\n",
"source = x_test[:100][valid][:32]\n",
"guide = x_test[100:200][valid][:32]"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "c900ec77",
"metadata": {
"scrolled": true
},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "755d5f9376cd4c9f97782ad8405d5775",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Feature Adversaries PyTorch: 0%| | 0/100 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Accuracy on adversarial test batch: 3.125%\n",
"Average perturbation: 0.13725492358207703%\n"
]
}
],
"source": [
"# Step 6: Generate adversarial test examples\n",
"attack = FeatureAdversariesPyTorch(\n",
" classifier,\n",
" layer=7,\n",
" delta=35/255,\n",
" optimizer=None,\n",
" step_size=1/255,\n",
" max_iter=100,\n",
")\n",
"x_test_adv = attack.generate(source, guide)\n",
"\n",
"# Step 7: Evaluate the ART classifier on adversarial test examples\n",
"\n",
"predictions = classifier.predict(x_test_adv)\n",
"accuracy = np.sum(np.argmax(predictions, axis=1) == np.argmax(y_test[:100][valid][:32], axis=1)) / len(y_test[:100][valid][:32])\n",
"\n",
"dim = tuple(range(1, len(source.shape)))\n",
"pert = np.mean(np.amax(np.abs(source - x_test_adv), axis=dim))\n",
"print(\"Accuracy on adversarial test batch: {}%\".format(accuracy * 100))\n",
"print(\"Average perturbation: {}%\".format(pert))"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "68717bc9",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a6bc4bb0>"
]
},
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# Step 8: Inspect results\n",
"\n",
"# orig 7, guide 6\n",
"plt.imshow(x_test_adv[0,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "5aabd4ae",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a4ac74c0>"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 1, guide 5\n",
"plt.imshow(x_test_adv[2,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "1765536f",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a4ae6b50>"
]
},
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 4, guide 9\n",
"plt.imshow(x_test_adv[4,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "1361622b",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a4a17190>"
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 4, guide 2\n",
"plt.imshow(x_test_adv[6,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "e6f7cb47",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a49f48b0>"
]
},
"execution_count": 8,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 5, guide 9\n",
"plt.imshow(x_test_adv[8,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "1489dc86",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a4952ee0>"
]
},
"execution_count": 9,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 0, guide 8\n",
"plt.imshow(x_test_adv[10,...].squeeze())"
]
},
{
"cell_type": "markdown",
"id": "18bc254a",
"metadata": {},
"source": [
"\n",
"## Unconstrained variant of Feature Adversaries\n",
"\n",
"This variant approximates the original hard constraint problem of the paper. Any available PyTorch optimizer can be used. We will use Adam as a good default one.\n"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "70d9fbba",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "3976663efe8e41208e8fa7eef698efa4",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Feature Adversaries PyTorch: 0%| | 0/100 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"Accuracy on adversarial test batch: 0.0%\n",
"Average perturbation: 0.2517760694026947%\n"
]
}
],
"source": [
"# Step 6: Generate adversarial test examples\n",
"attack = FeatureAdversariesPyTorch(\n",
" classifier,\n",
" layer=7,\n",
" delta=61/255,\n",
" optimizer=optim.Adam,\n",
" optimizer_kwargs={\"lr\": 0.01},\n",
" lambda_=1.0,\n",
" max_iter=100,\n",
" random_start=True\n",
")\n",
"x_test_adv = attack.generate(source, guide)\n",
"\n",
"# Step 7: Evaluate the ART classifier on adversarial test examples\n",
"\n",
"predictions = classifier.predict(x_test_adv)\n",
"accuracy = np.sum(np.argmax(predictions, axis=1) == np.argmax(y_test[:100][valid][:32], axis=1)) / len(y_test[:100][valid][:32])\n",
"\n",
"dim = tuple(range(1, len(source.shape)))\n",
"pert = np.mean(np.amax(np.abs(source - x_test_adv), axis=dim))\n",
"print(\"Accuracy on adversarial test batch: {}%\".format(accuracy * 100))\n",
"print(\"Average perturbation: {}%\".format(pert))"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "78e7ed4e",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a48cd370>"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# Step 8: Inspect results\n",
"\n",
"# orig 7, guide 6\n",
"plt.imshow(x_test_adv[0,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "dce73b0f",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a48a2e50>"
]
},
"execution_count": 12,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 1, guide 5\n",
"plt.imshow(x_test_adv[2,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "8082690b",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a480a610>"
]
},
"execution_count": 13,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 4, guide 9\n",
"plt.imshow(x_test_adv[4,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "a68f96b7",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a47f0190>"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 4, guide 2\n",
"plt.imshow(x_test_adv[6,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "7a16e8fb",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a48319d0>"
]
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
"image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8rg+JYAAAACXBIWXMAAAsTAAALEwEAmpwYAAAYQklEQVR4nO3de3CcV3kG8OfdXV1Xkm3FtqzYkm9xDHYuDlZuTkpDTXMxZBzagZC0adqmNdOBGRhoKYVQ0ul0Js2UpPyRgRElxaRAgAGSACHBCcHGOI4tJ8ZWYse3yLZsXZzYsixZ0t7e/qEVowSd9yi70u625/nNeCTvo/N9R6t9tdo93zlHVBVE9P9fpNgdIKLCYLETBYLFThQIFjtRIFjsRIGIFfJk5VKhlYi7v6C60j7A+eGp7dB4InYccf9ezFRX2G0Hhuy8vMzMkc7YudE3TSTstjVVZqwRz/2S8YzmeL73fEiV5/FifO/qu0995y4vN3Pv/W4dOxq1j13lPvfwcB8SycEJf2h5FbuI3AzgKwCiAP5LVe+3vr4ScVwta93HW7HSPJ/ueiWHXk6OlNk/vEjcXRQjqy8y25Zt2Wsfu3m+mePsgBlLTbUzS71+1GybabnCzFOV9gOvbCBl5vKb3Waej8i7VtjnPtLpzNL9/XmdOza/2cxTHcdyPnZ0xiwzT162yJntbHvYmeX8Z7yIRAE8DOAWACsA3CEi9r1PREWTz2v2qwAcUtUjqpoA8BiA9VPTLSKaavkU+3wAx8f9vzN721uIyAYRaRORtiRG8jgdEeVj2t+NV9VWVW1R1ZYy2G9kEdH0yafYTwBoGvf/BdnbiKgE5VPsOwEsE5HFIlIO4KMAnpyabhHRVMt56E1VUyLyCQDPYHTo7RFVzWtsLDKUNHOZ1+DMkksb7baeISBN2uOiEp/tzCq67aEx9YyjZ2a4h84AYOjdc8zcdIn7PpuMmpfcw1cAkO7pNfPokkXO7Pwy930KANXtJ+1ze66NyBjDa8n3rzbbVnQPmrkm7MdqbP6FZp7ufcN97Cb7ZxbZ/LLRMfd1DXmNs6vqUwCeyucYRFQYvFyWKBAsdqJAsNiJAsFiJwoEi50oECx2okAUdD67T/rVA2Y+/MGrnFl5n2ec3HNuabnE/oI3jCmRvaftYy9aYOYZz9TdqsilZq473VNoh29132eTMbjKnn5bfcCeU64x9xTZqsNv2ic32gJA5HX7GgDMck8VHa7yHPucPc6eOd1n5nKBPU1Vosbz7IEOs2105gz3cfvd3xef2YkCwWInCgSLnSgQLHaiQLDYiQLBYicKhBRyY8cZFQ26Zt6dzjx13DOUYoguW2Lmqdm1dvtz9pJZkb5zziwz0z72+UV1Zl750x1m/n9Z5LJ3ObN03F65qOzYKTP3Ta/VlHvlW+9qwnU19rlPnzFzWW2vlIy9B93nrjGWWweQaXZP596+vxX9gycnHGnmMztRIFjsRIFgsRMFgsVOFAgWO1EgWOxEgWCxEwWioFNcNZE0x9KjdfZ49EjLMnf4y11mW3EPawIAEmvtpYVj7fudWaTePZYMANXP21NYfZsHW1MaASDdd9ZzhOKJDLi32R6ZZ49lx4zdaQEgirlmnpp/gTPTHfbOuq/de7GZR5L2pOmLvmP/TMQYS0+/aU+ZhpWr+/7mMztRIFjsRIFgsRMFgsVOFAgWO1EgWOxEgWCxEwWitJaSXrHIzMu373Nm6pmf7NuSuazfzoeMJZmrnt6d17nzXe65sse9Te9QY5XZVnyD/B417T1m/tq/updUjsbs+2X7mm+Z+aYhe5vuLWfd1z880GivIfAv9lR6XFF91Myfea+9NPnRf17uzGIDTWZb2fZbM3ceN6dWYycV6QBwDkAaQEpVW/I5HhFNn6l4Zn+fqrp3lieiksDX7ESByLfYFcAvRGSXiGyY6AtEZIOItIlIWxL2Om9ENH3y/TP+elU9ISJzAWwSkf2qumX8F6hqK4BWAKiT+sKtbklEb5HXM7uqnsh+7AXwYwD5va1MRNMm52IXkbiI1I59DuBGAO1T1TEimlr5/BnfAODHIjJ2nO+o6tNmi+pKyAr3etq6fY/ZfGjdlc6sclNuY49josftNcjjx9yvQDKXGPPsAUQOHTPzyp/YY77Ri5eaefrAYWdmj7IDx35gbwe977pHPUewHUi6tz6e6XmqOZ2xX/WdTNrbIv/dnF85s2eH3HPdAaB3xN4LYF6dPV99Rc1J+/ivGms3WNs5A1Dj8SAdW51ZzsWuqkcAXJ5reyIqLA69EQWCxU4UCBY7USBY7ESBYLETBaKwU1zPD0N3uZdVzmeqp28aqU+q256qafJtHZz7kQHYQ2sAELn83c7s4F32MtSHrvtaTn0a03r2QjMfSFc6s56kvXR4Y3mfmZ8YsYfe9sbcffun5z9sto3PdQ8ZAsCLP7OHLH2aOrfl1d5F1X1JOp/ZiQLBYicKBIudKBAsdqJAsNiJAsFiJwoEi50oECW1lLRvqici0Wk7d+b6Vfapt+52ZrLaPW0XgHltAQB0fO8yM0+esieqNj3jHslf+vfbzba4045fTw6Y+Q9O2ltdH925wJnNeyFttj3xPvu5qPnnKTPfWn+1M7v4sD2OPtzg3lJ5lH3u8r6kp33uosuWODM5+mtnxmd2okCw2IkCwWInCgSLnSgQLHaiQLDYiQLBYicKREmNs3tljHHZ0SWt3dSeVZ6YZW/5DGOufaLW/p05cO8KM1/T+LqZ79phz52u6nQva+ybS/+B99xk5gNXLjTzyp/tMvOLlrvns6f3HTTbLnu6wsyHbrQXN565t8+ZZdr3m23LbniPmSdr7dJJzCwzc/e9AqTfZ58bz7/kjFTd6zrwmZ0oECx2okCw2IkCwWInCgSLnSgQLHaiQLDYiQJRUuPsvnXja/Z0ObPEotlm28jml808vtked+2/0b02+9988XGz7T0zus38Y53Xmnn8ZMbM9WV7vrxl5N3zzbxswJ63HWtsMPPUDPdc/NgC+9zJZvtnKmn7KgLfWLol+iv3WDYATN/KCkDUGEcH7DrRzS84M+8zu4g8IiK9ItI+7rZ6EdkkIgezH+3V+omo6CbzZ/w3Adz8tts+B+A5VV0G4Lns/4mohHmLXVW3ADj9tpvXA9iY/XwjgNumtltENNVyfc3eoKpjL6C7AThfuInIBgAbAKAS1Tmejojylfe78aqqMOZbqGqrqraoaksZ7IkNRDR9ci32HhFpBIDsR3sbUyIqulyL/UkAd2c/vxvAE1PTHSKaLt7X7CLyXQA3AJgtIp0AvgTgfgDfF5F7ABwF8JHp7OSYhDHumonav7fKPWO6qc4TZr7ys3uc2eHhuWbbazuuM/O6W+z918tvtcfZ89nX3rtW/1X2XPrBVfb9qhH3OgPlZfb9Fvm1fW1EqC8Kq7rOOzNJuh8r3mJX1Tsc0Vpvr4ioZPByWaJAsNiJAsFiJwoEi50oECx2okCU1BRX3zBQdOVyZza4ZIbZ9txqe4iocuEcM09k3MNjTxyxh6cW/GnuU1ABIH7g7VMT3ir92qG8jm/asdeM44vtpaYzdcYl0r4pqmYKRGfZky0Hr1/mzCp/utM+uG/p8Zta7Paepc3Ln/ac36Bt7UY47Iz4zE4UCBY7USBY7ESBYLETBYLFThQIFjtRIFjsRIEQ9YwnTqU6qderozc68+gF9Wb79JvGeLO1nTOASG2tma/bfszM/zD+mjO77YlPmW2XPO7eRhfwLx2cD999OrjmIjOXlP34qHjG7nvk0ovdxx5Omm2RsUfaM3XuZaoBAO3u6w90ZMRu6xHzXF8wcIm9xLbFd59XH3Vv0f3C4UdwdqhrwkF+PrMTBYLFThQIFjtRIFjsRIFgsRMFgsVOFAgWO1EgCjqfXWurkbp6lfsLBu1x1+hM91h5+vBRs600NZr57XVPm/mX33AvB33kw18z2/75lTeY+aGHrzHzGd/ebuYWqYnn3BYAKn/pXkIbAKJz7W2VcXbQnXnmfGPIPTcbAKJpz1bWFy1yZsONNWbbsmd3mblPfPvrZj54zWJnVnVywGwrZ/rdYdp9vQmf2YkCwWInCgSLnSgQLHaiQLDYiQLBYicKBIudKBAFnc8+o6JB18y705mnjnfmfvBI1M49892bX7THo1Pq/r14ee1xs+1N8VfNfGbEHi++9+TNZn7800ud2Ui9vbFxVc+QmXeutdcBmLPbvjai4zb3WPq8LfZzzaxnDpg5MvZjV2bWObPzyz3bRSfsn0kkYT+eNGZ/b9HNxnbUnpqMNc5zZttOfQ9nE725zWcXkUdEpFdE2sfddp+InBCR3dl/63zHIaLimsyf8d8EMNFTy0Oquir776mp7RYRTTVvsavqFgD2/kNEVPLyeYPuEyKyJ/tnvnPTLRHZICJtItKWSNuvD4lo+uRa7F8FsBTAKgBdAL7s+kJVbVXVFlVtKY96FggkommTU7Grao+qplU1A+DrAK6a2m4R0VTLqdhFZPx80Q8BMPaQJaJS4B1nF5HvArgBwGwAPQC+lP3/KgAKoAPAx1S1y3eyGeVzdc2c2515qqvbbG+NL/ra+vT9xbVm/oUvbnRmV1b0mm0bY/bc6R0j9lj1VRVlZr7bWAP94d4/Mtt+vek3Zv7g6SVmfjplX59wOunOD/fbc+FPDdrHnvcZz7rz1nx5z1z65BzPfPf99rUVMMb4ASB9yD3f3XqcA8BAS7Mz2735KzjX1znhN+ddvEJV75jg5m/42hFRaeHlskSBYLETBYLFThQIFjtRIFjsRIEo/BTXC//MmaeOeoYzDMO35nddT+VPdph5dOYMZ3ak1T0UAgD7r380pz4Vwl8d+wMzb6xwbw8MAHPLjWWNYU/v/c/etWbbDXM2m/ldu/7azJv/zf3YFmPJZQDAgQ4zzgx7lrk2Hi8AkO5z36967eVm27Jjp5zZtp7HcDbRwy2biULGYicKBIudKBAsdqJAsNiJAsFiJwoEi50oEIXdsrkshuT8emcueYyz+8bJ82WNizY/ZP/O/MA/3Gof+2SPmUfq7OmWT+39pZlb/rv512Z+Jn3ezD994iYz/+25pnfcpzEHEw1mHq9M2AeIuZfRHmqsNpuW177LzMs67GnNmvBsP97gXso6vd3eJnvw5hZnltnqng7NZ3aiQLDYiQLBYicKBIudKBAsdqJAsNiJAsFiJwpEQcfZZSSB2GvusXTPDOO85DO/2Gdktr0tsryQ+/UDAJB+095q74MHbnFm9RWDZtuo2OsZbHlhpZkvftIe666574Qze3zZM2Zbn8dq7Z9Zsts95lyx094e3Lc+wkj9AjOP7+gw83SPPU5vqfj5Tmcm6r4ugs/sRIFgsRMFgsVOFAgWO1EgWOxEgWCxEwWCxU4UiIKOs2eqKjB8xWJnXvasPZ6cD984eqzJHjdNHXePy9a0HbPbmmn+kjcYu2W/YG8dfG/jz8186e32fPdLF91p5g8ufMJIK822Ps1x+/Gyv9l9jYAYP08g//URpvOakVx5n9lFpElEnheRV0XkFRH5ZPb2ehHZJCIHsx9nTX93iShXk/kzPgXgM6q6AsA1AD4uIisAfA7Ac6q6DMBz2f8TUYnyFruqdqnqS9nPzwHYB2A+gPUANma/bCOA26apj0Q0Bd7RG3QisgjAFQBeBNCgqmMvFrsBTLhgmIhsEJE2EWlLJu3rtIlo+ky62EWkBsAPAXxKVd+ym5+O7g454YwKVW1V1RZVbSkri+fVWSLK3aSKXUTKMFro31bVH2Vv7hGRxmzeCCD3aTxENO28Q28iIgC+AWCfqj44LnoSwN0A7s9+tMZYAACRwWFUvHjAmWd8BzAk37/azFPxqJmL7+SrLnRGvmGaofX5bSdd9YR9/HO3X+PMDrTbg0APxUbMfCjtniYKAD9b3Wrmx1PuJZsPJwfMth/45mfNfOGXtpl5bLn7+FptLyWtCXvqrqbsAdXo7AvMPP3Gm2ZuHnv5Rc5MOrY6s8mMs18H4C4Ae0Vkd/a2z2O0yL8vIvcAOArgI5PsKxEVgbfYVXUrgAk3dwewdmq7Q0TThZfLEgWCxU4UCBY7USBY7ESBYLETBUJGL34rjDqp16sl9zfwreV94/tO2Y3P2FNcpdbeFnlwpXv74EjSvg8zZa7BjFHxPcYUVQDnV8wz87J+9/bAp1fY48mPfvHLZv5At70lc2TiCyd/JxZxj/O/8u+XmW3jP3zRzH2krNzI7IEoqbKn3w6ucY91A0C58TMBgMjml83cYk3H3tb9HZwd6ZnwAcdndqJAsNiJAsFiJwoEi50oECx2okCw2IkCwWInCkRBx9lnxGbrtTXrnXm6v9+Z+cSWLDLzgZVzzVw9v/aquobc5+7pM9smmmfb547Z4/DlvfZyXula95bRqWp7Prpk7J//kT+x25efsdcJaGhzz/vOd7nmUMUWL3Rm2zr/B2dHujnOThQyFjtRIFjsRIFgsRMFgsVOFAgWO1EgWOxEgSjols2IRCF1tc7YHrEFUGGMJx/pMJtWenKf1Fr3uvSRfX1m2/IRzxrknjx95oyZ4xr3vHD/evn2OHvTJt91GLlvSG2tTwBM4toHz3r6+Zx7uq8BsPYS8H1f6U73+geadM+j5zM7USBY7ESBYLETBYLFThQIFjtRIFjsRIFgsRMFYjL7szcB+BaABgAKoFVVvyIi9wH4WwBjC7Z/XlWfss8WRabePc4u/efM5unF7vXTY1HP762Y/a1qrb2+erT9uDtssOerjyyYZea+sfB4e7eZZ/YedmaVQ8Nm28hly838fJO9nn58++tmrhe67xtJZcy2MuheQwDwj/CPrLvSmWU8awj4+MbpPcvp53WNgCaN6zKM9Skmc1FNCsBnVPUlEakFsEtENmWzh1T1P95BP4moSCazP3sXgK7s5+dEZB+A+dPdMSKaWu/oNbuILAJwBYCxfXk+ISJ7ROQREZnwb1UR2SAibSLSlkjZyysR0fSZdLGLSA2AHwL4lKr2A/gqgKUAVmH0mX/CTcNUtVVVW1S1pTwWz7/HRJSTSRW7iJRhtNC/rao/AgBV7VHVtKpmAHwdgOcdCyIqJm+xi4gA+AaAfar64LjbG8d92YcAtE9994hoqkzm3fjrANwFYK+I7M7e9nkAd4jIKowOMnQA+Jj3SIkEpOOkM/YtJa0x9+8mTdvDOEjb00jhmWZqOjtgxtFD9vBUzNhaGAAwu97OFze5sxr72JEeeyvrmj77exu+tNnMKzvedGa+acmyeqWZw26Oiqd22l9gGLrNM/1W7KE78SzRbk1x9cl12G4y78ZvBTDRd2aPqRNRSeEVdESBYLETBYLFThQIFjtRIFjsRIFgsRMForBLSXtE6+rMPGUMXaZPnXKHkzn3nDlmbh0/umyJ2VaaPdtFv/Sqmae67Cmu8OXWsXNuOSrWcczMM/HcL5HWXa+Y+cgt7imsAFDebyyrnEibbaset8ey9drLzbzMc/2CdY1BdPYFZltZ6L6uQk66t9jmMztRIFjsRIFgsRMFgsVOFAgWO1EgWOxEgWCxEwVC1DPvdkpPJnIKwNFxN80G8EbBOvDOlGrfSrVfAPuWq6ns20JVnfCikYIW+++dXKRNVVuK1gFDqfatVPsFsG+5KlTf+Gc8USBY7ESBKHaxtxb5/JZS7Vup9gtg33JVkL4V9TU7ERVOsZ/ZiahAWOxEgShKsYvIzSLymogcEpHPFaMPLiLSISJ7RWS3iLQVuS+PiEiviLSPu61eRDaJyMHsR3s/6ML27T4ROZG973aLyLoi9a1JRJ4XkVdF5BUR+WT29qLed0a/CnK/Ffw1u4hEARwA8McAOgHsBHCHqtorOBSIiHQAaFHVol+AISLvBTAA4Fuqekn2tgcAnFbV+7O/KGep6j+WSN/uAzBQ7G28s7sVNY7fZhzAbQD+EkW874x+fQQFuN+K8cx+FYBDqnpEVRMAHgOwvgj9KHmqugXA6bfdvB7AxuznGzH6YCk4R99Kgqp2qepL2c/PARjbZryo953Rr4IoRrHPB3B83P87UVr7vSuAX4jILhHZUOzOTKBBVbuyn3cDaChmZybg3ca7kN62zXjJ3He5bH+eL75B9/uuV9X3ALgFwMezf66WJB19DVZKY6eT2sa7UCbYZvx3innf5br9eb6KUewnAIxfMW9B9raSoKonsh97AfwYpbcVdc/YDrrZj71F7s/vlNI23hNtM44SuO+Kuf15MYp9J4BlIrJYRMoBfBTAk0Xox+8RkXj2jROISBzAjSi9raifBHB39vO7ATxRxL68Rals4+3aZhxFvu+Kvv25qhb8H4B1GH1H/jCALxSjD45+LQHw2+y/V4rdNwDfxeifdUmMvrdxD4ALADwH4CCAZwHUl1DfHgWwF8AejBZWY5H6dj1G/0TfA2B39t+6Yt93Rr8Kcr/xclmiQPANOqJAsNiJAsFiJwoEi50oECx2okCw2IkCwWInCsT/An2vctPR9CHeAAAAAElFTkSuQmCC\n",
"text/plain": [
"<Figure size 432x288 with 1 Axes>"
]
},
"metadata": {
"needs_background": "light"
},
"output_type": "display_data"
}
],
"source": [
"# orig 5, guide 9\n",
"plt.imshow(x_test_adv[8,...].squeeze())"
]
},
{
"cell_type": "code",
"execution_count": 16,
"id": "8efcd647",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"<matplotlib.image.AxesImage at 0x7f33a4731460>"
]
},
"execution_count": 16,
"metadata": {},
"output_type": "execute_result"
},
{
"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": [
"# orig 0, guide 8\n",
"plt.imshow(x_test_adv[10,...].squeeze())"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"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": "3.9.7"
}
},
"nbformat": 4,
"nbformat_minor": 5
}