Spaces:
Sleeping
Sleeping
File size: 4,878 Bytes
b83b2c9 77b48d9 b83b2c9 77b48d9 b83b2c9 77b48d9 b83b2c9 77b48d9 b83b2c9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 | {
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "1d50af37",
"metadata": {
"vscode": {
"languageId": "plaintext"
}
},
"outputs": [],
"source": [
"from pathlib import pathlib\n",
"import matplotlib.pyplot as pyplot\n",
"import torch\n",
"from PIL import Image\n",
"from torchvision import transformers\n",
"from models import VGGEncoder"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ca8cc62f",
"metadata": {
"vscode": {
"languageId": "plaintext"
}
},
"outputs": [],
"source": [
"ROOT =Path.cwd()\n",
"EXAMPLES =ROOT/\"examples\"\n",
"VGG_WEIGHTS=ROOT/\"models\"/vgg_normalised.ptn\n",
"device = torch.device(\"cuda\" if torch.cuda.is available() else \"cpu\")\n",
"image_size = 512\n",
"\n",
"image_paths = {\n",
" \"Original Brad Pitt\": EXAMPLES / \"brad pitt.jpg\",\n",
" \"Stylized Brad Pitt A\": EXAMPLES / \"stylized_brad_pitt.jpg\", \n",
" \"Stylized Brad Pitt B\": EXAMPLES / \"stylized_brad pitt (1).jpg\"\n",
"}\n",
"\n",
"transform = transforms.Compose([\n",
" transforms.Resize( (image_size, image_size)),\n",
" transforms.ToTensor(),\n",
"])\n",
"\n",
"encoder = VGGEncoder(str(VGG_WEIGHTS), test=False).to(device).eval()\n",
"\n",
"print(\"Using device:\", device)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "70516ffa",
"metadata": {
"vscode": {
"languageId": "plaintext"
}
},
"outputs": [],
"source": [
"def load_image(path):\n",
" image = Image.open(path).convert(\"RGB\")\n",
" tensor = transform(image).unsqueeze(0).to(device)\n",
" return image, tensor\n",
"\n",
"\n",
"def show_images (paths_dict):\n",
" fig, axes = plt.subplots(1, len (paths_dict), figsize=(16, 5))\n",
" for ax, (title, path) in zip (axes, paths_dict.items()):\n",
" image = Image.open(path).convert(\"RGB\")\n",
" ax.imshow(image)\n",
" ax.set_title(title)\n",
" ax.axis(\"off\")\n",
" plt.tight_layout()\n",
"\n",
"\n",
"def extract_features (tensor):\n",
" with torch.no_grad():\n",
" h1, h2, h3, h4 encoder (tensor)\n",
" return {\n",
" \"relul 1\": h1,\n",
" \"relu2 1\": h2,\n",
" \"relu3 1\": h3,\n",
" \"relu4 1\": h4,\n",
" }\n",
"\n",
"\n",
"def activation_map (feature_tensor):\n",
" activation = feature_tensor[0].mean(dim=0).detach().cpu()\n",
" activation = activation - activation.min()\n",
" activation = activation / (activation.max() + 1e-8)\n",
" return activation"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cd571425",
"metadata": {
"vscode": {
"languageId": "plaintext"
}
},
"outputs": [],
"source": [
"image_tensors = {} \n",
"feature_bank = {}\n",
"\n",
"for name, path in image_paths.items():\n",
" _, tensor = load_image(path)\n",
" image_tensors[name] = tensor\n",
" feature_bank[name] = extract_features(tensor)\n",
"\n",
"for name, features in feature_bank.items():\n",
" shapes = {layer: tuple(feature.shape) for layer, feature in features.items()}\n",
" print (name)\n",
" print(shapes)\n",
" print()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "82e8746e",
"metadata": {
"vscode": {
"languageId": "plaintext"
}
},
"outputs": [],
"source": [
"layers_to_show = [\"relul_1\", \"relu2_1\", \"relu3_1\", \"relu4_1\"]\n",
"row_labels = [\"Input Image\"] + layers_to_show\n",
"num_rows = len(row_labels)\n",
"num_cols = len(image_paths)\n",
"\n",
"fig, axes = plt.subplots (num_rows, num_cols, figsize=(5* num_cols, 3.6 * num_rows))\n",
"\n",
"for col, (name, path) in enumerate(image_paths.items()):\n",
" image = Image.open (path).convert(\"RGB\")\n",
" axes[0, col].imshow(image)\n",
" axes[0, col].set_title(name, fontsize=13)\n",
" axes [0, col].axis(\"off\")\n",
"\n",
"for row, layer in enumerate (layers_to_show, start=1):\n",
" axes[row, col].imshow(activation_map(feature_bank[name][layer]))\n",
" axes[row, col].axis(\"off\")\n",
"\n",
"for row, label in enumerate(row_labels):\n",
" axes[row, 0].set_ylabel(label, fontsize=12, rotation=90, labelpad=18)\n",
" \n",
"plt.suptitle(\"VGG Content Features Across Original and Stylized Brad Pitt Images\", fontsize=16, y=1.02)\n",
"plt.tight_layout()"
]
}
],
"metadata": {
"language_info": {
"name": "python"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|