File size: 10,807 Bytes
d41b862 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 | {
"cells": [
{
"cell_type": "code",
"execution_count": 2,
"id": "71c25a25",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"cuda\n"
]
},
{
"ename": "NameError",
"evalue": "name 'exit' is not defined",
"output_type": "error",
"traceback": [
"\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)",
"Cell \u001b[0;32mIn[2], line 55\u001b[0m\n\u001b[1;32m 53\u001b[0m torch\u001b[38;5;241m.\u001b[39mcuda\u001b[38;5;241m.\u001b[39mmanual_seed(SEED)\n\u001b[1;32m 54\u001b[0m torch\u001b[38;5;241m.\u001b[39mbackends\u001b[38;5;241m.\u001b[39mcudnn\u001b[38;5;241m.\u001b[39mbenchmark \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mTrue\u001b[39;00m\n\u001b[0;32m---> 55\u001b[0m model, aug \u001b[38;5;241m=\u001b[39m \u001b[43mload_model\u001b[49m\u001b[43m(\u001b[49m\u001b[43mdiffusion_ver\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mVER\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mimage_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mSIZE\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mnum_timesteps\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mTIMESTEP\u001b[49m\u001b[43m,\u001b[49m\u001b[43mdecoder_only\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mFalse\u001b[39;49;00m\u001b[43m)\u001b[49m\n",
"File \u001b[0;32m~/code/DeCLIP/extractor_sd.py:186\u001b[0m, in \u001b[0;36mload_model\u001b[0;34m(config_path, seed, diffusion_ver, image_size, num_timesteps, block_indices, decoder_only, encoder_only, resblock_only)\u001b[0m\n\u001b[1;32m 184\u001b[0m dataset_cfg \u001b[38;5;241m=\u001b[39m cfg\u001b[38;5;241m.\u001b[39mdataloader\u001b[38;5;241m.\u001b[39mtest\n\u001b[1;32m 185\u001b[0m \u001b[38;5;28mprint\u001b[39m(cfg\u001b[38;5;241m.\u001b[39mtrain\u001b[38;5;241m.\u001b[39mdevice)\n\u001b[0;32m--> 186\u001b[0m \u001b[43mexit\u001b[49m()\n\u001b[1;32m 187\u001b[0m aug \u001b[38;5;241m=\u001b[39m instantiate(dataset_cfg\u001b[38;5;241m.\u001b[39mmapper)\u001b[38;5;241m.\u001b[39maugmentations\n\u001b[1;32m 188\u001b[0m model \u001b[38;5;241m=\u001b[39m instantiate_odise(cfg\u001b[38;5;241m.\u001b[39mmodel)\n",
"\u001b[0;31mNameError\u001b[0m: name 'exit' is not defined"
]
}
],
"source": [
"import os\n",
"from training.utils import build_vfm\n",
"os.environ['CUDA_VISIBLE_DEVICES'] = '3'\n",
"import torch\n",
"import os\n",
"import pandas as pd\n",
"import numpy as np\n",
"from PIL import Image\n",
"from tqdm import tqdm\n",
"import torch.nn.functional as F\n",
"from extractor_sd import load_model, process_features_and_mask, get_mask\n",
"from third_party.utils.utils_correspondence import co_pca, resize, find_nearest_patchs, find_nearest_patchs_replace\n",
"import matplotlib.pyplot as plt\n",
"import sys\n",
"from extractor_dino import ViTExtractor\n",
"from sklearn.decomposition import PCA as sklearnPCA\n",
"import math\n",
"from sklearn.cluster import KMeans\n",
"from scipy.optimize import linear_sum_assignment\n",
"from torchvision import transforms\n",
"\n",
"def preprocess_pil(pil_image):\n",
" prep = transforms.Compose([\n",
" transforms.ToTensor(),\n",
" transforms.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225))\n",
" ])\n",
" prep_img = prep(pil_image)[None, ...]\n",
" return prep_img\n",
"\n",
"MASK = True\n",
"VER = \"v1-5\"\n",
"PCA = False\n",
"CO_PCA = True\n",
"PCA_DIMS = [256, 256, 256]\n",
"SIZE =960\n",
"EDGE_PAD = False\n",
"\n",
"FUSE_DINO = 1\n",
"ONLY_DINO = 0\n",
"MODEL_SIZE = 'base'\n",
"DRAW_DENSE = 1\n",
"DRAW_SWAP = 1\n",
"TEXT_INPUT = False\n",
"SEED = 42\n",
"TIMESTEP = 100\n",
"\n",
"DIST = 'l2' if FUSE_DINO and not ONLY_DINO else 'cos'\n",
"if ONLY_DINO:\n",
" FUSE_DINO = True\n",
"\n",
"np.random.seed(SEED)\n",
"torch.manual_seed(SEED)\n",
"torch.cuda.manual_seed(SEED)\n",
"torch.backends.cudnn.benchmark = True\n",
"model, aug = load_model(diffusion_ver=VER, image_size=SIZE, num_timesteps=TIMESTEP,decoder_only=False)\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "75082563",
"metadata": {},
"outputs": [],
"source": [
"def compute_pair_feature(model, aug, save_path, files, category, mask=False, dist='cos', real_size=960):\n",
" img_size = 840\n",
" stride = 14 \n",
" device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
" extractor=build_vfm(\"dinov2-B\").to(device).eval().to(torch.float32)\n",
" patch_size = 14\n",
" num_patches = int(patch_size / stride * (img_size // patch_size - 1) + 1)\n",
" input_text = None\n",
"\n",
" # Load image 1\n",
" img1 = Image.open(files[0]).convert('RGB')\n",
" img1_input = resize(img1, real_size, resize=True, to_pil=True, edge=EDGE_PAD)\n",
" img1 = resize(img1, img_size, resize=True, to_pil=True, edge=EDGE_PAD)\n",
"\n",
" # Load image 2\n",
" img2 = Image.open(files[1]).convert('RGB')\n",
" img2_input = resize(img2, real_size, resize=True, to_pil=True, edge=EDGE_PAD)\n",
" img2 = resize(img2, img_size, resize=True, to_pil=True, edge=EDGE_PAD)\n",
"\n",
" with torch.no_grad():\n",
" features1 = process_features_and_mask(model, aug, img1_input, input_text=input_text, mask=False, raw=True)\n",
" features2 = process_features_and_mask(model, aug, img2_input, input_text=input_text, mask=False, raw=True)\n",
" processed_features1, processed_features2 = co_pca(features1, features2, PCA_DIMS)\n",
" img1_desc = processed_features1.reshape(1, 1, -1, num_patches**2).permute(0,1,3,2)\n",
" img2_desc = processed_features2.reshape(1, 1, -1, num_patches**2).permute(0,1,3,2)\n",
" img1_batch = preprocess_pil(img1).to(device)\n",
" img1_desc_dino = extractor.get_intermediate_layers(img1_batch)[0].unsqueeze(1)\n",
" img2_batch = preprocess_pil(img2).to(device)\n",
" img2_desc_dino = extractor.get_intermediate_layers(img2_batch)[0].unsqueeze(1)\n",
" img1_desc = img1_desc / img1_desc.norm(dim=-1, keepdim=True)\n",
" img2_desc = img2_desc / img2_desc.norm(dim=-1, keepdim=True)\n",
" img1_desc_dino = img1_desc_dino / img1_desc_dino.norm(dim=-1, keepdim=True)\n",
" img2_desc_dino = img2_desc_dino / img2_desc_dino.norm(dim=-1, keepdim=True)\n",
" img1_desc = torch.cat((img1_desc, img1_desc_dino), dim=-1)\n",
" img2_desc = torch.cat((img2_desc, img2_desc_dino), dim=-1)\n",
" if DRAW_DENSE:\n",
" mask1 = get_mask(model, aug, img1, category[0])\n",
" mask2 = get_mask(model, aug, img2, category[-1])\n",
" img1_desc_reshaped = img1_desc.permute(0,1,3,2).reshape(-1, img1_desc.shape[-1], num_patches, num_patches)\n",
" img2_desc_reshaped = img2_desc.permute(0,1,3,2).reshape(-1, img2_desc.shape[-1], num_patches, num_patches)\n",
" trg_dense_output, src_color_map = find_nearest_patchs(mask2, mask1, img2, img1, img2_desc_reshaped, img1_desc_reshaped, mask=mask)\n",
"\n",
" if not os.path.exists(f'{save_path}/{category[0]}'):\n",
" os.makedirs(f'{save_path}/{category[0]}')\n",
" fig_colormap, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 8))\n",
" ax1.axis('off')\n",
" ax2.axis('off')\n",
" ax1.imshow(src_color_map)\n",
" ax2.imshow(trg_dense_output)\n",
" fig_colormap.savefig(f'{save_path}/{category[0]}_colormap.png')\n",
" plt.close(fig_colormap)\n",
" \n",
" if DRAW_SWAP:\n",
" if not DRAW_DENSE:\n",
" mask1 = get_mask(model, aug, img1, category[0])\n",
" mask2 = get_mask(model, aug, img2, category[-1])\n",
"\n",
" if (ONLY_DINO or not FUSE_DINO) and not DRAW_DENSE:\n",
" img1_desc = img1_desc / img1_desc.norm(dim=-1, keepdim=True)\n",
" img2_desc = img2_desc / img2_desc.norm(dim=-1, keepdim=True)\n",
" \n",
" img1_desc_reshaped = img1_desc.permute(0,1,3,2).reshape(-1, img1_desc.shape[-1], num_patches, num_patches)\n",
" img2_desc_reshaped = img2_desc.permute(0,1,3,2).reshape(-1, img2_desc.shape[-1], num_patches, num_patches)\n",
" trg_dense_output, src_color_map = find_nearest_patchs_replace(mask2, mask1, img2, img1, img2_desc_reshaped, img1_desc_reshaped, mask=mask, resolution=156)\n",
" if not os.path.exists(f'{save_path}/{category[0]}'):\n",
" os.makedirs(f'{save_path}/{category[0]}')\n",
" fig_colormap, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 8))\n",
" ax1.axis('off')\n",
" ax2.axis('off')\n",
" ax1.imshow(src_color_map)\n",
" ax2.imshow(trg_dense_output)\n",
" fig_colormap.savefig(f'{save_path}/{category[0]}/swap.png')\n",
" plt.close(fig_colormap)\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e4d1bdc6",
"metadata": {},
"outputs": [],
"source": [
"def process_images(src_img_path,trg_img_path):\n",
"\n",
" categories = [['dog'], ['dog']]\n",
" files = [src_img_path, trg_img_path]\n",
" save_path = './my_results_vis' + f'/{trg_img_path.split(\"/\")[-1].split(\".\")[0]}_{src_img_path.split(\"/\")[-1].split(\".\")[0]}'\n",
" result = compute_pair_feature(model, aug, save_path, files, mask=MASK, category=categories, dist=DIST)\n",
" return result\n",
"\n",
"src_img_path = \"demo_images/dog.jpg\"\n",
"trg_img_path = \"demo_images/whitedog.jpg\"\n",
"\n",
"result = process_images(src_img_path, trg_img_path)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "declip",
"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": "3.9.0"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|