1<!doctype html><html lang=en-us><head><meta charset=utf-8><meta name=viewport content="width=device-width,initial-scale=1"><meta name=google-site-verification content="7MghMiEezWCdLZfbrgYR845yxCJj-g02bSAnoythP1A"><meta http-equiv=X-UA-Compatible content="IE=edge"><meta name=generator content="Source Themes Academic 4.5.0"><meta name=author content="Gordon Qian"><meta name=description content='<!DOCTYPE html> 2 3 4 5 6 7 8 9 10 11 12 13 Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors 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 Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors 54 55 ICLR 2024 56 57 58 59 60 61 62 Guocheng 63 Qian1,2 64 65 66 Jinjie 67 Mai1 68 69 70 Abdullah 71 Hamdi3 72 73 74 Jian 75 Ren2 76 77 78 79 80 Aliaksandr 81 Siarohin2 82 83 84 Bing 85 Li1 86 87 88 Hsin-Ying 89 Lee2 90 91 92 Ivan 93 Skorokhodov1,2 94 95 96 97 98 Peter 99 Wonka1 100 101 102 Sergey 103 Tulyakov2 104 105 106 Bernard 107 Ghanem1 108 109 110 111 112 1King Abdullah University of Science and 113 Technology (KAUST) 114 115 116 2Snap Inc. 117 118 119 3Visual Geometry Group, University of 120 Oxford 121 122 123 124 125 126 127 128 129 130 131 132 Arxiv 133 134 135 136 137 138 Code 139 140 141 142 143 144 145 146 147 148 149 150 151 152 Abstract 153 154 155 We present "Magic123", a two-stage coarse-to-fine solution for high-quality, textured 3D meshes 156 generation from a single unposed image in the wild using both 2D and 3D priors. In the first stage, 157 we optimize a neural radiance field to produce a coarse geometry. In the second stage, we adopt a 158 memory-efficient differentiable mesh representation to yield a high-resolution mesh with a visually 159 appealing texture. In both stages, the 3D content is learned through reference view supervision and 160 novel views guided by both 2D and 3D diffusion priors. We introduce a single tradeoff parameter 161 between the 2D and 3D priors to control exploration (more imaginative) and exploitation (more 162 precise) of the generated geometry. Additionally, We employ textual inversion and monocular depth 163 regularization to encourage consistent appearances across views and to prevent degenerate solutions, 164 respectively. 165 Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as validated 166 through extensive experiments on synthetic benchmarks and diverse real-world images. 167 168 '><link rel=alternate hreflang=en-us href=https://guochengqian.github.io/project/magic123/><meta name=theme-color content="#b76531"><link rel=stylesheet href=https://cdnjs.cloudflare.com/ajax/libs/academicons/1.8.6/css/academicons.min.css integrity="sha256-uFVgMKfistnJAfoCUQigIl+JfUaP47GrRKjf6CTPVmw=" crossorigin=anonymous><link rel=stylesheet href=https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.5.1/css/all.min.css integrity="sha512-DTOQO9RWCH3ppGqcWaEA1BIZOC6xxalwEsw9c2QQeAIftl+Vegovlnee1c9QX4TctnWMn13TZye+giMm8e2LwA==" crossorigin=anonymous><link rel=stylesheet href=https://cdnjs.cloudflare.com/ajax/libs/fancybox/3.2.5/jquery.fancybox.min.css integrity="sha256-ygkqlh3CYSUri3LhQxzdcm0n1EQvH2Y+U5S2idbLtxs=" crossorigin=anonymous><link rel=stylesheet href=https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.15.6/styles/github.min.css crossorigin=anonymous title=hl-light><link rel=stylesheet href=https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.15.6/styles/dracula.min.css crossorigin=anonymous title=hl-dark disabled><link rel=stylesheet href="https://fonts.googleapis.com/css?family=Newsreader:500,600,700%7CManrope:400,500,600,700,800%7CIBM+Plex+Mono:400,500,600&display=swap"><link rel=stylesheet href=/css/academic.min.c06754a7cf79d154c937a526cfcdc457.css><link rel=manifest href=/manifest.webmanifest><link rel=icon href="/img/icon.ico?v=202607072"><link rel=apple-touch-icon href="/img/icon.ico?v=202607072"><link rel=canonical href=https://guochengqian.github.io/project/magic123/><meta property="twitter:card" content="summary_large_image"><meta property="og:site_name" content="Gordon Qian"><meta property="og:url" content="https://guochengqian.github.io/project/magic123/"><meta property="og:title" content=" | Gordon Qian"><meta property="og:description" content='<!DOCTYPE html> 169 170 171 172 173 174 175 176 177 178 179 180 Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors 221 222 ICLR 2024 223 224 225 226 227 228 229 Guocheng 230 Qian1,2 231 232 233 Jinjie 234 Mai1 235 236 237 Abdullah 238 Hamdi3 239 240 241 Jian 242 Ren2 243 244 245 246 247 Aliaksandr 248 Siarohin2 249 250 251 Bing 252 Li1 253 254 255 Hsin-Ying 256 Lee2 257 258 259 Ivan 260 Skorokhodov1,2 261 262 263 264 265 Peter 266 Wonka1 267 268 269 Sergey 270 Tulyakov2 271 272 273 Bernard 274 Ghanem1 275 276 277 278 279 1King Abdullah University of Science and 280 Technology (KAUST) 281 282 283 2Snap Inc. 284 285 286 3Visual Geometry Group, University of 287 Oxford 288 289 290 291 292 293 294 295 296 297 298 299 Arxiv 300 301 302 303 304 305 Code 306 307 308 309 310 311 312 313 314 315 316 317 318 319 Abstract 320 321 322 We present "Magic123", a two-stage coarse-to-fine solution for high-quality, textured 3D meshes 323 generation from a single unposed image in the wild using both 2D and 3D priors. In the first stage, 324 we optimize a neural radiance field to produce a coarse geometry. In the second stage, we adopt a 325 memory-efficient differentiable mesh representation to yield a high-resolution mesh with a visually 326 appealing texture. In both stages, the 3D content is learned through reference view supervision and 327 novel views guided by both 2D and 3D diffusion priors. We introduce a single tradeoff parameter 328 between the 2D and 3D priors to control exploration (more imaginative) and exploitation (more 329 precise) of the generated geometry. Additionally, We employ textual inversion and monocular depth 330 regularization to encourage consistent appearances across views and to prevent degenerate solutions, 331 respectively. 332 Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as validated 333 through extensive experiments on synthetic benchmarks and diverse real-world images. 334 335 '><meta property="og:image" content="https://guochengqian.github.io/img/avatar.jpg"><meta property="twitter:image" content="https://guochengqian.github.io/img/avatar.jpg"><meta property="og:locale" content="en-us"><meta property="article:modified_time" content="2026-06-02T14:28:55-07:00">
335<script type=application/ld+json>{"@context":"https://schema.org","@type":"Article","mainEntityOfPage":{"@type":"WebPage","@id":"https://guochengqian.github.io/project/magic123/"},"headline":"","datePublished":"0001-01-01T00:00:00Z","dateModified":"2026-06-02T14:28:55-07:00","author":{"@type":"Person","name":"Gordon Qian"},"publisher":{"@type":"Organization","name":"Gordon Qian","logo":{"@type":"ImageObject","url":"https://guochengqian.github.io/img/icon.ico"}},"description":"\u003c!DOCTYPE html\u003e\n\u003chtml\u003e\n\n\u003chead\u003e\n \u003cscript id=\"MathJax-script\" async src=\"https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js\"\u003e\u003c/script\u003e\n\n \u003c!-- Google tag (gtag.js) --\u003e\n \u003cscript async src=\"https://www.googletagmanager.com/gtag/js?id=G-XB3PR2Y1TQ\"\u003e\u003c/script\u003e\n \u003cscript\u003e\n window.dataLayer = window.dataLayer || [];\n\n function gtag() {\n dataLayer.push(arguments);\n }\n gtag('js', new Date());\n\n gtag('config', 'G-XB3PR2Y1TQ');\n \u003c/script\u003e\n\n \u003cmeta charset=\"utf-8\"\u003e\n \u003cmeta name=\"viewport\" content=\"width=device-width, initial-scale=1.0, shrink-to-fit=no\"\u003e\n \u003ctitle\u003eMagic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors\u003c/title\u003e\n \u003clink rel=\"stylesheet\" href=\"https://cdnjs.cloudflare.com/ajax/libs/twitter-bootstrap/4.5.0/css/bootstrap.min.css\"\u003e\n \u003clink href='https://fonts.googleapis.com/css?family=Source+Sans+Pro:300,400,500,600' rel='stylesheet'\n type='text/css'\u003e\n \u003clink rel=\"stylesheet\" href=\"assets/css/Highlight-Clean.css\"\u003e\n \u003clink rel=\"stylesheet\" href=\"assets/css/styles.css\"\u003e\n\n \u003clink rel=\"apple-touch-icon\" sizes=\"180x180\" href=\"/dragon.png\"\u003e\n \u003clink rel=\"icon\" type=\"image/png\" sizes=\"32x32\" href=\"/dragon-32x32.png\"\u003e\n \u003clink rel=\"icon\" type=\"image/png\" sizes=\"16x16\" href=\"/dragon-16x16.png\"\u003e\n\n\n \u003clink rel=\"manifest\" href=\"/site.webmanifest\"\u003e\n\n \u003cmeta property=\"og:site_name\"\n content=\"Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors\" /\u003e\n \u003cmeta property=\"og:type\" content=\"video.other\" /\u003e\n \u003cmeta property=\"og:title\"\n content=\"Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors\" /\u003e\n \u003cmeta property=\"og:description\"\n content=\"Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors, 2023.\" /\u003e\n \u003cmeta property=\"og:url\" content=\"https://guochengqian.github.io/project/magic123/\" /\u003e\n \u003cmeta property=\"og:image\" content=\"https://guochengqian.github.io/project/magic123/assets/images/pull.jpg\" /\u003e\n \u003c!-- todo change the sample image here --\u003e\n\n \u003cmeta property=\"article:publisher\" content=\"https://guochengqian.github.io/project/magic123/\" /\u003e\n \u003cmeta name=\"twitter:card\" content=\"summary_large_image\" /\u003e\n \u003cmeta name=\"twitter:title\"\n content=\"Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors\" /\u003e\n \u003cmeta name=\"twitter:description\"\n content=\"We combine 2D and 3D priors to generate photo-realistic 3D objects from a single image.\" /\u003e\n \u003cmeta name=\"twitter:url\" content=\"https://guochengqian.github.io/project/magic123/\" /\u003e\n \u003cmeta name=\"twitter:image\" content=\"https://guochengqian.github.io/project/magic123/assets/images/pull.jpg\" /\u003e\n \u003c!-- \u003cmeta name=\"twitter:site\" content=\"\" /\u003e --\u003e\n\n \u003cscript src=\"assets/js/video_comparison.js\"\u003e\u003c/script\u003e\n \u003cscript type=\"module\" src=\"https://unpkg.com/@google/[email protected]/dist/model-viewer.min.js\"\u003e\u003c/script\u003e\n\u003c/head\u003e\n\n\u003cbody\u003e\n\n \u003cstyle\u003e\n .project-links {\n margin-top: 0.25rem;\n margin-bottom: 2rem;\n }\n .project-link-item {\n margin-bottom: 0.25rem;\n }\n .project-link-icon {\n height: 64px;\n width: auto;\n display: inline-block;\n }\n .hero-gif {\n width: 100%;\n height: auto;\n display: block;\n margin: 14px auto 0;\n }\n .example-generated-objects,\n .example-generated-objects h2,\n .example-generated-objects p {\n text-align: center;\n }\n .example-generated-objects .examples-grid .row {\n justify-content: center;\n }\n .example-generated-objects .examples-grid .col-sm-1 {\n justify-content: center !important;\n text-align: center;\n }\n .example-generated-objects .examples-grid img,\n .example-generated-objects .examples-grid video,\n .tradeoff-media img,\n .tradeoff-media video,\n .quantitative-results-media img {\n width: 100%;\n max-width: 100%;\n height: auto;\n display: block;\n margin-left: auto;\n margin-right: auto;\n }\n .tradeoff-media {\n width: 100%;\n }\n .tradeoff-media img,\n .tradeoff-media video {\n margin-bottom: 16px;\n }\n .tradeoff-media \u003e img,\n .tradeoff-media \u003e video {\n width: 100% !important;\n max-width: none !important;\n min-width: 100%;\n height: auto !important;\n max-height: none !important;\n display: block;\n box-sizing: border-box;\n object-fit: contain;\n }\n .footer .footer-icons {\n display: flex;\n justify-content: center;\n align-items: center;\n gap: 16px;\n text-align: center;\n margin-bottom: 10px;\n }\n .footer .icon-link {\n display: inline-flex;\n align-items: center;\n justify-content: center;\n min-width: 36px;\n min-height: 36px;\n font-size: 18px;\n text-decoration: none;\n }\n .footer .content {\n text-align: center;\n }\n @media (max-width: 575px) {\n .project-link-icon {\n height: 52px;\n }\n .project-links {\n margin-bottom: 1.5rem;\n }\n }\n \u003c/style\u003e\n\n \u003cdiv class=\"highlight-clean\" style=\"padding-bottom: 10px;\"\u003e\n \u003cdiv class=\"container\" style=\"max-width: 100%;\"\u003e\n \u003cbr\u003e\n \u003ch1 class=\"text-center\"\u003e\u003cb\u003eMagic123\u003c/b\u003e: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors\u003c/h1\u003e\n \u003cbr\u003e\n \u003ch2 class=\"text-center\"\u003e\u003cb\u003eICLR 2024\u003c/b\u003e\u003c/h2\u003e\n \u003cbr\u003e\n \u003c/div\u003e\n\n \u003cdiv class=\"container author-container\"\u003e\n \u003cdiv class=\"row authors\"\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"https://guochengqian.github.io/\"\u003eGuocheng\n Qian\u003csup\u003e1,2\u003c/sup\u003e \u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"https://cemse.kaust.edu.sa/people/person/jinjie-mai\"\u003eJinjie\n Mai\u003csup\u003e1\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"https://abdullahamdi.com/\"\u003eAbdullah\n Hamdi\u003csup\u003e3\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"https://alanspike.github.io/\"\u003eJian\n Ren\u003csup\u003e2\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv class=\"row authors\"\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca class=\"text-center\"\n href=\"https://aliaksandrsiarohin.github.io/aliaksandr-siarohin-website/\"\u003eAliaksandr\n Siarohin\u003csup\u003e2\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca class=\"text-center\" href=\"https://cemse.kaust.edu.sa/people/person/bing-li\"\u003eBing\n Li\u003csup\u003e1\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"http://hsinyinglee.com/\"\u003eHsin-Ying\n Lee\u003csup\u003e2\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"https://universome.github.io/\"\u003eIvan\n Skorokhodov\u003csup\u003e1,2\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv class=\"row authors\"\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"https://peterwonka.net/\"\u003ePeter\n Wonka\u003csup\u003e1\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"http://www.stulyakov.com/\"\u003eSergey\n Tulyakov\u003csup\u003e2\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch5\u003e\u003ca href=\"https://www.bernardghanem.com/\"\u003eBernard\n Ghanem\u003csup\u003e1\u003c/sup\u003e\u003c/a\u003e\u003c/h5\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv class=\"row affiliations\"\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch6 class=\"text-center\"\u003e\u003ca class=\"text-center\"\u003e\u003csup\u003e1\u003c/sup\u003eKing Abdullah University of Science and\n Technology (KAUST)\u003c/a\u003e\u003c/h6\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch6 class=\"text-center\"\u003e\u003ca class=\"text-center\"\u003e\u003csup\u003e2\u003c/sup\u003eSnap Inc.\u003c/a\u003e\u003c/h6\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col\"\u003e\n \u003ch6 class=\"text-center\"\u003e\u003ca class=\"text-center\"\u003e\u003csup\u003e3\u003c/sup\u003eVisual Geometry Group, University of\n Oxford\u003c/a\u003e\u003c/h6\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n\n \u003cdiv class=\"container align-items-center project-links\"\u003e\n \u003cdiv class=\"row text-center mx-auto justify-content-center\"\u003e\n \u003cdiv class=\"col-sm-4 col-md-3 project-link-item\"\u003e\n \u003ca href=\"https://arxiv.org/abs/2306.17843\" target=\"_blank\" class=\"link-block w-inline-block\"\u003e\n \u003cimg src=\"assets/images/arxiv_paper.png\" alt=\"paper\" class=\"img-fluid project-link-icon\" /\u003e\u003c/a\u003e\n \u003cbr\u003e\n \u003cdiv class=\"text-block-2\"\u003e\u003cstrong class=\"bold-text-nerf_v2\"\u003eArxiv\u003c/strong\u003e\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv class=\"col-sm-4 col-md-3 project-link-item\"\u003e\n \u003ca href=\"https://github.com/guochengqian/Magic123\" target=\"_blank\" class=\"link-block w-inline-block\"\u003e\n \u003cimg src=\"assets/images/github.png\" alt=\"github\" class=\"img-fluid project-link-icon\" /\u003e\u003c/a\u003e\n \u003cbr\u003e\n \u003cdiv class=\"text-block-2\"\u003e\u003cstrong class=\"bold-text-nerf_v2\"\u003eCode\u003c/strong\u003e\u003c/div\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n \u003c/div\u003e\n\n \u003c/div\u003e\n \u003c!-- \u003cdiv class=\"banner\"\u003e\n \u003cvideo class=\"video lazy\"\n poster=\"https://dreamfusion-cdn.ajayj.com/sept28/banner_1x6_customhue_A.jpg\"\n autoplay loop playsinline muted\u003e\n \u003csource data-src=\"https://dreamfusion-cdn.ajayj.com/sept28/banner_1x6_customhue_A.mp4\" type=\"video/mp4\" style=\"width: 100%; height: auto;\"\u003e\u003c/source\u003e\n \u003c/video\u003e\n \u003c/div\u003e --\u003e\n \u003cvideo src=\"assets/videos/pull.mp4\" autoplay loop playsinline muted class=\"hero-gif\" type=\"video/mp4\"\u003e\u003c/video\u003e\n\n\n \u003chr class=\"divider\" /\u003e\n \u003cdiv class=\"container\" style=\"max-width: col-md-8;\"\u003e\n \u003cdiv class=\"row\"\u003e\n \u003cdiv class=\"col-sm-12\"\u003e\n \u003ch2\u003eAbstract\u003c/h2\u003e\n \u003cp\u003e\n \u003c!-- \u003cstrong\u003e --\u003e\n We present \"Magic123\", a two-stage coarse-to-fine solution for high-quality, textured 3D meshes\n generation from a single unposed image in the wild using both 2D and 3D priors. In the first stage,\n we optimize a neural radiance field to produce a coarse geometry. In the second stage, we adopt a\n memory-efficient differentiable mesh representation to yield a high-resolution mesh with a visually\n appealing texture. In both stages, the 3D content is learned through reference view supervision and\n novel views guided by both 2D and 3D diffusion priors. We introduce a single tradeoff parameter\n between the 2D and 3D priors to control exploration (more imaginative) and exploitation (more\n precise) of the generated geometry. Additionally, We employ textual inversion and monocular depth\n regularization to encourage consistent appearances across views and to prevent degenerate solutions,\n respectively.\n Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as validated\n through extensive experiments on synthetic benchmarks and diverse real-world images.\n \u003c!-- \u003c/strong\u003e --\u003e\n \u003c/p\u003e"}</script>
335<title>| Gordon Qian</title></head><body id=top data-spy=scroll data-offset=70 data-target=#TableOfContents class=theme-claude><aside class=search-results id=search><div class=container><section class=search-header><div class="row no-gutters justify-content-between mb-3"><div class=col-6><h1>Search</h1></div><div class="col-6 col-search-close"><a class=js-search href=#><i class="fas fa-times-circle text-muted" aria-hidden=true></i></a></div></div><div id=search-box><input name=q id=search-query placeholder=Search... autocapitalize=off autocomplete=off autocorrect=off spellcheck=false type=search></div></section><section class=section-search-results><div id=search-hits></div></section></div></aside><nav class="navbar navbar-light fixed-top navbar-expand-lg py-0 compensate-for-scrollbar" id=navbar-main><div class=container><a class=navbar-brand href=/>Gordon Qian</a> 336<button type=button class=navbar-toggler data-toggle=collapse data-target=#navbar aria-controls=navbar aria-expanded=false aria-label="Toggle navigation"> 337<span><i class="fas fa-bars"></i></span></button><div class="collapse navbar-collapse" id=navbar><ul class="navbar-nav mr-auto"><li class=nav-item><a class=nav-link href=/#about><span>Home</span></a></li><li class=nav-item><a class=nav-link href=/#news><span>News</span></a></li><li class=nav-item><a class=nav-link href=/#focused-areas><span>Focus Areas</span></a></li><li class=nav-item><a class=nav-link href=/#featured><span>Publications</span></a></li><li class=nav-item><a class=nav-link href=/#experience><span>Experience</span></a></li></ul><ul class="navbar-nav ml-auto"><li class=nav-item><a class=nav-link href=/blog/><span>Blog</span></a></li><li class=nav-item><a class="nav-link js-search" href=#><i class="fas fa-search" aria-hidden=true></i></a></li><li class=nav-item><a class="nav-link js-dark-toggle" href=#><i class="fas fa-moon" aria-hidden=true></i></a></li></ul></div></div></nav><article class="article article-project"><div class="article-container pt-3"><h1></h1><div class=article-metadata><span class=article-date>Last updated on 338Jun 2, 2026</span><div class=share-box aria-hidden=true><ul class=share><li><a href="https://twitter.com/intent/tweet?url=https://guochengqian.github.io/project/magic123/&text=" target=_blank rel=noopener class=share-btn-twitter><i class="fab fa-twitter"></i></a></li><li><a href="https://www.facebook.com/sharer.php?u=https://guochengqian.github.io/project/magic123/&t=" target=_blank rel=noopener class=share-btn-facebook><i class="fab fa-facebook-f"></i></a></li><li><a href="mailto:?subject=&body=https://guochengqian.github.io/project/magic123/" target=_blank rel=noopener class=share-btn-email><i class="fas fa-envelope"></i></a></li><li><a href="https://www.linkedin.com/shareArticle?url=https://guochengqian.github.io/project/magic123/&
338title=" target=_blank rel=noopener class=share-btn-linkedin><i class="fab fa-linkedin-in"></i></a></li><li><a href="https://web.whatsapp.com/send?text=%20https://guochengqian.github.io/project/magic123/" target=_blank rel=noopener class=share-btn-whatsapp><i class="fab fa-whatsapp"></i></a></li><li><a href="https://service.weibo.com/share/share.php?url=https://guochengqian.github.io/project/magic123/&title=" target=_blank rel=noopener class=share-btn-weibo><i class="fab fa-weibo"></i></a></li></ul></div></div></div><div class=article-container><div class=article-style><!doctype html><html><head>
338<script id=MathJax-script async src=https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js></script>
vendor: 63 bytes, line 338
338<script async src="https://www.googletagmanager.com/gtag/js?id=
338G-XB3PR2Y1TQ
vendor: 11 bytes, line 338
338"></script>
338<script>window.dataLayer=window.dataLayer||[];function gtag(){dataLayer.push(arguments)}gtag("js",new Date),gtag("config","G-XB3PR2Y1TQ")</script>
338<meta charset=utf-8><meta name=viewport content="width=device-width,initial-scale=1,shrink-to-fit=no"><title>Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors</title><link rel=stylesheet href=https://cdnjs.cloudflare.com/ajax/libs/twitter-bootstrap/4.5.0/css/bootstrap.min.css><link href='https://fonts.googleapis.com/css?family=Source+Sans+Pro:300,400,500,600' rel=stylesheet type=text/css><link rel=stylesheet href=assets/css/Highlight-Clean.css><link rel=stylesheet href=assets/css/styles.css><link rel=apple-touch-icon sizes=180x180 href=/dragon.png><link rel=icon type=image/png sizes=32x32 href=/dragon-32x32.png><link rel=icon type=image/png sizes=16x16 href=/dragon-16x16.png><link rel=manifest href=/site.webmanifest><meta property="og:site_name" content="Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors"><meta property="og:type" content="video.other"><meta property="og:title" content="Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors"><meta property="og:description" content="Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors, 2023."><meta property="og:url" content="https://guochengqian.github.io/project/magic123/"><meta property="og:image" content="https://guochengqian.github.io/project/magic123/assets/images/pull.jpg"><meta property="article:publisher" content="https://guochengqian.github.io/project/magic123/"><meta name=twitter:card content="summary_large_image"><meta name=twitter:title content="Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors"><meta name=twitter:description content="We combine 2D and 3D priors to generate photo-realistic 3D objects from a single image."><meta name=twitter:url content="https://guochengqian.github.io/project/magic123/"><meta name=twitter:image content="https://guochengqian.github.io/project/magic123/assets/images/pull.jpg">
338<script src=assets/js/video_comparison.js></script>
338<script type=module src=https://unpkg.com/@google/[email protected]/dist/model-viewer.min.js></script>
338</head><body><style>.project-links{margin-top:.25rem;margin-bottom:2rem}.project-link-item{margin-bottom:.25rem}.project-link-icon{height:64px;width:auto;display:inline-block}.hero-gif{width:100%;height:auto;display:block;margin:14px auto 0}.example-generated-objects,.example-generated-objects h2,.example-generated-objects p{text-align:center}.example-generated-objects .examples-grid .row{justify-content:center}.example-generated-objects .examples-grid .col-sm-1{justify-content:center!important;text-align:center}.example-generated-objects .examples-grid img,.example-generated-objects .examples-grid video,.tradeoff-media img,.tradeoff-media video,.quantitative-results-media img{width:100%;max-width:100%;height:auto;display:block;margin-left:auto;margin-right:auto}.tradeoff-media{width:100%}.tradeoff-media img,.tradeoff-media video{margin-bottom:16px}.tradeoff-media>img,.tradeoff-media>video{width:100%!important;max-width:none!important;min-width:100%;height:auto!important;max-height:none!important;display:block;box-sizing:border-box;object-fit:contain}.footer .footer-icons{display:flex;justify-content:center;align-items:center;gap:16px;text-align:center;margin-bottom:10px}.footer .icon-link{display:inline-flex;align-items:center;justify-content:center;min-width:36px;min-height:36px;font-size:18px;text-decoration:none}.footer .content{text-align:center}@media(max-width:575px){.project-link-icon{height:52px}.project-links{margin-bottom:1.5rem}}</style><div class=highlight-clean style=padding-bottom:10px><div class=container style=max-width:100%><br><h1 class=text-center><b>Magic123</b>: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors</h1><br><h2 class=text-center><b>ICLR 2024</b></h2><br></div><div class="container author-container"><div class="row authors"><div class=col><h5><a href=https://guochengqian.github.io/>Guocheng 339Qian<sup>1,2</sup></a></h5></div><div class=col><h5><a href=https://cemse.kaust.edu.sa/people/person/jinjie-mai>Jinjie 340Mai<sup>1</sup></a></h5></div><div class=col><h5><a href=https://abdullahamdi.com/>Abdullah 341Hamdi<sup>3</sup></a></h5></div><div class=col><h5><a href=https://alanspike.github.io/>Jian 342Ren<sup>2</sup></a></h5></div></div><div class="row authors"><div class=col><h5><a class=text-center href=https://aliaksandrsiarohin.github.io/aliaksandr-siarohin-website/>Aliaksandr 343Siarohin<sup>2</sup></a></h5></div><div class=col><h5><a class=text-center href=https://cemse.kaust.edu.sa/people/person/bing-li>Bing 344Li<sup>1</sup></a></h5></div><div class=col><h5><a href=http://hsinyinglee.com/>Hsin-Ying 345Lee<sup>2</sup></a></h5></div><div class=col><h5><a href=https://universome.github.io/>Ivan 346Skorokhodov<sup>1,2</sup></a></h5></div></div><div class="row authors"><div class=col><h5><a href=https://peterwonka.net/>Peter 347Wonka<sup>1</sup></a></h5></div><div class=col><h5><a href=http://www.stulyakov.com/>Sergey 348Tulyakov<sup>2</sup></a></h5></div><div class=col><h5><a href=https://www.bernardghanem.com/>Bernard 349Ghanem<sup>1</sup></a></h5></div></div><div class="row affiliations"><div class=col><h6 class=text-center><a class=text-center><sup>1</sup>King Abdullah University of Science and 350Technology (KAUST)</a></h6></div><div class=col><h6 class=text-center><a class=text-center><sup>2</sup>Snap Inc.</a></h6></div><div class=col><h6 class=text-center><a class=text-center><sup>3</sup>Visual Geometry Group, University of 351Oxford</a></h6></div></div></div></div><div class="container align-items-center project-links"><div class="row text-center mx-auto justify-content-center"><div class="col-sm-4 col-md-3 project-link-item"><a href=https://arxiv.org/abs/2306.17843 target=_blank class="link-block w-inline-block"><img src=assets/images/arxiv_paper.png alt=paper class="img-fluid project-link-icon"></a><br><div class=text-block-2><strong class=bold-text-nerf_v2>Arxiv</strong></div></div><div class="col-sm-4 col-md-3 project-link-item"><a href=https://github.com/guochengqian/Magic123 target=_blank class="link-block w-inline-block"><img src=assets/images/github.png alt=github class="img-fluid project-link-icon"></a><br><div class=text-block-2><strong class=bold-text-nerf_v2>Code</strong></div></div></div></div></div><video src=assets/videos/pull.mp4 autoplay loop playsinline muted class=hero-gif type=video/mp4></video><hr class=divider><div class=container style=max-width:col-md-8><div class=row><div class=col-sm-12><h2>Abstract</h2><p>We present "Magic123", a two-stage coarse-to-fine solution for high-quality, textured 3D meshes 352generation from a single unposed image in the wild using both 2D and 3D priors. In the first stage, 353we optimize a neural radiance field to produce a coarse geometry. In the second stage, we adopt a 354memory-efficient differentiable mesh representation to yield a high-resolution mesh with a visually 355appealing texture. In both stages, the 3D content is learned through reference view supervision and 356novel views guided by both 2D and 3D diffusion priors. We introduce a single tradeoff parameter 357between the 2D and 3D priors to control exploration (more imaginative) and exploitation (more 358precise) of the generated geometry. Additionally, We employ textual inversion and monocular depth 359regularization to encourage consistent appearances across views and to prevent degenerate solutions, 360respectively. 361Magic123 demonstrates a significant improvement over previous image-to-3D techniques, as validated 362through extensive experiments on synthetic benchmarks and diverse real-world images.</p></div></div></div><div class=container style=max-width:col-md-8><div class="row captioned_videos"><div class=col-sm-12><img src=assets/images/magic123_pipeline.png alt="Image description" class=img-fluid><h6 class=caption style=text-align:left><b>Magic123 pipeline.</b> Magic123 is a two-stage 363coarse-to-fine framework for high-quality 3643D generation from a reference image. Magic123 is guided by the reference image, constrained by the 365monocular depth estimation from the image, and driven by a joint 2D and 3D diffusion prior to dream 366up novel views. At the coarse stage, we optimize an Instant-NGP neural radiance field (NeRF) to 367reconstruct a coarse geometry. At the fine stage, we initialize a DMTet mesh from the NeRF output 368and optimize a high-resolution mesh and texture directly. Textural inversion is used in both stages 369to generate object-preserving geometry and view-consistent textures.</h6></div></div></div><hr class=divider><div class="container example-generated-objects" style=max-width:col-md-8><div class="row text-center"><div class=col-sm-12><h2>Example generated objects</h2><p>Magic123 generates photo-realistic 3D objects from a single unposed image.</p></div></div><div class="container align-items-center examples-grid"><div class=row><div class="col-sm-1 d-flex align-items-center justify-content-start p-0"><div class=method-title>Input</div></div><div class=col-sm-2><img src=assets/videos/reference/bear.png alt="Bear Image" class=img-fluid></div><div class=col-sm-2><img src=assets/videos/reference/dragon.png alt="Dragon Image" class=img-fluid></div><div class=col-sm-2><img src=assets/videos/reference/horse.png alt="Horse Image" class=img-fluid></div><div class=col-sm-2><img src=assets/videos/reference/teapot.png alt="Sparrow Image" class=img-fluid></div></div><div class=row><div class="col-sm-1 d-flex align-items-center justify-content-start p-0"><div class=method-title><a href=https://github.com/openai/shap-e>Shap-E</a></div></div><div class=col-sm-2><video src=assets/videos/shape_e/bear.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/shape_e/dragon.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/shape_e/horse.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/shape_e/teapot.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div></div><div class=row><div class="col-sm-1 d-flex align-items-center justify-content-start p-0"><div class=method-title><a href=https://ku-cvlab.github.io/3DFuse/>3D Fuse</a></div></div><div class=col-sm-2><video src=assets/videos/3d_fuse/bear.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/3d_fuse/dragon.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/3d_fuse/horse.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/3d_fuse/teapot.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div></div><div class=row><div class="col-sm-1 d-flex align-items-center justify-content-start p-0"><div class=method-title>
369<a href=https://vita-group.github.io/NeuralLift-360/>Neural Lift</a></div></div><div class=col-sm-2><video src=assets/videos/neural_lift/bear.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/neural_lift/dragon.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/neural_lift/horse.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/neural_lift/teapot.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div></div><div class=row><div class="col-sm-1 d-flex align-items-center justify-content-start p-0"><div class=method-title><a href=https://lukemelas.github.io/realfusion/>Real Fusion</a></div></div><div class=col-sm-2><video src=assets/videos/realfusion/bear.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/realfusion/dragon.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/realfusion/horse.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/realfusion/teapot.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div></div><div class=row><div class="col-sm-1 d-flex align-items-center justify-content-start p-0"><div class=method-title><b>Magic123</b><br><b>(Ours)</b></div></div><div class=col-sm-2><video src=assets/videos/magic123/bear.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/magic123/dragon.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/magic123/horse.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div><div class=col-sm-2><video src=assets/videos/magic123/teapot.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div></div></div></div><hr class=divider><div class=container style=max-width:col-md-8><div class=row><div class=col-sm-12><h2>Tradeoff between 2D and 3D priors in Magic123</h2><p>We compare single image reconstructions for three cases: a cactus (common object), two stacked 370donuts (less common object), and a dragon statue (uncommon object). 371Magic123 with only 2D prior (on the right) favors geometry exploration, generating 3D content 372with 373more imagination but potentially lacking 3D consistency. Magic123 with only 3D prior prioritizes 374geometry exploitation, resulting in precise yet potentially simplified geometry with reduced 375details. Magic123 thus proposes to use both 2D and 3D prior and introduces a tradeoff parameter 376\( \lambda_{2D/3D} \) to control the geometry exploration and exploitation. We provide a 377balanced point 378\( \lambda_{2D/3D}=1 \) , with which Magic123 consistently offers 379identity-preserving 3D content with fine-grained geometry and visually appealing texture.</p></div></div><div class="row justify-content-center" style=padding:0;margin:0><div class="col-sm-8 text-center tradeoff-media"><img src=assets/images/priors.jpeg alt="Image description" class=img-fluid><div style=height:30px></div><img src=assets/images/2d_3d.png alt="Image description" class=img-fluid> 380<video src=assets/videos/human_examples.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video> 381<video src=assets/videos/lambda_2d.mp4 autoplay loop playsinline muted type=video/mp4 style=width:100%;height:auto></video></div></div></div></div><hr class=divider><div class=container style=max-width:col-md-8><div class=row><div class=col-sm-12><h2>Effects of the coarse-to-fine pipeline</h2><p>Qualitative comparisons of the coarse and fine stages between Magic123 with only 2D prior \( 382\lambda_{2D}=1,\lambda_{3D}=0 \), 383Magic123 with only 3D prior \( \lambda_{2D}=0,\lambda_{3D}=40 \), and Magic123 \( 384\lambda_{2D}=1,\lambda_{3D}=40 \).</p></div></div><div class="row justify-content-center" style=padding:0;margin:0><div class="col-sm-12 text-center"><img src=assets/images/coarse_2_fine.png alt="Image description" class=img-fluid><div style=height:10px></div></div></div></div><hr class=divider><div class=container style=max-width:col-md-8><div class=row><div class=col-sm-12><h2>Quantitative Results</h2><p>To demonstrate the effectiveness of the proposed Magic123, we evaluate its performance on the 385NeRF4 and RealFusion15 datasets. We conduct a comprehensive and quantitative comparison with 386multiple baselines for both datasets, as shown in Table I. Notably, our method achieves Top-1 387performance across all the metrics when compared to previous SOTA approaches. This remarkable 388performance demonstrates the superiority of Magic123 and its ability to generate high-quality 3D 389representations.</p></div></div><div class="row justify-content-center" style=padding:0;margin:0><div class="col-sm-12 text-center quantitative-results-media"><img src=assets/images/results.jpeg alt="Image description" class=img-fluid></div></div></div><hr class=divider><footer class=footer><div class=container><div class="content has-text-centered footer-icons"><a class=icon-link href=https://arxiv.org/abs/2306.17843 aria-label="Magic123 paper"><i class="fa-solid fa-file-pdf" aria-hidden=true></i> 390</a><a class=icon-link href=https://github.com/guochengqian/Magic123 aria-label="Magic123 GitHub"><i class="fab fa-github" aria-hidden=true></i></a></div><div class="columns is-centered"><div class="column is-8"><div class=content><p>This website is licensed under a <a rel=license href=http://creativecommons.org/licenses/by-sa/4.0/>Creative 391Commons Attribution-ShareAlike 4.0 International License</a>.</p><p>We borrow the <a href=https://dreamfusion3d.github.io/>source code</a> for our 392website. 393We sincerely appreciate DreamFusion authors for their awesome templates.</p></div></div></div></div></footer>
393<script src=assets/js/yall.js></script>
393<script>yall({observeChanges:!0})</script>
393<script src=assets/js/scripts.js></script>
393<script src=https://cdnjs.cloudflare.com/ajax/libs/jquery/3.5.1/jquery.min.js></script>
393<script src=https://cdnjs.cloudflare.com/ajax/libs/twitter-bootstrap/4.5.0/js/bootstrap.bundle.min.js></script>
393</body></html></div></div></article>
393<script src=https://cdnjs.cloudflare.com/ajax/libs/jquery/3.4.1/jquery.min.js integrity="sha256-CSXorXvZcTkaix6Yvo6HppcZGetbYMGWSFlBw8HfCJo=" crossorigin=anonymous></script>
393<script src=https://cdnjs.cloudflare.com/ajax/libs/jquery.imagesloaded/4.1.4/imagesloaded.pkgd.min.js integrity="sha256-lqvxZrPLtfffUl2G/e7szqSvPBILGbwmsGE1MKlOi0Q=" crossorigin=anonymous></script>
393<script src=https://cdnjs.cloudflare.com/ajax/libs/jquery.isotope/3.0.6/isotope.pkgd.min.js integrity="sha256-CBrpuqrMhXwcLLUd5tvQ4euBHCdh7wGlDfNz8vbu/iI=" crossorigin=anonymous></script>
393<script src=https://cdnjs.cloudflare.com/ajax/libs/fancybox/3.2.5/jquery.fancybox.min.js integrity="sha256-X5PoE3KU5l+JcX+w09p/wHl9AzK333C4hJ2I9S5mD4M=" crossorigin=anonymous></script>
393<script src=https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.15.6/highlight.min.js integrity="sha256-aYTdUrn6Ow1DDgh5JTc3aDGnnju48y/1c8s1dgkYPQ8=" crossorigin=anonymous></script>
393<script src=https://cdnjs.cloudflare.com/ajax/libs/highlight.js/9.15.6/languages/r.min.js></script>
393<script>hljs.initHighlightingOnLoad()</script>
393<script>const search_config={indexURI:"/index.json",minLength:1,threshold:.3},i18n={no_results:"No results found",placeholder:"Search...",results:"results found"},content_type={post:"Posts",project:"Projects",publication:"Publications",talk:"Talks"}</script>
393<script id=search-hit-fuse-template type=text/x-template> 394 <div class="search-hit" id="summary-{{key}}"> 395 <div class="search-hit-content"> 396 <div class="search-hit-name"> 397 <a href="{{relpermalink}}">{{title}}</a> 398 <div class="article-metadata search-hit-type">{{type}}</div> 399 <p class="search-hit-description">{{snippet}}</p> 400 </div> 401 </div> 402 </div> 403 </script>
403<script src=https://cdnjs.cloudflare.com/ajax/libs/fuse.js/3.2.1/fuse.min.js integrity="sha256-VzgmKYmhsGNNN4Ph1kMW+BjoYJM2jV5i4IlFoeZA9XI=" crossorigin=anonymous></script>
403<script src=https://cdnjs.cloudflare.com/ajax/libs/mark.js/8.11.1/jquery.mark.min.js integrity="sha256-4HLtjeVgH0eIB3aZ9mLYF6E8oU5chNdjU6p6rrXpl9U=" crossorigin=anonymous></script>
403<script src=/js/academic.min.b89d1011e4e05f2d9ac6fafcb0f556d1.js></script>
403<script>(function(){function n(e){var t=e.previousElementSibling;return t&&t.classList.contains("video-placeholder")?t:null}function t(e){e.style.display="block";var t=n(e);t&&(t.style.display="none")}function e(e){e.muted=!0,e.defaultMuted=!0,e.playsInline=!0,e.setAttribute("muted",""),e.setAttribute("playsinline",""),e.setAttribute("webkit-playsinline",""),t(e);var n=e.play();n&&typeof n.catch=="function"&&n.catch(function(){})}function s(n){if(n.dataset.mobileAutoplayReady==="true")return;n.dataset.mobileAutoplayReady="true",n.addEventListener("loadeddata",function(){e(n)}),n.addEventListener("canplay",function(){e(n)}),n.addEventListener("play",function(){t(n)}),n.readyState>=2&&e(n)}function o(){var e=document.querySelectorAll(".publication-image-container .video-content");e.forEach(s)}document.addEventListener("DOMContentLoaded",function(){o(),document.addEventListener("visibilitychange",function(){if(document.hidden)return;document.querySelectorAll(".publication-image-container .video-content").forEach(e)}),["touchstart","pointerdown"].forEach(function(t){document.addEventListener(t,function n(){document.querySelectorAll(".publication-image-container .video-content").forEach(e),document.removeEventListener(t,n)},{passive:!0})})})})()</script>
403<div class=container><footer class=site-footer><p class=powered-by>© 2026 Gordon Guocheng Qian. Thanks for the template from wowchemy-hugo-themes. 404<span class=float-right aria-hidden=true><a href=# class=back-to-top><span class=button_icon><i class="fas fa-chevron-up fa-2x"></i></span></a></span></p></footer></div><div id=modal class="modal fade" tabindex=-1 role=dialog aria-hidden=true><div class=modal-dialog><div class=modal-content><div class=modal-header><h5 class=modal-title>Cite</h5><button type=button class=close data-dismiss=modal aria-label=Close> 405<span aria-hidden=true>×</span></button></div><div class=modal-body><pre><code class="tex hljs"></code></pre></div><div class=modal-footer><a class="btn btn-outline-primary my-1 js-copy-cite" href=# target=_blank><i class="fas fa-copy"></i> Copy 406</a><a class="btn btn-outline-primary my-1 js-download-cite" href=# target=_blank><i class="fas fa-download"></i> Download</a><div id=modal-error></div></div></div></div></div></body></html>
Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.