{"id":43,"date":"2026-05-08T02:11:32","date_gmt":"2026-05-08T02:11:32","guid":{"rendered":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/?page_id=43"},"modified":"2026-05-08T02:58:09","modified_gmt":"2026-05-08T02:58:09","slug":"method","status":"publish","type":"page","link":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/","title":{"rendered":"Method"},"content":{"rendered":"\n
\n
\n
\n
  • <\/li><\/ul>\n\n<\/div>\n<\/div>\n\n\n\n
    <\/div>\n<\/div>\n\n\n\n
    <\/div>\n<\/div>\n\n\n\n
    \n
    \n\n

    We present a novel, multi-stage framework for high-fidelity 3D generation combining a chain-of-thought compositional pipeline with DINO-based feature injection. First, vision-language models semantically decompose a complex input image into isolated parts, utilizing 2D generative models to isolate the components before reconstructing them via 3D generative models. To ensure multi-view consistency and structural coherence during this synthesis, we introduce a feature disentanglement technique using DINOv2. By performing channel-space PCA and variance analysis on the extracted features, we decouple stable, view-invariant structural channels from unstable, view-specific detail channels. This targeted injection mechanism allows us to seamlessly swap high-dispersion channels from a reference view into the structurally consistent channels of a target view, achieving zero-shot view-conditioning while preserving rigid geometric integrity.<\/p>\n\n\n\n

    Mathematical Formulation of DINO Feature Injection<\/h3>\n\n\n\n

    The feature injection approach can be formally described using the following equations:<\/p>\n\n\n\n

    1. Feature Extraction:<\/strong>
    Given an input view I_view, we extract the intermediate token features using DINOv2. The resulting feature map has a channel dimension C = 1024.<\/p>\n\n\n\n\n\n

    F<\/mi>v<\/mi>i<\/mi>e<\/mi>w<\/mi><\/mrow><\/msub>=<\/mo>\u03a6<\/mi><\/mrow>DINOv2<\/mtext><\/msub>(<\/mo>I<\/mi>v<\/mi>i<\/mi>e<\/mi>w<\/mi><\/mrow><\/msub>)<\/mo>\u2208<\/mo>\u211d<\/mi>C<\/mi>\u00d7<\/mo>H<\/mi>\u00d7<\/mo>W<\/mi><\/mrow><\/msup><\/mrow>F_{view} = \\Phi_{\\text{DINOv2}}(I_{view}) \\in \\mathbb{R}^{C \\times H \\times W}<\/annotation><\/semantics><\/math><\/div>\n\n\n\n

    2. Variance Analysis (Disentanglement):<\/strong><\/p>\n\n\n\n

    To separate global structure from view-specific details, we compute the spatial variance for each channel c across multiple views. Channels with high variance represent view-dependent details.<\/p>\n\n\n\n

    \u03c3<\/mi>c<\/mi>2<\/mn><\/msubsup>=<\/mo>1<\/mn>N<\/mi><\/mfrac>\u2211<\/mo>i<\/mi>=<\/mo>1<\/mn><\/mrow>N<\/mi><\/munderover><\/mrow>(<\/mo>F<\/mi>v<\/mi>i<\/mi><\/msub>(<\/mo>c<\/mi>)<\/mo><\/mrow><\/msubsup>\u2212<\/mo>\u03bc<\/mi>(<\/mo>c<\/mi>)<\/mo><\/mrow><\/msup>)<\/mo><\/mrow>2<\/mn><\/msup><\/mrow>\\sigma_c^2 = \\frac{1}{N} \\sum_{i=1}^{N} \\left( F_{v_i}^{(c)} – \\mu^{(c)} \\right)^2<\/annotation><\/semantics><\/math><\/div>\n\n\n\n

    3. Channel Selection (Masking):<\/strong><\/p>\n\n\n\n

    We define a channel-wise binary mask M based on a variance threshold tau. Channels exceeding this threshold are classified as “High-Dispersion” (M_c = 1), while the rest are “Structurally Consistent” (M_c = 0).<\/p>\n\n\n\n\n\n

    M<\/mi>c<\/mi><\/msub>=<\/mo>{<\/mo>1<\/mn>,<\/mo><\/mrow><\/mtd>if <\/mtext>\u03c3<\/mi>c<\/mi>2<\/mn><\/msubsup>><\/mo>\u03c4<\/mi><\/mrow><\/mtd><\/mtr>0<\/mn>,<\/mo><\/mrow><\/mtd>if <\/mtext>\u03c3<\/mi>c<\/mi>2<\/mn><\/msubsup>\u2264<\/mo>\u03c4<\/mi><\/mrow><\/mtd><\/mtr><\/mtable><\/mo><\/mrow><\/mrow>M_c = \\begin{cases} 1, & \\text{if } \\sigma_c^2 > \\tau \\\\ 0, & \\text{if } \\sigma_c^2 \\le \\tau \\end{cases}<\/annotation><\/semantics><\/math><\/div>\n\n\n\n

    4. Targeted Feature Injection:<\/strong><\/p>\n\n\n\n

    The novel feature map F_new is reconstructed by mixing the high-dispersion channels from the front\/reference view with the structurally consistent channels from the back\/target view using the mask M, where \\odot denotes channel-wise multiplication.<\/p>\n\n\n\n

    F<\/mi>n<\/mi>e<\/mi>w<\/mi><\/mrow><\/msub>=<\/mo>M<\/mi>\u2299<\/mo>F<\/mi>f<\/mi>r<\/mi>o<\/mi>n<\/mi>t<\/mi><\/mrow><\/msub>+<\/mo>(<\/mo>\ud835\udfcf<\/mn>\u2212<\/mo>M<\/mi>)<\/mo>\u2299<\/mo>F<\/mi>b<\/mi>a<\/mi>c<\/mi>k<\/mi><\/mrow><\/msub><\/mrow>F_{new} = M \\odot F_{front} + (\\mathbf{1} – M) \\odot F_{back}\n<\/annotation><\/semantics><\/math><\/div>\n\n<\/div><\/div>\n\n\n\n
    <\/div>\n<\/div>\n\n\n\n
    \n
    <\/div>\n\n\n\n
    \n
    <\/div>\n\n\n\n
    <\/div>\n<\/div>\n\n\n\n
    \"\"<\/figure>\n\n\n\n
    <\/div>\n\n\n\n
    \n\n\n\n
    <\/div>\n<\/div>\n\n\n\n
    \n
    \n
    \n

    Twenty Twenty-Five<\/p>\n\n\n\n

    email@example.com
    +1 555 349 1806<\/p>\n<\/div>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

    Twenty Twenty-Five email@example.com+1 555 349 1806<\/p>\n","protected":false},"author":266,"featured_media":62,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-43","page","type-page","status-publish","has-post-thumbnail","hentry"],"yoast_head":"\nMethod - 3D Asset Generation: Semantic Part Decomposition<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Method - 3D Asset Generation: Semantic Part Decomposition\" \/>\n<meta property=\"og:description\" content=\"Twenty Twenty-Five email@example.com+1 555 349 1806\" \/>\n<meta property=\"og:url\" content=\"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/\" \/>\n<meta property=\"og:site_name\" content=\"3D Asset Generation: Semantic Part Decomposition\" \/>\n<meta property=\"article:modified_time\" content=\"2026-05-08T02:58:09+00:00\" \/>\n<meta property=\"og:image\" content=\"http:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-content\/uploads\/sites\/143\/2026\/05\/Screenshot-2026-05-07-at-10.39.51-PM.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1426\" \/>\n\t<meta property=\"og:image:height\" content=\"572\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data1\" content=\"3 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/method\\\/\",\"url\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/method\\\/\",\"name\":\"Method - 3D Asset Generation: Semantic Part Decomposition\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/method\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/method\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/wp-content\\\/uploads\\\/sites\\\/143\\\/2026\\\/05\\\/Screenshot-2026-05-07-at-10.39.51-PM.png\",\"datePublished\":\"2026-05-08T02:11:32+00:00\",\"dateModified\":\"2026-05-08T02:58:09+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/method\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/method\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/method\\\/#primaryimage\",\"url\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/wp-content\\\/uploads\\\/sites\\\/143\\\/2026\\\/05\\\/Screenshot-2026-05-07-at-10.39.51-PM.png\",\"contentUrl\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/wp-content\\\/uploads\\\/sites\\\/143\\\/2026\\\/05\\\/Screenshot-2026-05-07-at-10.39.51-PM.png\",\"width\":1426,\"height\":572},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/method\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Method\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/#website\",\"url\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/\",\"name\":\"3D Asset Generation: Semantic Part Decomposition\",\"description\":\"\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf1\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Method - 3D Asset Generation: Semantic Part Decomposition","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/","og_locale":"en_US","og_type":"article","og_title":"Method - 3D Asset Generation: Semantic Part Decomposition","og_description":"Twenty Twenty-Five email@example.com+1 555 349 1806","og_url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/","og_site_name":"3D Asset Generation: Semantic Part Decomposition","article_modified_time":"2026-05-08T02:58:09+00:00","og_image":[{"width":1426,"height":572,"url":"http:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-content\/uploads\/sites\/143\/2026\/05\/Screenshot-2026-05-07-at-10.39.51-PM.png","type":"image\/png"}],"twitter_card":"summary_large_image","twitter_misc":{"Est. reading time":"3 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/","url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/","name":"Method - 3D Asset Generation: Semantic Part Decomposition","isPartOf":{"@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/#website"},"primaryImageOfPage":{"@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/#primaryimage"},"image":{"@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/#primaryimage"},"thumbnailUrl":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-content\/uploads\/sites\/143\/2026\/05\/Screenshot-2026-05-07-at-10.39.51-PM.png","datePublished":"2026-05-08T02:11:32+00:00","dateModified":"2026-05-08T02:58:09+00:00","breadcrumb":{"@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/#primaryimage","url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-content\/uploads\/sites\/143\/2026\/05\/Screenshot-2026-05-07-at-10.39.51-PM.png","contentUrl":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-content\/uploads\/sites\/143\/2026\/05\/Screenshot-2026-05-07-at-10.39.51-PM.png","width":1426,"height":572},{"@type":"BreadcrumbList","@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/method\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/"},{"@type":"ListItem","position":2,"name":"Method"}]},{"@type":"WebSite","@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/#website","url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/","name":"3D Asset Generation: Semantic Part Decomposition","description":"","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"}]}},"_links":{"self":[{"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/pages\/43","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/users\/266"}],"replies":[{"embeddable":true,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/comments?post=43"}],"version-history":[{"count":4,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/pages\/43\/revisions"}],"predecessor-version":[{"id":69,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/pages\/43\/revisions\/69"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/media\/62"}],"wp:attachment":[{"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf1\/wp-json\/wp\/v2\/media?parent=43"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}