{"id":58,"date":"2026-05-06T16:52:17","date_gmt":"2026-05-06T16:52:17","guid":{"rendered":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/?page_id=58"},"modified":"2026-05-08T02:06:45","modified_gmt":"2026-05-08T02:06:45","slug":"current-work","status":"publish","type":"page","link":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/","title":{"rendered":"Current & Proposed work"},"content":{"rendered":"\n

Improving HOSIG<\/h2>\n\n\n\n

We are building a baseline HOI\/HSI pipeline on top of HOSIG[2], whose base model was trained on only 3 objects using the right hand. To improve generalization, we are retraining it with a larger, filtered dataset from TRUMANS[5] covering greater object diversity and closer grasping poses.<\/p>\n\n\n\n

\"\"
common SGAP errors<\/figcaption><\/figure>\n\n\n\n

In doing so, we identified the SGAP module as the primary source of common artifacts, including body-scene penetrations, hand-object penetrations, and unnatural poses for out-of-distribution objects. <\/p>\n\n\n\n

SGAP works in two main stages, the GNet cVAE pose generation phase followed up by an iterative contact based loss optimisation on hand and object contact.<\/p>\n\n\n\n

\"\"
SGAP works in two phases<\/figcaption><\/figure>\n\n\n\n

Improving the base GNet [stage 1]<\/h2>\n\n\n\n

We retrained the SGAP model with a the new dataset and refined losses to address this. The results below show a clear improvement in grasping quality, specially for the penetration metrics.<\/p>\n\n\n

\n
\"\"
Orange = New dataset, Blue = Original Sgap dataset <\/figcaption><\/figure>\n<\/div>\n\n\n
\"\"
results of model retraining with modified loss scheme<\/figcaption><\/figure>\n\n\n\n

Improving pose refinement [stage 2]<\/h2>\n\n\n\n

We replace the original single-step, full-body optimization with a 3-stage approach that prioritizes hand\/arm refinement before body pose, removes overpowering losses like gaze, and applies a lower learning rate for body adjustments<\/p>\n\n\n\n

\"\"
blue shows the original coarase pose<\/figcaption><\/figure>\n\n\n\n

Video gen based HOI<\/h2>\n\n\n\n

One of the core challenges in HOI generation is generalizing to novel objects and diverse motion patterns. We explore using video generation as a prior to address this \u2014 leveraging WAN[6] to synthesize videos from captions and a starting frame, using the first frame’s depth and scene geometry as a conditioning signal. Object trajectories and hand poses are then recovered in parallel via FoundationPose[6] and HaMeR[7], producing a 3D HOI output.<\/p>\n\n\n\n

We are also exploring how to integrate video-based HOI generation with full-body motion, toward a unified pipeline.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Proposed framework<\/h2>\n\n\n\n

Ultimately, these components will be integrated into a cohesive pipeline, enhanced with a VLM planner to reason over visual inputs and determine appropriate interactions, and a human-in-the-loop framework for controllability<\/p>\n\n\n\n

\"\"
Proposed framework<\/figcaption><\/figure>\n","protected":false},"excerpt":{"rendered":"

Improving HOSIG We are building a baseline HOI\/HSI pipeline on top of HOSIG[2], whose base model was trained on only 3 objects using the right hand. To improve generalization, we are retraining it with a larger, filtered dataset from TRUMANS[5] covering greater object diversity and closer grasping poses. In doing so, we identified the SGAP … <\/p>\n

Continue reading “Current & Proposed work”<\/span><\/a><\/p>\n","protected":false},"author":271,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-58","page","type-page","status-publish","hentry"],"yoast_head":"\nCurrent & Proposed work - Scene-Aware, User-Guided Generation of 3D Human-Object Interactions<\/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\/2026teamf4\/current-work\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Current & Proposed work - Scene-Aware, User-Guided Generation of 3D Human-Object Interactions\" \/>\n<meta property=\"og:description\" content=\"Improving HOSIG We are building a baseline HOI\/HSI pipeline on top of HOSIG[2], whose base model was trained on only 3 objects using the right hand. To improve generalization, we are retraining it with a larger, filtered dataset from TRUMANS[5] covering greater object diversity and closer grasping poses. In doing so, we identified the SGAP … Continue reading "Current & Proposed work"\" \/>\n<meta property=\"og:url\" content=\"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/\" \/>\n<meta property=\"og:site_name\" content=\"Scene-Aware, User-Guided Generation of 3D Human-Object Interactions\" \/>\n<meta property=\"article:modified_time\" content=\"2026-05-08T02:06:45+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-content\/uploads\/sites\/146\/2026\/05\/image-7.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1335\" \/>\n\t<meta property=\"og:image:height\" content=\"375\" \/>\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\\\/2026teamf4\\\/current-work\\\/\",\"url\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/current-work\\\/\",\"name\":\"Current & Proposed work - Scene-Aware, User-Guided Generation of 3D Human-Object Interactions\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/current-work\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/current-work\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/wp-content\\\/uploads\\\/sites\\\/146\\\/2026\\\/05\\\/image-7-1024x288.png\",\"datePublished\":\"2026-05-06T16:52:17+00:00\",\"dateModified\":\"2026-05-08T02:06:45+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/current-work\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/current-work\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/current-work\\\/#primaryimage\",\"url\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/wp-content\\\/uploads\\\/sites\\\/146\\\/2026\\\/05\\\/image-7.png\",\"contentUrl\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/wp-content\\\/uploads\\\/sites\\\/146\\\/2026\\\/05\\\/image-7.png\",\"width\":1335,\"height\":375},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/current-work\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Current & Proposed work\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/#website\",\"url\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/\",\"name\":\"Scene-Aware, User-Guided Generation of 3D Human-Object Interactions\",\"description\":\"Sambhav Rohatgi, Alex Zhao, Yizhou Zhao, Mose Sakashita, Koichiro Niinuma, L\u00e1szl\u00f3 A. Jeni\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/mscvprojects.ri.cmu.edu\\\/2026teamf4\\\/?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":"Current & Proposed work - Scene-Aware, User-Guided Generation of 3D Human-Object Interactions","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\/2026teamf4\/current-work\/","og_locale":"en_US","og_type":"article","og_title":"Current & Proposed work - Scene-Aware, User-Guided Generation of 3D Human-Object Interactions","og_description":"Improving HOSIG We are building a baseline HOI\/HSI pipeline on top of HOSIG[2], whose base model was trained on only 3 objects using the right hand. To improve generalization, we are retraining it with a larger, filtered dataset from TRUMANS[5] covering greater object diversity and closer grasping poses. In doing so, we identified the SGAP … Continue reading \"Current & Proposed work\"","og_url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/","og_site_name":"Scene-Aware, User-Guided Generation of 3D Human-Object Interactions","article_modified_time":"2026-05-08T02:06:45+00:00","og_image":[{"width":1335,"height":375,"url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-content\/uploads\/sites\/146\/2026\/05\/image-7.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\/2026teamf4\/current-work\/","url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/","name":"Current & Proposed work - Scene-Aware, User-Guided Generation of 3D Human-Object Interactions","isPartOf":{"@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/#website"},"primaryImageOfPage":{"@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/#primaryimage"},"image":{"@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/#primaryimage"},"thumbnailUrl":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-content\/uploads\/sites\/146\/2026\/05\/image-7-1024x288.png","datePublished":"2026-05-06T16:52:17+00:00","dateModified":"2026-05-08T02:06:45+00:00","breadcrumb":{"@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/#primaryimage","url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-content\/uploads\/sites\/146\/2026\/05\/image-7.png","contentUrl":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-content\/uploads\/sites\/146\/2026\/05\/image-7.png","width":1335,"height":375},{"@type":"BreadcrumbList","@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/current-work\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/"},{"@type":"ListItem","position":2,"name":"Current & Proposed work"}]},{"@type":"WebSite","@id":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/#website","url":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/","name":"Scene-Aware, User-Guided Generation of 3D Human-Object Interactions","description":"Sambhav Rohatgi, Alex Zhao, Yizhou Zhao, Mose Sakashita, Koichiro Niinuma, L\u00e1szl\u00f3 A. Jeni","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/?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\/2026teamf4\/wp-json\/wp\/v2\/pages\/58","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-json\/wp\/v2\/users\/271"}],"replies":[{"embeddable":true,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-json\/wp\/v2\/comments?post=58"}],"version-history":[{"count":3,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-json\/wp\/v2\/pages\/58\/revisions"}],"predecessor-version":[{"id":92,"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-json\/wp\/v2\/pages\/58\/revisions\/92"}],"wp:attachment":[{"href":"https:\/\/mscvprojects.ri.cmu.edu\/2026teamf4\/wp-json\/wp\/v2\/media?parent=58"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}