Reasoning about the 3D world from multi-view images remains a fundamental challenge for Multimodal Large Language Models (MLLMs). While modern MLLMs handle single-image inputs effectively, they struggle to integrate evidence across viewpoints into a coherent 3D understanding. A growing body of work attempts to close this gap by injecting 3D awareness into MLLMs, either by boosting fine-grained pixel-level cross-view correspondence or by fusing features from 3D geometry foundation models, yet a substantial gap to human reasoning persists. In this work, we revisit human spatial reasoning, which suggests that rather than relying on fine-grained geometry cues, humans roughly identify common objects across views, infer the relative geometry between viewpoints, and assemble a coarse 3D layout of the scene. Inspired by this process, we introduce Imagine3D-LLM, an MLLM that learns to assemble a similar compact 3D representation of the scene and conditions its answer on this representation. Concretely, we append a small set of learnable summary tokens after the image tokens, decode them into a compact 3D Gaussian Splatting representation supervised by a photometric reconstruction loss, and train jointly with the standard next-token prediction objective. Notably, although only the summary tokens receive direct reconstruction supervision, this objective also induces stronger cross-frame correspondence within the LLM’s underlying image features, suggesting that learning to reconstruct propagates 3D-aware signals throughout the model. As a result, Imagine3D-LLM consistently outperforms prior approaches across multiple spatial reasoning and 3D understanding benchmarks, suggesting that imagining the scene can be more effective than being told its pixel-wise geometry.<\/p>\n","protected":false},"excerpt":{"rendered":"
Reasoning about the 3D world from multi-view images remains a fundamental challenge for Multimodal Large Language Models (MLLMs). While modern MLLMs handle single-image inputs effectively, they struggle to integrate evidence across viewpoints into a coherent 3D understanding. A growing body of work attempts to close this gap by injecting 3D awareness into MLLMs, either by […]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-author-ordering":[{"type":"text","value":"Jaewoo Jung","user_id":0},{"type":"text","value":"Hyeonseo Yu","user_id":0},{"type":"text","value":"Honggyu An","user_id":0},{"type":"text","value":"Jisang Han","user_id":0},{"type":"text","value":"Mungyeom Kim","user_id":0},{"type":"text","value":"Minkyeong Jeon","user_id":0},{"type":"text","value":"Heeseong Shin","user_id":0},{"type":"text","value":"WonJun Moon","user_id":0},{"type":"text","value":"Federico Tombari","user_id":0},{"type":"text","value":"Dániel Baráth","user_id":0},{"type":"user_nicename","value":"Marc Pollefeys","user_id":"36191"},{"type":"text","value":"Seungryong Kim","user_id":0},{"type":"text","value":"Sungjin Hong","user_id":0}],"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"","msr_isbn":"","msr_issue":"","msr_journal":"","msr_number":"","msr_organization":"NeurIPS","msr_pages_string":"","msr_page_range_start":"","msr_page_range_end":"","msr_series":"","msr_volume":"","msr_copyright":"","msr_conference_name":"NeurIPS 2026","msr_doi":"","msr_arxiv_id":"2609.38177","msr_mag_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_release_tracker_id":"","msr_highlight_type":"","msr_date_display_format":"","msr_main_download_label":"","msr_external_link_label":"","msr_doi_label":"","msr_published_date":"2026-09-29","msr_startdate":"","msr_presentation_date":"","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_year":2026,"msr_month":9,"msr_day":29,"msr_microsoftintellectualproperty":false,"msr_pub_id":"4caeeedd27419579b82095b2f0c49e5d32787cc6","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2609.38177","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_original_fields_of_study":[],"msr_s2_paper_id":"4caeeedd27419579b82095b2f0c49e5d32787cc6","msr_s2_pdf_url":"","msr_citation_count_updated":"","msr_citation_count":2,"msr_influential_citations":0,"msr_reference_count":96,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[{"provider":"s2","id":"4caeeedd27419579b82095b2f0c49e5d32787cc6"},{"provider":"arxiv","id":"2609.38177"},{"provider":"corpusid","id":"292474206"}],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13556,13562,13551],"msr-publication-type":[193716],"msr-publisher":[],"msr-publication-cta":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[246691,263185],"msr-conference":[259048],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1190659","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-research-area-computer-vision","msr-research-area-graphics-and-multimedia","msr-locale-en_us","msr-field-of-study-computer-science","msr-field-of-study-computer-vision-and-pattern-recognition"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2026-09-29","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"NeurIPS","msr_how_published":"","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":0,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2609.38177","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_citation_count":2,"msr_citation_count_updated":"","msr_s2_paper_id":"4caeeedd27419579b82095b2f0c49e5d32787cc6","msr_influential_citations":0,"msr_reference_count":96,"msr_arxiv_id":"2609.38177","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[],"msr-author-ordering":[{"type":"text","value":"Jaewoo Jung","user_id":0,"rest_url":false},{"type":"text","value":"Hyeonseo Yu","user_id":0,"rest_url":false},{"type":"text","value":"Honggyu An","user_id":0,"rest_url":false},{"type":"text","value":"Jisang Han","user_id":0,"rest_url":false},{"type":"text","value":"Mungyeom Kim","user_id":0,"rest_url":false},{"type":"text","value":"Minkyeong Jeon","user_id":0,"rest_url":false},{"type":"text","value":"Heeseong Shin","user_id":0,"rest_url":false},{"type":"text","value":"WonJun Moon","user_id":0,"rest_url":false},{"type":"text","value":"Federico Tombari","user_id":0,"rest_url":false},{"type":"text","value":"Dániel Baráth","user_id":0,"rest_url":false},{"type":"user_nicename","value":"Marc Pollefeys","user_id":36191,"rest_url":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/microsoft-research\/v1\/researchers?person=Marc Pollefeys"},{"type":"text","value":"Seungryong Kim","user_id":0,"rest_url":false},{"type":"text","value":"Sungjin Hong","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[602418],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"inproceedings","related_content":[],"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190659","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":4,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190659\/revisions"}],"predecessor-version":[{"id":1191030,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190659\/revisions\/1191030"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1190659"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1190659"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1190659"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1190659"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=1190659"},{"taxonomy":"msr-publication-cta","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-cta?post=1190659"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1190659"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1190659"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1190659"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1190659"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1190659"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1190659"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1190659"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1190659"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}