BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Department of Electrical and Computer Engineering (HKUECE) 電機與計算機工程系 - ECPv6.15.20//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Department of Electrical and Computer Engineering (HKUECE) 電機與計算機工程系
X-ORIGINAL-URL:https://ece.hku.hk
X-WR-CALDESC:Events for Department of Electrical and Computer Engineering (HKUECE) 電機與計算機工程系
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:Asia/Hong_Kong
BEGIN:STANDARD
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
TZNAME:HKT
DTSTART:20250101T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20260326T170000
DTEND;TZID=Asia/Hong_Kong:20260326T180000
DTSTAMP:20260511T074929
CREATED:20260320T094513Z
LAST-MODIFIED:20260320T094513Z
UID:115340-1774544400-1774548000@ece.hku.hk
SUMMARY:RPG Seminar – Scaling Up Spatial Awareness: High-Fidelity Data Synthesis for 3D Scene Understanding
DESCRIPTION:Zoom Link:\nhttps://hku.zoom.us/j/91627715757?pwd=ByKZvbK3QYx8VSWXVoNGBsZXTpFEz3.1 \nAbstract\nSpatial understanding constitutes a fundamental pillar of human-level intelligence\, yet its advancement is currently bottlenecked by the scarcity of diverse\, high-fidelity 3D data. Existing research predominantly relies on domain-specific or manually annotated datasets\, creating a critical void: the absence of a principled\, scalable engine capable of synthesizing high-quality spatial data at scale. To address this\, we elucidate the core design principles for robust spatial data generation and introduce OpenSpatial—an open-source engine engineered for high fidelity\, massive scalability\, and broad task diversity. OpenSpatial adopts 3D bounding boxes as the foundational primitive to architect a comprehensive data hierarchy across five essential dimensions: Spatial Measurement\, Spatial Relationship\, Camera Perception\, Multi-view Consistency\, and Scene-Aware Reasoning. Leveraging this infrastructure\, we curate OpenSpatial-3M\, a large-scale dataset that enables models to transition from simple recognition to sophisticated spatial intelligence. Extensive evaluations demonstrate that models trained on our synthesized data achieve state-of-the-art performance across a wide spectrum of benchmarks\, showing substantial and consistent improvements over existing baselines. Furthermore\, we provide a systematic analysis of how synthesized data attributes influence the emergence of spatial perception in vision-language models. By open-sourcing both the engine and the 3M-scale dataset\, we offer a versatile foundation to accelerate future research in generalized 3D scene understanding. \nSpeaker\nMr. Jianhui Liu\nDepartment of Electrical and Computer Engineering\nThe University of Hong Kong \nBiography of the Speaker\nMr. Jianhui Liu is a PhD candidate with the Department of Electrical and Electronic Engineering at the University of Hong Kong. He received the B.Eng. degree in Intelligent Science and Technology from Xidian University in 2021. His research interest lies in machine learning and computer vision\, focusing on Multimodal Large Language Models (MLLMs) for reasoning\, agent\, long video\, spatial intelligence\, unified models\, and their real-world grounding and applications. \nOrganiser\nProf. Xiaojuan Qi\nDepartment of Electrical and Computer Engineering\, The University of Hong Kong \nAll are welcome.
URL:https://ece.hku.hk/events/20260326/
CATEGORIES:Seminar
ATTACH;FMTTYPE=image/jpeg:https://ece.hku.hk/wp-content/uploads/2024/11/rpg-seminar.jpg
END:VEVENT
END:VCALENDAR