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-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:20240101T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20250515T150000
DTEND;TZID=Asia/Hong_Kong:20250515T160000
DTSTAMP:20260509T211546
CREATED:20250603T034632Z
LAST-MODIFIED:20250603T034632Z
UID:111567-1747321200-1747324800@ece.hku.hk
SUMMARY:Distributed Mixture-of-Expert Systems at the Wireless Edge (Duplicate)
DESCRIPTION:Zoom Link: https://hku.zoom.us/j/91757354553?pwd=tHpInMTglaIVMJLek0ydP0vddHihh8.1 \nMeeting ID: 917 5735 4553\nPasscode: 587193 \nAbstract\nExisting Video-to-Audio (V2A) models typically generate sound based solely on visual input\, offering limited user control. To address this limitation\, we propose a multimodal controllable V2A system that conditions audio generation on a variety of user inputs– such as text\, images\, or audio– in addition to video. Our approach leverages the ImageBind model to align these diverse input modalities into a shared representation\, which is then used to guide audio generation. \nDue to the lack of multimodal datasets for audio generation\, we constructed a training dataset comprising large-scale text-audio data complemented with a limited amount of video-audio data to enable controllable and context-aware audio generation. To further enhance generative quality\, we introduce several data ensemble strategies: (1) Source Balancing\, which maintains a trade-off between concept diversity and sample diversity\, and (2) two synchronization techniques– Audio Feature Selector (AFS) and Audio Peak IoU Matching (APIM)– to improve temporal alignment between video and generated audio. \nOur system enables flexible and precise audio generation that aligns closely with multimodal user intent. Finally\, we introduce a novel benchmark with a cross-sample evaluation framework\, designed to standardize assessments of multimodal V2A systems by evaluating consistency and diversity across input-output combinations. Our method achieves state-of-the-art performance\, demonstrating significant improvements in audio quality\, synchronization\, and input-condition alignment. \nSpeaker\nMr. HE Ruifei\nDepartment of Electrical and Electronic Engineering\nThe University of Hong Kong \nSpeaker’s Biography\nMr. Ruifei He is a final-year Ph.D. student in Department of Electrical and Electronic Engineering at the University of Hong Kong. He obtained his B.Eng. degree in the Department of Automation at Zhejiang University. His research focuses on data-centric/efficient learning (e.g. synthetic/generative data\, mixing multi-modal data\, and semi-supervised learning) for computer vision tasks. \n\nAll are welcome!
URL:https://ece.hku.hk/events/20250515-1/
LOCATION:Online via Zoom
CATEGORIES:Highlights,Seminar
ATTACH;FMTTYPE=image/jpeg:https://ece.hku.hk/wp-content/uploads/2024/11/rpg-seminar.jpg
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Hong_Kong:20250515T150000
DTEND;TZID=Asia/Hong_Kong:20250515T160000
DTSTAMP:20260509T211546
CREATED:20250603T035131Z
LAST-MODIFIED:20250603T035131Z
UID:111572-1747321200-1747324800@ece.hku.hk
SUMMARY:End-to-end High-quality Posterior Ocular Shape Reconstruction in Ophthalmology
DESCRIPTION:Zoom Link: https://hku.zoom.us/j/98147018160?pwd=KEuU1XQtISq3HQpWJyF6itZMJ1hYY5.1\nMeeting ID: 981 4701 8160\nPassword: 294231 \nAbstract\nAccurately estimating morphological changes of the Posterior Eyeball Shape (PES) is a critical task in ophthalmology\, since the PES is a crucial factor in many clinical applications\, such as myopia prevention\, surgical planning\, and disease screening. However\, existing imaging devices are constrained by limited field-of-view (FOV) and insufficient resolution\, thus providing insufficient diagnostic information for surgeons to make accurate decisions. Previous segment-based reconstruction methods suffer from two main drawbacks: first\, common imaging modalities can’t provide intact enough shape details as constraints\, thus requiring intricate pre- and post-processing; second\, existing data representations struggle to trade-off between computational efficiency and reconstruction quality\, thus hindering end-to-end reconstruction with fine-grained shape details. \nBenefiting from our more efficient 2D representation in polar coordinate\, we propose a novel task of reconstructing intact 3D PES based on purely small-FOV OCT scans and introduces a novel Posterior Eyeball Shape Network (PESNet) to accomplish this task. The proposed PESNet equips the Siamese structure that incorporates anatomical information of the eyeball as guidance. To capture more detailed information\, we introduce a Polar Voxelization Block (PVB) that transfers sparse input point clouds to a dense representation. Furthermore\, we propose a Radius-wise Fusion Block (RFB) that fuses correlative hierarchical features from the two branches. Finally\, this high-cost reconstruction task is compressed into the 2D surface map regression task. The experiments indicate that our method achieves state-of the-art performance\, providing a well-represented complete posterior eyeball shape on both healthy and patient cases. This result demonstrates that our method offers a significant improvement over existing methods in accurately reconstructing the complete 3D posterior eyeball shape. This achievement has important implications for clinical applications. \nSpeaker\nMr. ZHANG Jiaqi\nDepartment of Electrical and Electronic Engineering\nThe University of Hong Kong \nSpeaker’s Biography\nMr. Jiaqi Zhang obtained his B.Sc. degree from Northeastern University (NEU) in China and his M.Res. degree from the National University of Singapore (NUS) in Singapore. He is currently pursuing the Ph.D. in the Department of Electrical and Electronic Engineering at the University of Hong Kong (HKU)\, under the supervision of Prof. Xiaojuan Qi. His research focuses on medical image processing\, 3D reconstruction\, AIGC\, and representation learning. \n\nAll are welcome!
URL:https://ece.hku.hk/events/20250515-2/
LOCATION:Online via Zoom
CATEGORIES:Highlights,Seminar
ATTACH;FMTTYPE=image/jpeg:https://ece.hku.hk/wp-content/uploads/2024/11/rpg-seminar.jpg
END:VEVENT
END:VCALENDAR