• Invited Talk 1: Minju Baek

Ph.D. student @ Chung-Ang University (CAU)

• Bio.: Minju Baek is a Ph.D. student at the Graduate School of Artificial Intelligence, Chung-Ang University, advised by Professor Joonki Paik. Her research focuses on infrared (IR) object detection, domain adaptation, and domain generalization, with an emphasis on bridging the modality gap between RGB and IR data for robust perception in challenging real-world conditions such as UAV and aerial imagery. Her recent work includes CLIP-driven RoI-level semantic alignment for infrared object detection (ICIP 2026) and structure-aware phase-based dual alignment for robust UAV object detection (ICPR 2026).

• Title: Asymmetric Dual-Teacher Learning for Label-Free Thermal Adaptation of Frozen Detectors

  • Abstract: Thermal object detection is essential for reliable perception under adverse visibility conditions, yet its practical deployment is hindered by the high cost of target-domain bounding-box annotations and the substantial modality gap between RGB and thermal imagery. Existing RGB-to-thermal unsupervised domain adaptation methods typically require labeled RGB source data and adapt the detector itself, while modality translation approaches preserve a pretrained detector by adapting only the input space but still rely on target thermal annotations. In this work, we introduce a \textbf{label-free thermal adaptation} framework that requires neither source RGB data nor target thermal annotations during adaptation. Starting from an off-the-shelf RGB-pretrained detector, we keep the detector completely frozen and optimize only a lightweight modality translator using pseudo-labels generated by the detector itself. To address confirmation bias and prediction drift in this self-referential learning process, we propose a \textbf{dual-teacher pseudo-labeling strategy} that combines a fixed identity teacher, which provides a stable prediction anchor, with an adaptive EMA teacher that reflects the evolving translator. We further apply asymmetric confidence filtering to account for their different exposure to the self-training feedback loop, followed by IoU-based merging to construct reliable pseudo-labels.

• Invited Talk 2: Dorota Kamińska

Associate Professor @ Lodz University of Technology

• Bio.: Dorota Kamińska is an Associate Professor at Lodz University of Technology and the founder and head of the Voxel Research Lab. Her research lies at the intersection of extended reality, affective computing, human–computer interaction, and human-centered AI, with a strong focus on immersive technologies for education, healthcare, accessibility, and inclusive design. She has authored more than 100 scientific publications and has been recognized among the World’s Top 2% Scientists in the field of Information and Communication Technologies. She leads and contributes to numerous international research initiatives exploring how AI and immersive technologies can create more adaptive, engaging, and human-centered experiences.

• Title: One Metaverse, Many Users: Human-Centered Evaluation of Immersive Technologies Across the Lifespan

  • Abstract: Immersive technologies are increasingly envisioned as a foundation for future interactive and AI-enhanced environments. Yet, their usability and effectiveness strongly depend on the characteristics, abilities, and expectations of the people who enter these virtual worlds. A system designed around a young, technologically experienced user may create very different challenges for a child, an older adult, or a person with specific cognitive or sensory needs. This talk draws on a series of user studies involving immersive applications evaluated with participants from different age groups, including children and adolescents, young adults, and older adults. Across educational, wellbeing, cognitive, and everyday-life scenarios, the studies reveal recurring differences in interaction strategies, cognitive workload, attention, navigation, understanding of virtual cues, physical interaction, presence, and tolerance of immersive stimulation. Rather than treating age and user diversity as secondary usability factors, these observations suggest that they should become integral components of immersive system design. The talk discusses practical lessons for designing more inclusive virtual environments and considers how behavioral, contextual, and affective information could support future adaptive and human-centered AI systems capable of adjusting immersive experiences to the needs of individual users. Ultimately, building meaningful metaverse environments requires not only increasingly intelligent virtual worlds, but also a better understanding of the diverse humans who inhabit them.

• Keynote Speaker: Victoria Vesna

Professor @ UCLA Design | Media Arts; Director, Art|Sci Center

• Bio.: Victoria Vesna is a media artist and professor at UCLA whose work bridges art, science, and technology. As Director of the UCLA Art|Sci Center, she leads interdisciplinary collaborations exploring data, sound, biotechnology, networked systems, and immersive media. Her work has been exhibited internationally and has received numerous awards. She is also North American Editor of AI & Society: Journal of Knowledge, Culture, and Communication  (Springer/Verlag, UK). Through her research and artistic practice, she invites audiences to engage critically and creatively with the cultural, social, ethical, and environmental implications of emerging technologies. 

• Title: Across Scales and Cultures: Art–Science Perspectives on AI and Immersive Worlds

  • Abstract: This keynote draws on three decades of art–science collaborations using databases, telepresence, immersive environments, and AR/VR to create culturally responsive experiences. A central focus is scale—from the microscopic and atomic to the planetary—and the ways collaborations with microscopists, physicists, environmental scientists, and other researchers reshape how data, bodies, and technological systems are perceived. Selected artistic projects will be examined to show how artistic practice can reveal biases embedded in datasets and computational systems while opening alternative approaches to intelligence, embodiment, and participation. Case studies from the UCLA Art|Sci Center will highlight collaborative methods grounded in cultural awareness, authorship, and accountability, proposing ways for AI to engage one world through many lenses without reducing difference to categories or stereotypes. 

• Invited Talk 3: Haley Marks

Project Scientist @ California NanoSystems Institute (UCLA)

• Bio.: Haley Marks, PhD is a project scientist at the California NanoSystems Institute (UCLA), supporting the Advanced Light Microscopy & Spectroscopy (ALMS) shared resource facility. Her work focuses on quantitative imaging pipelines combining correlative imaging modalities including second harmonic generation, fluorescence lifetime imaging, Raman scattering, super-resolution, and multiphoton microscopy. Working across dozens of research groups, she builds customized image segmentation and analysis workflows which turn raw acquisitions into reproducible, metadata-rich measurements, producing trustworthy datasets for specimens ranging from clinical biopsies to cultural artifacts.

• Title: Going Beyond RGB Colorspace Using Spectroscopy, Microscopy, and Machine Learning for Quantitative Digitization of Biological and Cultural Specimens

  • Abstract: Conventional digitization of microscopic samples typically record appearance under a single illumination condition, and on one dimensional scale. Herein we describe correlative imaging workflows which instead recover chemically, spectrally, and structurally specific measurements using a combination of microscopy techniques. Second harmonic and fluorescence lifetime imaging with phasor analysis separate free and protein-bound metabolites without exogenous labels from extracellular matrix components, providing metabolic readout in live cells, organ-on-chip constructs, and tissue biopsies under multiphoton excitation. For additional chemical specificity, Raman scattering returns molecular and mineral fingerprints, applied to samples ranging from kidney stones to pigment identification for painting authentication, while second harmonic generation and correlative atomic force microscopy resolve nanoscale structural organization. As accurate quantification depends heavily on calibrated excitation and detection systems and on spectral standards in order to keep measurements comparable across instruments and sessions, random forest pixel classifiers and U-Net segmentation convert these acquisitions into per-object measurements at scale, producing the confidence in image measurements which multimodal models require for trustworthy artifact description and retrieval.

• Invited Talk 4: Fengqing Maggie Zhu 

Associate Professor @ Purdue University

• Bio.: Fengqing Maggie Zhu is an Associate Professor of the Elmore Family School of Electrical and Computer Engineering at Purdue University, West Lafayette, Indiana. She received the B.S.E.E. (with highest distinction), M.S. and Ph.D. degrees in Electrical and Computer Engineering from Purdue University in 2004, 2006 and 2011, respectively. Prior to joining Purdue in 2015, she was a Staff Researcher at Futurewei Technologies, where she received a Certification of Recognition for Core Technology Contribution. Her research interests include smart health with a focus on image based dietary assessment and wearable sensor data analysis, visual coding for machines, and application-driven visual data analytics.  She is the recipient of an NSF CISE Research Initiation Initiative (CRII) award, a Google Faculty Research Award, and an ESI and trainee poster award for the NIH Precision Nutrition workshop. Her group’s work on visual coding for machines has received the Best Algorithms Paper Award at the Winter Conference on Applications of Computer Vision (WACV) 2023 and the Best Paper Finalists at the Picture Coding Symposium (PCS) 2022. She is currently serving as the Chair of the IEEE MMSP-TC (2026-2028) and an Elected Member of the IVMSP-TC (2025-2027). She is also an Associate Editor for the IEEE Transactions on Multimedia (2025-2027). She has served on the organizing and program committees of major conferences in her field and received recognition such as the Outstanding Area Chair for ICME 2021. Dr. Zhu is a senior member of the IEEE.

• Title: Scalable 3D Gaussian Splatting for the Metaverse: From Pose-free Gaussian Splatting to Attribute-Aware Edge Streaming

  • Abstract: Real-time photorealistic scene perception and rendering form the foundation of interactive metaverse environments. While 3D Gaussian Splatting (3DGS) has emerged as an attractive representation for novel view synthesis, its deployment faces two fundamental challenges: 1) conventional reconstruction pipelines rely on structure-from-motion preprocessing that is difficult to obtain in real-world scenarios, and 2) the resulting large-scale Gaussian representations demand efficient storage and transmission. In this talk, I present recent advances to overcome these challenges, enabling unconstrained real-world capture and efficient streaming for interactive 3D/4D spaces.