Call for Special Sessions
MMM 2027 invites high-quality submissions to the accepted Special Sessions. Special Sessions provide focused venues for emerging and timely research topics within multimedia modeling, multimedia understanding, multimodal learning, generative AI, spatial intelligence, embodied systems, and related application domains.
Special Session papers will be included in the MMM 2027 proceedings and will follow the same formatting, length, anonymisation, and double-blind review guidelines as regular papers. All submitted papers, including invited papers, will undergo rigorous peer review coordinated by the MMM 2027 Technical Programme Committee and the Special Session organisers.
Authors should select the most relevant Special Session track when submitting their papers through the MMM 2027 submission system.
In total five special sessions have been accepted for MMM 2027.
1. Embodied Multimedia Modeling
Scope
Embodied Multimedia Modeling addresses a fundamental shift in how intelligent systems perceive, represent, and interact with the world. Classical multimedia research treats content as a static artifact to be acquired, indexed, and delivered—a model centered on passive consumption. Embodied Multimedia Modeling breaks this assumption: it places a purposeful agent at the center of the multimedia pipeline, where perception drives action and action reshapes perception in a continuous closed loop.
See details and Call for Papers
The paradigm is timely since that large vision-language-action models, world models, and immersive XR hardware have collectively elevated Embodied AI from a theoretical ambition to deployable reality. Yet the modeling challenges that arise at the intersection of embodiment and multimedia remain largely unaddressed by existing venues: How should a multimodal scene representation be structured so that an agent can plan actions over it? How can generative models produce content that is not merely plausible but physically and contextually consistent with an agent's environment? How do we evaluate immersion, presence, and task success in ways that go beyond pixel-level fidelity?
This special session brings these questions squarely into the MMM community, where expertise in multimedia modeling (e.g., structured representation, semantic indexing, cross-modal retrieval, and content understanding) offers a complementary perspective to the control-theoretic focus of robotics venues. We specifically target the modeling abstractions that sit between low-level sensor streams and high-level agent behavior: scene graphs for embodied retrieval, task-conditioned generative models, physiological and interaction signals as evaluation channels, and semantic communication pipelines that close the loop between perception and action.
This session continues the momentum of the inaugural Workshop on Embodied Multimedia held at IEEE ICME 2026 and the first tutorial and survey on Embodied Multimedia, and brings the agenda to MMM where the community's core strength in multimedia modeling can enrich and be enriched by the emerging field of Embodied AI.
Organisers
- Yang Liu, Tongji University, China
- Peng Sun, University of Ottawa, Canada
- Dingkang Yang, Fysics AI, China
- Jing Cheng, East China Normal University, China
Topics of Interest
Topics include, but are not limited to:
- Active and embodied scene modelling
- Agent-centric scene representation
- Next-best-view planning and active multi-sensor fusion
- 3D scene graph construction for embodied retrieval
- Multimodal indexing and retrieval for embodied agents
- Retrieval-augmented generation in dynamic environments
- Memory-efficient indexing of egocentric streams
- Generative embodied content and world models
- Physically consistent scene generation
- Avatar synthesis and motion synthesis
- Human-avatar and human-robot interaction
- Semantic and task-oriented multimedia communication
- Task-aware compression and edge-cloud collaboration
- VR/AR multimedia modelling and immersive interaction
- Haptic rendering and adaptive XR content delivery
- Physiological signal analysis for evaluating immersion and interaction comfort
- Large vision-language-action models for grounding, reasoning, and planning
- Embodied media arts, creative interaction, and AI-mediated artistic expression
Submission Types
- Full research papers, following the standard MMM submission format
2. From Micro to Macro: Video Understanding in the Era of Multimodal Large Models
Scope
This special session invites researchers and practitioners to explore recent advances in video understanding in the era of multimodal large models. As the field rapidly shifts from conventional task-specific pipelines to foundation-model-based paradigms, a central challenge is no longer only how to recognize or localize video content, but how to effectively adapt, extend, and design large models for diverse video understanding tasks across different temporal and semantic granularities. This challenge spans a wide range of problems, from subtle short-term action recognition and temporal localization to long-range activity understanding, event reasoning, and semantic interpretation in long videos.
See details and Call for Papers
Organisers
- Yanbin Hao, Hefei University of Technology, China
- Yi Tan, Nanyang Technological University, Singapore
- Hao Zhang, Huazhong University of Science and Technology, China
Topics of Interest
Topics include, but are not limited to:
- Micro-action and fine-grained action recognition
- Action localisation and detection
- Long-video understanding
- Video retrieval
- Multimodal large model-based video understanding, grounding, and generation
- Local-global temporal modelling for video perception
- Efficient video token/frame selection and memory mechanisms
- Video question answering, reasoning, and explanation
- Human behaviour understanding in complex video scenes
- Benchmarks and datasets for multi-granularity video understanding
- Real-world applications of video understanding
Submission Types
- Full research papers, following the standard MMM special session format
3. Multimodal Agentic Systems and Applications
Scope
Multimodal agentic systems are an emerging frontier in artificial intelligence, where large foundation models endowed with reasoning, planning, memory, and tool-use capabilities autonomously perceive, interpret, and act upon multimodal data, including text, images, video, audio, and sensor signals. Unlike traditional unimodal agents, these systems enable robust, context-aware, and generalizable decision-making across diverse real-world domains. This special session invites original contributions from the multimedia community as well as adjacent fields such as information retrieval, data mining, web, and beyond. We aim to bring together researchers across these broad AI communities to foster interaction, stimulate cross-disciplinary discussions, and catalyze collaborative efforts. We welcome both foundational advances and application-oriented research.
See details and Call for Papers
Organisers
- Yunshan Ma, Singapore Management University, Singapore
- Yujuan Ding, The Hong Kong Polytechnic University, Hong Kong SAR
- Jian Kang, Mohamed bin Zayed University of Artificial Intelligence, United Arab Emirates
- Zhaochun Ren, Leiden University, The Netherlands
- See-Kiong Ng, National University of Singapore, Singapore
Topics of Interest
Topics include, but are not limited to:
- Multimodal perception and reasoning for agentic systems
- Agentic workflow orchestration and task planning
- Multi-agent collaboration, communication, and negotiation
- Retrieval-augmented generation for multimodal agents
- Memory mechanisms and continual learning for agents
- Evaluation benchmarks and metrics for multimodal agents
- Human-agent interaction and explainable agentic systems
- Ethical considerations, safety, and alignment of autonomous multimodal agents
- Agentic systems for information retrieval, web search, and data mining
- Financial applications, including forecasting, document analysis, question answering, and decision-making
- Cybersecurity applications, including threat intelligence, automated attack and defence, and vulnerability detection
- Creative and cultural applications
- Social science and humanities applications
- Healthcare, education, robotics, and other vertical-domain applications
Submission Types
- Full research paper (12 pages)
- Survey paper (12 pages)
- Benchmark and resource paper (12 pages)
4. Reconstruction and Generation of Humans, Objects, and Scenes for Spatial Intelligence
Scope
Spatial intelligence—the ability of artificial systems to perceive, interpret, and interact with physical environments—has emerged as a cornerstone of modern intelligent systems. As an inherently interdisciplinary field, it bridges computer vision, robotics, geometric deep learning, and cognitive science to enable machines not just to observe the world, but to actively understand and manipulate it. Despite significant advancements have been witnessed in neural implicit representations, 3D reconstruction, and embodied AI, achieving human-level spatial understanding remains an open challenge—particularly human-object-environment collaborative cognitive processes.
See details and Call for Papers
This special session aims to consolidate recent progress in reconstruction and generation of humans, objects, and scenes, which serve as foundational components of spatial intelligence. Specifically, this SI seeks to address: 1) Semantically-Driven Scene Reconstruction with High Geometric Fidelity; 2) Physically Consistent Human-Object Reconstruction; 3) Spatio-Temporal Scene Generation via Human-Object-Environment Co-Cognition. By bridging these research frontiers, we aim to catalyze integrated approaches that support robust perception, generalizable generation, and interactive reasoning—laying the groundwork for next-generation spatially intelligent systems in robotics, AR/VR, simulation, and digital twins.
Organisers
- Jian Ma, Tianjin University, China
- Yuhang Ming, Hangzhou Dianzi University, China
- Buzhen Huang, Tianjin University, China
Topics of Interest
Topics include, but are not limited to:
- 3D human body and face reconstruction from monocular or multimodal inputs
- Neural implicit representations and generative modelling of humans, objects, and scenes
- NeRF, 3D Gaussian Splatting, diffusion models, and related 3D generation methods
- Object-centric reconstruction and tracking in dynamic environments
- Human-scene-object interaction modelling for spatial understanding
- Multi-view, multimodal, and cross-domain 3D perception
- Compression and representation learning for large-scale spatial data
- Digital twin modelling from visual or sensory inputs
- Scene generation for simulation and embodied AI
- Spatial intelligence for VR/AR, navigation, video generation, robotics, digital humans, remote sensing, and autonomous driving
Submission Types
- Full research papers
- Survey papers
5. Multimodal Computer Vision in Medicine: Bridging Data Modalities for Clinical Impact
Scope
To deploy high performance multimodal computer vision models into the real-world and address clinical practice requirements, many aspects should be taken into account, namely, critical technical, data-centric, and operational challenges. While these models, such as vision-language models and integrated fusion networks, showed remarkable results by combining images with clinical notes, lab tests, and genomics, they face several implementation challenges. An important issue is data fragmentation, as clinical systems store patient data in incompatible silos, and synchronizing these different modalities across different timestamps remains as a demanding question. Additionally, healthcare environments struggle with the "black box" nature of these models, since clinicians require explainability and traceability to understand exactly what happened behind the scenes, for example, which features are adopted by the machines and their influence on the diagnosis to maintain legal and ethical standards. Finally, models frequently suffer from domain shift, meaning the algorithms using only one source dataset often fail or tend to be biased when applied to other patient demographics in new clinical settings. For that reason, this special session aims to bring together researchers and practitioners to present the latest research advances in the domain and discuss the future collaborations to bridge the gap between multimodal computer vision research and clinical practice.
See details and Call for Papers
This special session directly fits with the core themes of multimedia modeling by addressing the heterogeneity of cross-modal clinical data, in order to improve final outcomes. In modern medical practice, patient data includes from many sources of media, diagnostic imaging, time-series biosignals, unstructured physician notes, and tabular laboratory results. The challenge of using all of these into a comprehensive analytical framework requires fundamental multimedia research problems, such as cross-modal representation learning and multimodal fusion. Furthermore, by emphasizing explainable AI (XAI) and human-in-the-loop evaluation, this special session bridges the gap between theoretical and multimedia modeling and real-world clinical applications, providing a vital forum to discuss artificial intelligence models that are interpretable, reliable, and safe for physicians as well as patients.
Organisers
- Klaus Schöffmann, Institute of Information Technology (ITEC), Klagenfurt University, Austria
- Thien Tan Tri Tai Truyen, MD FACC, Smidt Heart Institute, Cedars-Sinai Medical Center, Los Angeles, USA
- Yankun Wu, Premium Research Institute for Human Metaverse Medicine (PRIMe), The University of Osaka, Japan
- Vi Ngoc Tuong Ly, MD PhD, School of Computing, Dublin City University, Ireland
Topics of Interest
Topics include, but are not limited to:
- Explainable AI (XAI) and multimodal interpretability
- Interactive AI and human-in-the-loop validation
- Workflow integration and EHR/PACS compatibility
- Handling domain shifts and algorithmic bias
Submission Types
- Full research papers, following the standard MMM submission format
For inquiries regarding special sessions, please contact the special session chairs at ss@mmm2027.net.