The 1st Workshop on
Human-Centered Multimodal Intelligence in the Wild

The emergence of multimodal foundation models (MLLMs) has catalyzed a paradigm shift in human-centered vision. While traditional systems often struggle with the open-world complexity of human behavior, the integration of large-scale pre-training with cross-modal reasoning offers paths toward truly adaptive intelligence. This workshop addresses the critical transition from constrained benchmarks to in-the-wild deployment, where visual intelligence must navigate the ambiguity of human intent and the noise of unstructured environments.

The Human-Centered Multimodal Intelligence in the Wild (HCMIW) workshop bridges the gap between theoretical MLLMs and practical, human-aligned applications. We focus on the synergy between robust perception and sophisticated reasoning, aiming to redefine how vision systems interpret human presence across heterogeneous sensors and dynamic contexts.

To address these challenges, HCMIW invites contributions across five critical axes:

• MLLMs for Human-Centric Tasks: Novel paradigms for detection, tracking, and activity understanding, with emphasis on efficient adaptation and from-scratch training tailored for real-world constraints.
• Perception in Adverse Conditions: Advancing robustness against in-the-wild variables, including extreme lighting, rapid motion, severe occlusions, and complex human-object interactions.
• Heterogeneous Sensor Fusion: Beyond RGB—leveraging infrared, event-based cameras, LiDAR, and thermal sensing to build multi-spectral understanding of human activity.
• Generalization & Domain Transfer: Strategies for bridging the gap between massive general-purpose datasets and specialized, high-stakes human-centric domains.
• Next-Generation Benchmarking: Developing wild-first datasets and evaluation metrics that prioritize human-facing performance, safety, and reliability over static accuracy.