9 June 2026

Our group has presented a new paper at the 20th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2026) in Kyoto, Japan. In this paper, we explore the use of image-to-video models to generate deictic gestures at scale. Our results demonstrate strong potential for image-to-video models as a scalable, low-effort approach to creating high-fidelity gesture data for HRI and related fields. The synthetic videos provide meaningful variability and novelty in the gesture dataset, which is useful for downstream detection, recognition, and classification tasks. These findings suggest that generative video pipelines are an important tool for future gesture and HRI research, helping accelerate dataset creation and enabling more diverse interaction systems that support non-verbal communication.

You can find more details here: Paper - Dataset