Training data for the Referential Gesture Challenge at the 1st HSI Workshop (ECCV 2026). Contains ~2K pointing gesture clips from MM-Conv with synchronized speech, motion, and 3D scene graphs and ~1K pointing gesture from synthetic dataset with synchronized speech, motion, and 3D targets.
Current top-level structure:
data/
annotations/ # released training annotations
audio/ # released training audio… See the full description on the dataset page:
https://huggingface.co/datasets/hsi-workshop/referential-gesture-challenge.
