---
title: README
emoji: 🌖
colorFrom: red
colorTo: blue
sdk: static
pinned: false
---
# Human–Scene Interaction Workshop

Data, benchmarks, and demos for the Workshop on Human–Scene Interaction Workshop.


## Challenge: Scene-Aware Referential Gesture Generation at ECCV26


Given speech, a 3D target coordinate, and a scene, generate SMPL-X pointing gestures that are temporally aligned, spatially accurate, and referentially grounded. Submissions are scored on three axes: temporal alignment, spatial accuracy, and referent recall.

📄 **[Challenge Paper](./hsi_challenge_release_v1.pdf)** · 📦 **[Data & Baseline](https://huggingface.co/datasets/hsi-workshop/referential-gesture-challenge)** · 🎯 **[Interactive Demo](https://huggingface.co/spaces/mm-conv-scene-gesture/referential-gesture-challenge-demo)**

## Resources

| Resource | Description |
|----------|-------------|
| [MM-Conv](https://huggingface.co/datasets/mm-conv-scene-gesture/demo-data) | ~2K pointing clips from naturalistic VR dialogue with 3D scene graphs |
| [SGS-HSI](https://huggingface.co/datasets/mm-conv-scene-gesture/demo-data) | 1,138 synthetic single-target pointing clips |
| OmniControl-PT baseline | Reference baseline (code & weights coming soon) |

## Timeline

| Milestone | Date |
|-----------|------|
| Challenge opens | May 5, 2026 |
| Submission deadline | July 7, 2026 |
| Results announced | July 31, 2026 |
| Workshop | October 2026 |

## Organizers

Jonas Beskow (KTH) · Rishabh Dabral (MPI) · Anna Deichler (KTH) · Fethiye Irmak Doğan (Cambridge) · Anindita Ghosh (MPI) ·