Call for participation: GENEA Challenge 2026 on speech-driven gesture generation Challenge period: Until June 28 Location: Interactive Social Agents Workshop @ ECCV 2026, Malmö, Sweden Website: https://genea-workshop.github.io/2026/challenge/ *********************************************************************
Motivation
*********************
It is difficult to reproduce co-speech gesture-generation systems trained by other people, and each publication often uses its own evaluation methodology that often deviates from best practices. This makes it difficult to assess the state of the art in the field.
To better compare and understand methods for gesture generation and evaluation, we are continuing the GENEA (Generation and Evaluation of Non-verbal Behaviour for Embodied Agents) challenges, previously run in 2020, 2022, and 2023, wherein gesture-generation approaches from different authors are evaluated side by side on a high-quality avatar in a set of large-scale crowdsourced studies.
Held in conjunction with the Interactive Social Avatars workshop at ECCV 2026, this year’s challenge invites submissions of both novel models and previously published systems. New research may be submitted as a formal workshop paper, while published systems should be accompanied by a structured technical description document.
Data and tasks
*********************
The challenge will be using the Seamless Interaction dataset from Meta FAIR. The challenge will perform four evaluations:
-
Core task 1: Motion realism
-
Core task 2: Speech-gesture alignment
-
Optional task 1: Semantic appropriateness for a specific spoken word (using the Seamless Interaction spoken gesture game subset)
-
Optional task 2: Dyadic appropriateness for/alignment to interlocutor behaviour
The evaluations will be based on the methodologies of prior challenges and the recent GENEA Leaderboard paper (CVPR Findings 2026). Challenge submissions will also be placed on the ongoing GENEA living leaderboard, which will add a new leaderboard on the Seamless Interaction dataset following this challenge.
Timeline
*********************
Apr. 15 – Official challenge soft launch
June 21 – Test-set inputs released to participants
June 28 – Deadline for participants to submit generated motion
July 10 – Evaluation results released to participants
July 15 – Paper submission deadline
Aug. 3 – Author notifications
Aug. 8 – Camera-ready paper deadline
Sept. 8 or 9 – Challenge presentations at ECCV Workshop
Registration
*********************
To register for the challenge, please follow this link:
https://forms.gle/MhorH9X3PmLsbQ5e8
Submission format
*********************
Challenge participants are required to make sure that there is a reproducible description of their system available, so that everyone can learn from the challenge by connecting what was done to the resulting system performance. We provide three pathways to do this:
-
Submit a full, 8-page paper for publication at the ECCV workshop
-
Submit a 4-page non-archival technical report, enabling authors to publish a full paper of their work at another conference of their choice
-
Participants that enter a previously published/described method into the challenge may submit a 1-page non-archival extended abstract that describe all changes made to train their system for the challenge
We look forward to seeing your contributions to the challenge!
Organising committee
********************************
Rajmund Nagy – KTH Royal Institute of Technology, Sweden
Hendric Voß – Bielefeld University, Germany
Mihail Tsakov – Independent researcher, the Netherlands
Teodor Nikolov – Motorica AB, Sweden
Silvia Arellano García – KTH Royal Institute of Technology, Sweden
Taras Kucherenko – SEED, Electronic Arts, Sweden
Youngwoo Yoon – ETRI, South Korea
Gustav Eje Henter – KTH Royal Institute of Technology, Sweden / Motorica AB, Sweden
Questions about the challenge can be sent to genea-challenge@googlegroups.com.




May 5th, 2026
Daniela Lopez de Luise
Posted in