Summary
Inter-X++ is a large-scale benchmark for human-human interaction (HHI) perception and synthesis, addressing fundamental limitations of prior datasets such as low-fidelity motion, missing dexterous hand detail, and scarce multimodal annotations. Collected with a novel hybrid motion capture system, it provides 11,388 high-fidelity interaction sequences totaling over 8.1 million frames with precise whole-body motion and detailed finger joints. The dataset is enriched with multi-faceted annotations: hierarchical fine-grained text descriptions, interaction categories, causal interaction order, subject relationships and personalities, vertex-level contact maps, and physics-based regularization constraints. Building on these annotations, the authors establish a unified testbed spanning four categories of downstream tasks, symmetrically covering both generation and perception paradigms. They also propose OpenHHI, a unified HHI representation and modeling framework that jointly optimizes interaction reconstruction and semantic understanding. Extensive experiments show OpenHHI achieves state-of-the-art performance on both generation and perception tasks. The paper is available on arXiv (2608.20312).
Paper Overview
Research Area: Computer Vision (CV)
Authors: Liang Xu, Chengqun Yang, Zili Lin, Xintao Lv
Published: 2026-08-22
arXiv: 2608.20312
Abstract
The ability to perceive and synthesize human-human interactions is fundamental to developing intelligent digital human systems. However, existing datasets and modeling methods suffer from fundamental limitations: low-fidelity kinematics, lack of dexterous hand gestures, and severely insufficient multimodal annotations.
This paper introduces Inter-X++, a comprehensive, large-scale benchmark captured via a novel hybrid motion capture system. It provides:
- 11,388 high-fidelity interaction sequences with over 8.1 million frames
- Precise whole-body motion with detailed finger joints
The dataset is further enriched with multi-faceted annotations, including:
- Hierarchical fine-grained text descriptions
- Interaction categories
- Causal interaction order
- Subject relationships and personalities
- Vertex-level contact maps and physics-based regularization constraints
Leveraging these annotations, the authors establish a
unified testbed covering four categories of downstream tasks, symmetrically spanning both generation and perception paradigms.
They also propose OpenHHI, a unified HHI representation and modeling framework that jointly optimizes interaction reconstruction and semantic understanding. Extensive experiments demonstrate that OpenHHI achieves state-of-the-art performance on both generation and perception tasks.
---
*Auto-collected on 2026-08-22*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178633819