Go to Zero: Towards Zero-shot Motion Generation with Million-scale Data

2025-07-09

AI Summary

This work addresses three key challenges in text-driven zero-shot human motion generation: limited generalization capability, insufficient training data scale, and the absence of standardized evaluation protocols. Methodologically, we introduce MotionMillion—the largest high-quality motion dataset to date (2 million sequences)—and propose MotionMillion-Eval, the first comprehensive benchmark specifically designed for zero-shot motion generation evaluation. Leveraging an efficient annotation pipeline and a scalable architecture, we train a 7B-parameter end-to-end generative model. Our contributions include substantial improvements in zero-shot generalization across domains and for complex compositional motions, achieving state-of-the-art performance on multiple novel benchmarks. Furthermore, we publicly release both the code and dataset, establishing a robust foundation and a unified evaluation standard for the community.

Technology Category

• Computer Vision: Motion & Tracking
• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Generating diverse and natural human motion sequences based on textual descriptions constitutes a fundamental and challenging research area within the domains of computer vision, graphics, and robotics. Despite significant advancements in this field, current methodologies often face challenges regarding zero-shot generalization capabilities, largely attributable to the limited size of training datasets. Moreover, the lack of a comprehensive evaluation framework impedes the advancement of this task by failing to identify directions for improvement. In this work, we aim to push text-to-motion into a new era, that is, to achieve the generalization ability of zero-shot. To this end, firstly, we develop an efficient annotation pipeline and introduce MotionMillion-the largest human motion dataset to date, featuring over 2,000 hours and 2 million high-quality motion sequences. Additionally, we propose MotionMillion-Eval, the most comprehensive benchmark for evaluating zero-shot motion generation. Leveraging a scalable architecture, we scale our model to 7B parameters and validate its performance on MotionMillion-Eval. Our results demonstrate strong generalization to out-of-domain and complex compositional motions, marking a significant step toward zero-shot human motion generation. The code is available at https://github.com/VankouF/MotionMillion-Codes.

Problem

• Achieving zero-shot generalization in text-to-motion generation
• Addressing limited dataset size for training motion models
• Developing a comprehensive benchmark for motion generation evaluation

Innovation

• Largest human motion dataset with 2M sequences
• Comprehensive benchmark for zero-shot evaluation
• Scalable 7B model for generalization performance