HyperMotionX: The Dataset and Benchmark with DiT-Based Pose-Guided Human Image Animation of Complex Motions

TL;DR

This paper introduces HyperMotionX, a new dataset and benchmark for evaluating pose-guided human image animation of complex motions, along with a DiT-based model that enhances animation quality in dynamic scenarios.

Contribution

It presents a novel DiT-based baseline with a spatial low-frequency enhancement module and introduces the HyperMotionX dataset and benchmark for complex human motion animation evaluation.

Findings

Improved structural stability in dynamic human motion sequences

Enhanced appearance consistency in animations

Effective evaluation of complex motion animation models

Abstract

Recent advances in diffusion models have significantly improved conditional video generation, particularly in the pose-guided human image animation task. Although existing methods are capable of generating high-fidelity and time-consistent animation sequences in regular motions and static scenes. However there are still obvious limitations when facing complex human body motions that contain highly dynamic, non-standard motions, and the lack of a high-quality benchmark for evaluation of complex human motion animations. To address this challenge, we propose a concise yet powerful DiT-based human animation generation baseline and design spatial low-frequency enhanced RoPE, a novel module that selectively enhances low-frequency spatial feature modeling by introducing learnable frequency scaling. Furthermore, we introduce the Open-HyperMotionX Dataset and HyperMotionX Bench, which provide high-quality human pose annotations and curated video clips for evaluating and improving pose-guided human image animation models under complex human motion conditions. Our method significantly improves structural stability and appearance consistency in highly dynamic human motion sequences. Extensive experiments demonstrate the effectiveness of our dataset and proposed approach in advancing the generation quality of complex human motion image animations. The codes, model weights, and dataset have been made publicly available at https://vivocameraresearch.github.io/hypermotion/