HyperMotionX Bench
HyperMotionX Bench
One-line summary: 2025 public pose-guided human-image-animation bench for complex motions, released in the same paper as the authors' DiT baseline.
What it is
Open-HyperMotionX Dataset plus HyperMotionX Bench: human pose annotations and curated clips for evaluating pose-guided animation under highly dynamic, non-standard motion. Same paper introduces a DiT baseline with spatial low-frequency enhanced RoPE. From 2026-08-21-academic-research-independent-avatar-generation-benchmarks.
Why it matters to ai-video-generation
It is the only retrieved named pose-driven animation bench. That fills part of the "is there any human-animation eval?" gap. It does not close independent-avatar-benchmarks: the bench is author-tied, and the abstract does not describe an independent multi-model leaderboard of animate-anyone-2 / mimicmotion / wan-animate.
Key facts
- Task: pose-guided human image animation under complex motion.
- Public code, weights, and dataset claimed (project page named in the abstract; not fetched).
- Citations in the 2026-08-21 Consensus retrieval: 6.
Strengths
- Diagnoses a missing complex-motion eval set and ships one.
Weaknesses
- Same paper as the model being improved — author-reported gains on their own bench.
- Not described as a live third-party leaderboard.