← ArchivePaper2023
Example-based Motion Synthesis via Generative Motion Matching
Weiyu Li, Xuelin Chen, Peizhuo Li, Olga Sorkine-Hornung, Baoquan Chen
Generative motion matching framework combining motion database search with generative models for smooth diverse motion synthesis.
Abstract
We present GenMM, a generative model that "mines" as many diverse motions as possible from a single or few example sequences. In stark contrast to existing data-driven methods, which typically require long offline training time, are prone to visual artifacts, and tend to fail on large and complex skeletons, GenMM inherits the training-free nature and the superior quality of the well-known Motion Matching method. GenMM can synthesize a high-quality motion within a fraction of a second, even with highly complex and large skeletal structures. At the heart of our generative framework lies the generative motion matching module, which utilizes the bidirectional visual similarity as a generative cost function to motion matching, and operates in a multi-stage framework to progressively refine a random guess using exemplar motion matches. In addition to diverse motion generation, we show the versatility of our generative framework by extending it to a number of scenarios that are not possible with motion matching alone, including motion completion, key frame-guided generation, infinite looping, and motion reassembly.
How to read this
- Category
- Method: example-based motion synthesis (training-free generative framework)
- Contributions
- GenMM, a training-free generative framework that mines diverse motions from one or a few example sequences
- A generative motion matching module using bidirectional visual similarity as a generative cost in a multi-stage refinement
- Versatile extensions: motion completion, keyframe-guided generation, infinite looping, and motion reassembly
- Context
- Combines the training-free, high-quality nature of Motion Matching (clavet-motionmatching-2016) with generative synthesis, positioned as an alternative to learned data-driven approaches like the Human Motion Diffusion Model (tevet-mdm-2022).Builds on: Motion Matching and The Road to Next-Gen Animation · Human Motion Diffusion Model
- Correctness
- Works from very limited examples and claims fast, artifact-resistant synthesis on complex skeletons; keep in mind it mines variation from the given exemplars, so output diversity is bounded by what the example sequences contain.
- Clarity
- Fairly accessible if you know motion matching; a first pass conveys the analogy, a second pass clarifies the bidirectional-similarity cost and the multi-stage scheme.
- How to read it
- Anchor on the contrast with both classic motion matching and learned diffusion; second pass is worth it to understand the generative cost function and how the multi-stage refinement avoids artifacts.
Built upon by
Nothing yet.
Related work
- Neural Animation Layering for Synthesizing Martial Arts Movements 2021 / SIGGRAPH
- Automated Extraction and Parameterization of Motions in Large Data Sets 2004 / SIGGRAPH
- Learning Robust and Scalable Motion Matching with Lipschitz Continuity and Sparse Mixture of Experts 2023 / MIG
- Interactive Character Control with Auto-Regressive Motion Diffusion Models 2024 / TOG
Keywords
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →