Research map / Video / Video generation
TMD-Bench: A Multi-Level Evaluation Paradigm for Music–Dance Co-Generation
Xiaoda Yang, Majun Zhang, Changhao Pan, Nick Huang, Yang Yuguang, Fan Zhuo, Pengfei Zhou, Jin Zhou, Sizhe Shan, Shan Yang, Miles Yang, Yang You and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train a unified joint music-dance diffusion model on rhythm-aligned data (RhyJAM) instead of using open-source audio-video models | RhyJAM vs Ovi · TMD-Bench alignment | VBCS / avg. alignment: VBCS 0.30 -> 0.50; Avg. 0.48 -> 0.59 | Improved |
| use a unified model instead of cascaded text-to-music-to-dance pipeline (ACE-Step+Wan 2.5) | RhyJAM vs Ace+Wan · TMD-Bench alignment | Avg. alignment: 0.51 -> 0.59 | Improved |
| use a cascaded text-to-music-to-dance pipeline instead of end-to-end | ACE-Step+Wan 2.5 · TMD-Bench alignment | ABHS: ABHS 0.25 (vs Sora 2 0.16) | Improved |
| commercial closed-source model (Sora 2) vs RhyJAM on rhythmic alignment | Sora 2 · TMD-Bench alignment | Avg. alignment: 0.59 vs 0.59 | No clear effect |
| closed-source models versus RhyJAM on video quality | RhyJAM · TMD-Bench video | Avg. video score: RhyJAM 0.71 vs Veo 3 0.78 / Wan 2.6 0.79 | Hurt |
| RhyJAM versus other open-source models on music rhythm semantics | RhyJAM · TMD-Bench music | Rhythm & Groove: 0.59 (best); next best 0.57 | Improved |
| use algorithmic physical metrics vs MLLM judge for alignment | Sora 2 · TMD-Bench alignment | ABHS vs high-level Align.: ABHS 0.16 vs Align. 0.85 | Mixed |
| RhyJAM versus closed-source on compositional music attributes (genre, ambiance, functional scenes) | RhyJAM · TMD-Bench music | Genre/Amb./Func.: Func. 0.07 vs Ovi/Sora 2 0.17 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- JavisDiT — JavisDiT: Joint Audio-Video Diffusion Transformer with Hierarchical Spatio-Temporal Prior Synchronization (ICLR 2026) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OmniDance: Multimodal Driven Dance Video Generation with Large-scale Internet Data (ECCV 2026 · similarity 0.92)
- Text Dictates, Music Decorates: Energy-based Attention for Editable Dance Generation (ECCV 2026 · similarity 0.92)
- DanceEditor: Towards Iterative Editable Music-driven Dance Generation with Open-Vocabulary Descriptions (ICCV 2025 · similarity 0.90)
- MDD: A Dataset for Text-and-Music Conditioned Duet Dance Generation (ICCV 2025 · similarity 0.90)
- MEGADance: Mixture-of-Experts Architecture for Genre-Aware 3D Dance Generation (NeurIPS 2025 · similarity 0.90)
- DanceAnyWay: Synthesizing Beat-Guided 3D Dances with Randomized Temporal Contrastive Learning (AAAI 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
