# CoMPAS3D: A Dataset and Benchmark for Interactive Motion

arXiv: https://arxiv.org/abs/2507.19684 · HTML: https://mlsanity.com/paper/2507.19684/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 19 of its 37 references that are in the map.
- First posted on arXiv 25 Jul 2025.

## Builds on (most cited first, 19 of 19)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution](https://mlsanity.com/paper/2409.12191/) (2024)
- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [Expressive Body Capture: 3D Hands, Face, and Body from a Single Image](https://mlsanity.com/paper/1904.05866/) (2019)
- [T2M-GPT: Generating Human Motion from Textual Descriptions with Discrete Representations](https://mlsanity.com/paper/2301.06052/) (2023)
- [NTU RGB+D 120: A Large-Scale Benchmark for 3D Human Activity Understanding](https://mlsanity.com/paper/1905.04757/) (2019)
- [Bailando: 3D Dance Generation by Actor-Critic GPT with Choreographic Memory](https://mlsanity.com/paper/2203.13055/) (2022)
- [InterGen: Diffusion-based Multi-human Motion Generation under Complex Interactions](https://mlsanity.com/paper/2304.05684/) (2023)
- [Single-Shot Multi-Person 3D Pose Estimation From Monocular RGB](https://mlsanity.com/paper/1712.03453/) (2017)
- [Inter-X: Towards Versatile Human-Human Interaction Analysis](https://mlsanity.com/paper/2312.16051/) (2023)
- [Hi4D: 4D Instance Segmentation of Close Human Interaction](https://mlsanity.com/paper/2303.15380/) (2023)
- [Multi-Person Extreme Motion Prediction](https://mlsanity.com/paper/2105.08825/) (2021)
- [ReMoS: 3D Motion-Conditioned Reaction Synthesis for Two-Person Interactions](https://mlsanity.com/paper/2311.17057/) (2023)
- [ReGenNet: Towards Human Action-Reaction Synthesis](https://mlsanity.com/paper/2403.11882/) (2024)
- [Duolando: Follower GPT with Off-Policy Reinforcement Learning for Dance Accompaniment](https://mlsanity.com/paper/2403.18811/) (2024)
- [InterDance:Reactive 3D Dance Generation with Realistic Duet Interactions](https://mlsanity.com/paper/2412.16982/) (2024)
- [Reconstructing Three-Dimensional Models of Interacting Humans](https://mlsanity.com/paper/2308.01854/) (2023)
- [Synergy and Synchrony in Couple Dances](https://mlsanity.com/paper/2409.04440/) (2024)
- [The MI-Motion Dataset and Benchmark for 3D Multi-Person Motion Prediction](https://mlsanity.com/paper/2306.13566/) (2023)

## Built on by (most cited first, 2 of 2)

- [SalsaAgent: A multimodal embodied language model for interactive dance generation](https://mlsanity.com/paper/2605.29219/) (2026)
- [WeLike2Party! In-Context Motion Transfer for Multi-Human Image Animation](https://mlsanity.com/paper/2609.36937/) (2026)

## Abstract (arXiv)

Socially interactive humanoid robots must engage with humans through their bodies, adapting in real time to a partner's movement, intent, and abilities. This requires models that understand not just how bodies move, but what movement means in a shared social context. Yet evaluation frameworks for interactive motion generation do not measure whether generated follower motion is legible within a shared movement vocabulary, nor whether it is appropriate to the partner's proficiency level. This gap has two causes: existing frameworks rely on kinematic metrics such as FID and beat alignment that cannot measure either property, and existing datasets lack the move annotations and proficiency variation needed. Salsa is well-suited as an evaluation domain: improvised, dyadic, and governed by a move vocabulary and judging criteria covering timing, musicality, technique, difficulty, partnering, and originality. We present CoMPAS3D, a motion capture dataset of improvised partner salsa paired with an evaluation framework covering kinematic quality, two objective metrics (move legibility and proficiency appropriateness), and six competition-based subjective dimensions. The dataset includes 3 hours of improvisation by 18 dancers spanning beginner, intermediate, and professional levels, with over 2,800 expert-annotated segments covering move types, errors, and stylistic elements. We define three benchmarks: move classification (analogous to transcription), proficiency estimation (fluency assessment), and follower generation (dialogue response). Fine-tuned vision-language models perform strongly on objective metrics applied to ground-truth motion sequences. Applied to Duolando and InterGen, the metrics reveal failures that kinematic metrics miss. Human evaluations confirm the gap between generated and ground-truth motion. CoMPAS3D, annotations, benchmark code, and baseline results are publicly available.