Research map / Video / Video generation
InterGen: Diffusion-Based Multi-human Motion Generation Under Complex Interactions
Han Liang, Wenqian Zhang, Wenxuan Li, Jingyi Yu A, Lan Xu
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 |
|---|---|---|---|
| use non-canonical shared-frame representation instead of canonical (ablation) | InterGen · InterHuman test | FID: 11.322 -> 5.918 | Improved |
| use non-canonical shared-frame representation instead of canonical (ablation) | InterGen · InterHuman test | R-Precision Top-1: 0.134 -> 0.371 | Improved |
| add mutual attention conditioning instead of concatenation (ablation) | InterGen · InterHuman test | FID: 6.273 -> 5.918 | Improved |
| add mutual attention conditioning instead of concatenation (ablation) | InterGen · InterHuman test | R-Precision Top-1: 0.303 -> 0.371 | Improved |
| add weights sharing between the two denoisers (ablation) | InterGen · InterHuman test | FID: 8.059 -> 5.918 | Improved |
| add weights sharing between the two denoisers (ablation) | InterGen · InterHuman test | R-Precision Top-1: 0.153 -> 0.371 | Improved |
| add masked joint distance map (DM) loss (ablation) | InterGen · InterHuman test | FID: 6.653 -> 5.918 | Improved |
| add relative orientation (RO) loss (ablation) | InterGen · InterHuman test | FID: 6.311 -> 5.918 | Improved |
| add regularization loss schedule t<=0.7T instead of none (ablation) | InterGen · InterHuman test | FID: 7.862 -> 5.918 | Improved |
| apply regularization loss at all timesteps instead of t<=0.7T (ablation) | InterGen · InterHuman test | FID: 5.918 -> 7.037 | Hurt |
| raise regularization threshold from 0.7T to 0.8T (ablation) | InterGen · InterHuman test | FID: 5.918 -> 6.521 | Hurt |
| replace CLIP-ViT-L/14 text encoder with LLaMA-7B (ablation) | InterGen · InterHuman test | FID: 5.918 -> 5.865 | Mixed |
Later papers that compared with this method
19 later accepted papers report InterGen as a baseline in their results tables; 18 of them report a better number on most of the results they share with it.
- InterEdit: Navigating Text-Guided Multi-Human 3D Motion Editing (ECCV 2026 · better on 7 of 7 shared results)
- Disentangled Hierarchical VAE for 3D Human-Human Interaction Generation (ICLR 2026 · better on 4 of 4 shared results)
- ReactDance: Hierarchical Representation for High-Fidelity and Coherent Long-Form Reactive Dance Generation (ICLR 2026 · better on 4 of 4 shared results)
- Text2Interact: High-Fidelity and Diverse Text-to-Two-Person Interaction Generation (ICLR 2026 · better on 4 of 4 shared results)
- Interact2Ar: Full-Body Human-Human Interaction Generation via Autoregressive Diffusion Models (CVPR 2026 · better on 4 of 4 shared results)
- InterCMDM: Block-Causal Diffusion for Autoregressive Human Interaction Generation (ECCV 2026 · better on 3 of 3 shared results)
- ARMFlow: AutoRegressive MeanFlow for Online 3D Human Reaction Generation (CVPR 2026 · better on 2 of 2 shared results)
- VersatileMotion: A Unified Framework for Motion Synthesis and Comprehension (ECCV 2026 · better on 2 of 2 shared results)
11 more not listed.
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.
- TEMOS — TEMOS: Generating Diverse Human Motions from Textual Descriptions (ECCV 2022) · better on 4 of 4 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
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.91)
- InterDreamer: Zero-Shot Text to 3D Dynamic Human-Object Interaction (NeurIPS 2024 · similarity 0.90)
- InterControl: Zero-shot Human Interaction Generation by Controlling Every Joint (NeurIPS 2024 · similarity 0.90)
- AMD: Autoregressive Motion Diffusion (AAAI 2024 · similarity 0.90)
- HOIAnimator: Generating Text-prompt Human-object Animations using Novel Perceptive Diffusion Models (CVPR 2024 · similarity 0.89)
- InterMask: 3D Human Interaction Generation via Collaborative Masked Modeling (ICLR 2025 · 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.
