Research map / Video / Action recognition
MultiAct: Long-Term 3D Human Motion Generation from Multiple Action Labels
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| generate transition and action motion jointly from previous motion and current action | MultiAct · BABEL generated transitions | FID_test: 5.85$^{\pm0.059}$ -> 0.67^{\pm0.027} | Helped |
| add separate latent embeddings for previous and current motion (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 1.43^{\pm0.073} -> 0.97^{\pm0.015} | Helped |
| do not give the current action label to PrevNet (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 1.45^{\pm0.092} -> 0.97^{\pm0.015} | Helped |
| use only the last frames rather than the whole previous motion as PrevNet input (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 1.10^{\pm0.092} -> 0.97^{\pm0.015} | Helped |
| use face-front canonicalization rather than no canonicalization (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 1.12^{\pm0.026} -> 0.97^{\pm0.015} | Helped |
| use face-front rather than zero-canonicalization (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | Acc. top1: 0.64^{\pm0.018} -> 0.64^{\pm0.009} | No clear effect |
| use 4 rather than 2 frames as PrevNet input (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 1.04^{\pm0.032} -> 0.97^{\pm0.015} | Helped |
| use 4 rather than 3 frames as PrevNet input (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 0.98^{\pm0.037} -> 0.97^{\pm0.015} | Helped |
| use 4 rather than 5 frames as PrevNet input (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 1.02^{\pm0.067} -> 0.97^{\pm0.015} | Helped |
| use 4 rather than 6 frames as PrevNet input (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 0.95^{\pm0.087} -> 0.97^{\pm0.015} | No clear effect |
| use Transformer rather than LSTM backbone (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 1.99^{\pm0.052} -> 0.97^{\pm0.015} | Helped |
| use Transformer rather than GRU backbone (ablation) | MACVAE · BABEL unseen previous motions, single-step action motion | FID_test: 2.92^{\pm0.104} -> 0.97^{\pm0.015} | Helped |
Closest papers by meaning
- AMD: Autoregressive Motion Diffusion (AAAI 2024 · similarity 0.87)
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.87)
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.87)
- 3DTrajMaster: Mastering 3D Trajectory for Multi-Entity Motion in Video Generation (ICLR 2025 · similarity 0.87)
- MotionGPT: Finetuned LLMs Are General-Purpose Motion Generators (AAAI 2024 · similarity 0.87)
- Move-in-2D: 2D-Conditioned Human Motion Generation (CVPR 2025 · similarity 0.86)
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.