Research map / Continual learning / Domain and prompt-based
Human Motion Forecasting in Dynamic Domain Shifts: A Homeostatic Continual Test-time Adaptation Framework
Qiongjie Cui, Huaijiang Sun, Bin Li, Jianfeng Lu, Weiqing Li
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 |
|---|---|---|---|
| update only domain-sensitive parameters (isolated update) instead of all parameters (ablation) | HoCoTTA · Human3.6M to GRAB | MPJPE at 1000ms (mm): 127.0 -> 119.2 | Improved |
| use 24 novel-domain augmentations instead of 16 (ablation) | HoCoTTA · Human3.6M to GRAB | MPJPE at 1000ms (mm): 123.4 -> 119.2 | Improved |
| increase augmentations from 24 to 32 (ablation) | HoCoTTA · Human3.6M to GRAB | MPJPE at 1000ms (mm): 119.2 -> 120.3 | No clear effect |
| set isolation quantile tau to 0.2 instead of 0.15 (ablation) | HoCoTTA · Human3.6M to GRAB | MPJPE at 1000ms (mm): 122.1 -> 119.2 | Improved |
| raise learning rate from 0.001 to 0.0015 (ablation) | HoCoTTA · Human3.6M to GRAB | MPJPE at 1000ms (mm): 119.2 -> 124.5 | Hurt |
| raise EMA momentum from 0.99 to 0.995 (ablation) | HoCoTTA · Human3.6M to GRAB | MPJPE at 1000ms (mm): 119.2 -> 122.7 | Hurt |
| HoCoTTA vs H/P-TTP on novel dataset adaptation | H/P-TTP · Human3.6M to GRAB, A1 passing | MPJPE at 400ms (mm): 45.4 -> 37.7 | Improved |
| HoCoTTA vs H/P-TTP on unseen categories and subjects | H/P-TTP · Human3.6M unseen categories | MPJPE at 1000ms (mm): 102.5 -> 99.6 | Improved |
| HoCoTTA vs H/P-TTP at short horizon on CMU Mocap unseen categories | H/P-TTP · CMU Mocap unseen categories | MPJPE at 160ms (mm): 13.4 vs 13.5 | No clear effect |
| HoCoTTA vs siMLPe backbone under standard splits | siMLPe · Human3.6M (setup-1) | MPJPE at 1000ms (mm): 109.4 -> 98.4 | Improved |
Closest papers by meaning
- ViDA: Homeostatic Visual Domain Adapter for Continual Test Time Adaptation (ICLR 2024 · similarity 0.89)
- DPCore: Dynamic Prompt Coreset for Continual Test-Time Adaptation (ICML 2025 · similarity 0.85)
- ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains (NeurIPS 2025 · similarity 0.85)
- A Versatile Framework for Continual Test-Time Domain Adaptation: Balancing Discriminability and Generalizability (CVPR 2024 · similarity 0.85)
- MoML: Online Meta Adaptation for 3D Human Motion Prediction (CVPR 2024 · similarity 0.85)
- ScaMo: Exploring the Scaling Law in Autoregressive Motion Generation Model (CVPR 2025 · similarity 0.84)
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.
