Research map / 3D vision / Human pose estimation
Few-Shot Human Motion Prediction via Meta-Learning
Liang-Yan Gui, Yu-Xiong Wang, Deva Ramanan, Jose M. F. Moura
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 |
|---|---|---|---|
| add meta-learned generic initialization (MAML) to adaptation-only model (ablation) | PAML · H3.6M Walking 80ms | mean angle error: 0.39 -> 0.35 | Improved |
| add adaptation network H to init-only model (ablation) | PAML · H3.6M Smoking 80ms | mean angle error: 0.53 -> 0.39 | Improved |
| add adaptation network H to init-only model (ablation) | PAML · H3.6M Discussion 80ms | mean angle error: 0.54 -> 0.41 | Improved |
| add meta-learned initialization to adaptation-only model (ablation) | PAML · H3.6M Eating 80ms | mean angle error: 0.50 -> 0.36 | Improved |
| meta-learning with PAML instead of fine-tuning transfer | residual sup. · H3.6M Walking 80ms | mean angle error: 0.44 -> 0.35 | Improved |
| PAML instead of multi-task learning | residual sup. · H3.6M Smoking 80ms | mean angle error: 0.71 -> 0.39 | Improved |
| action-agnostic instead of action-specific training from scratch | residual sup. · H3.6M Walking 80ms | mean angle error: 1.90 -> 1.78 | Improved |
| use 3-layer Leaky ReLU H instead of affine 1-layer H (ablation) | PAML · H3.6M Walking 80ms | mean angle error: 0.39 -> 0.35 | Improved |
| use Leaky ReLU instead of ReLU in 3-layer H (ablation) | PAML · H3.6M Walking 80ms | mean angle error: 0.37 -> 0.35 | Improved |
| go from 3 to 4 layers in H (ablation) | PAML · H3.6M Walking 80ms | mean angle error: 0.35 -> 0.36 | Hurt |
| PAML instead of MAML in few-shot classification | PAML · mini-ImageNet 5-way 1-shot | accuracy: 48.70% -> 53.26% | Improved |
| PAML instead of MAML in few-shot classification | PAML · mini-ImageNet 5-way 5-shot | accuracy: 63.11% -> 68.19% | Improved |
Closest papers by meaning
- MoML: Online Meta Adaptation for 3D Human Motion Prediction (CVPR 2024 · similarity 0.89)
- SimpliHuMoN: Simplifying Human Motion Prediction (NeurIPS 2026 · similarity 0.88)
- Generating Smooth Pose Sequences for Diverse Human Motion Prediction (ICCV 2021 · similarity 0.87)
- A Stochastic Conditioning Scheme for Diverse Human Motion Prediction (CVPR 2020 · similarity 0.87)
- Investigating Pose Representations and Motion Contexts Modeling for 3D Motion Prediction (TPAMI 2022 · similarity 0.87)
- LAL: Enhancing 3D Human Motion Prediction with Latency-aware Auxiliary Learning (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.
