Research map / Label-efficient and robust learning / Test-time adaptation
Test-time Personalizable Forecasting of 3D Human Poses
Qiongjie Cui, Huaijiang Sun, Jianfeng Lu, Weiqing Li, Bin Li, Hongwei Yi, Haofan Wang
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 H/P-TTP to SPGSN on H3.6M subject 5 | SPGSN · H3.6M setup-1.1 (1000ms) | MPJPE (mm): 109.6 -> 103.7 | Improved |
| add H/P-TTP to MSR on GRAB | MSR · GRAB setup-1.2 (1000ms) | MPJPE (mm): 178.6 -> 147.3 | Improved |
| add H/P-TTP to SPGSN on unseen subjects of GRAB | SPGSN · GRAB unseen subjects (average, 1000ms) | MPJPE (mm): 155.9 -> 135.5 | Improved |
| add H/P-TTP to SPGSN on unseen subjects of H3.6M | SPGSN · H3.6M unseen subjects (average, 1000ms) | MPJPE (mm): 107.0 -> 102.8 | Improved |
| add H/P-TTP to SPGSN on H3.6M unseen subjects | SPGSN · H3.6M unseen subjects (average, 1000ms) | PCK@150mm (%): 66.4 -> 70.9 | Improved |
| add H/P-TTP to PGBIG on H3.6M unseen subjects | PGBIG · H3.6M unseen subjects (average, 1000ms) | PCK@150mm (%): 65.7 -> 68.2 | Improved |
| add H/P-TTP to PGBIG on GRAB unseen subjects | PGBIG · GRAB unseen subjects (average, 1000ms) | PCK@150mm (%): 63.4 -> 66.4 | Improved |
| add implicit and explicit augmenters (ablation) | H/P-TTP (SPGSN) · H3.6M setup-2.1 (1000ms) | MPJPE (mm) | Improved |
| use dynamic test-time learning rate (ablation) | H/P-TTP (SPGSN) · H3.6M setup-2.1 (1000ms) | MPJPE (mm) | Improved |
| increase TTP gradient steps from 1 to 3 (ablation) | H/P-TTP (SPGSN) · H3.6M setup-2.1 (1000ms) | MPJPE (mm): 102.8 -> 104.8 | Hurt |
| change initial learning rate beta from 0.001 (ablation) | H/P-TTP (SPGSN) · H3.6M setup-2.1 (1000ms) | MPJPE (mm): 102.8 -> 103.4 (beta 0.0005), 103.1 (beta 0.0015) | No clear effect |
| change number of augmentations H=K from 8 to 6, 12 or 16 (ablation) | H/P-TTP (SPGSN) · H3.6M setup-2.1 (1000ms) | MPJPE (mm): 102.8 -> 109.2 (6), 107.4 (12), 105.5 (16) | Hurt |
Closest papers by meaning
- Test-Time Personalization with a Transformer for Human Pose Estimation (NeurIPS 2021 · similarity 0.87)
- PoseGPT: Quantization-Based 3D Human Motion Generation and Forecasting (ECCV 2022 · similarity 0.86)
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.84)
- MotionMap: Representing Multimodality in Human Pose Forecasting (CVPR 2025 · similarity 0.84)
- SimpliHuMoN: Simplifying Human Motion Prediction (NeurIPS 2026 · similarity 0.84)
- A Dual-Augmentor Framework for Domain Generalization in 3D Human Pose Estimation (CVPR 2024 · similarity 0.83)
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.
