Research map / 3D vision / Human pose estimation
Unified Pose Sequence Modeling
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 |
|---|---|---|---|
| add dynamic routing mechanism to the UPS encoder (H=2 blocks per layer vs. no routing) (ablation) | UPS · NTU120 xsub action recognition | Top-1 accuracy (%): 87.0 -> 89.3 | Helped |
| add dynamic routing mechanism to the UPS encoder (ablation) | UPS · Human3.6M 3D pose estimation | MPJPE (mm): 42.8 -> 40.8 | Helped |
| add dynamic routing mechanism to the UPS encoder (ablation) | UPS · NTU120 early action prediction (20%/40%/60%) | Top-1 accuracy (%): 53.02->53.25 (20%); 73.47->75.06 (40%); 84.15->85.35 (60%) | Helped |
| train one unified model on all four tasks (UPS) instead of separate per-task models (UPSseparate) (ablation) | UPS · NTU60/NTU120 3D action recognition | Top-1 accuracy (%): NTU60 xsub 89.6->92.6; NTU120 xsub 85.1->89.3 | Helped |
| train one unified model on all four tasks instead of separate per-task models (ablation) | UPS · Kinetics 400 2D action recognition | Top-1 accuracy (%): 36.2 -> 40.5 | Helped |
| train one unified model on all four tasks instead of separate per-task models (ablation) | UPS · Human3.6M 3D pose estimation | MPJPE (mm): 44.9 -> 40.8 | Helped |
| train one unified model on all four tasks instead of separate per-task models (ablation) | UPS · NTU60 early action prediction (20%) | Top-1 accuracy (%): 50.11 -> 53.25 | Helped |
| increase UPS encoder depth from 3 to 5 (ablation) | UPS · NTU120 action recognition, Human3.6M, early prediction | Top-1 accuracy (%) / MPJPE (mm): 89.3->89.2 (action); 40.8->40.8 (pose) | No clear effect |
| increase number of parallel blocks per layer from 2 to 5 (ablation) | UPS · NTU120 action recognition, Human3.6M, early prediction | Top-1 accuracy (%) / MPJPE (mm): 89.3->89.1 (action); 40.8->40.9 (pose) | No clear effect |
| increase number of parallel blocks per layer from 1 to 2 (enabling dynamic routing) (ablation) | UPS · NTU120 action recognition | Top-1 accuracy (%): 88.6 -> 89.3 | Helped |
| reduce encoder depth from 3 to 1 (ablation) | UPS · Human3.6M 3D pose estimation | MPJPE (mm): 40.8 -> 42.0 | Hurt |
| reduce encoder depth from 3 to 1 (ablation) | UPS · NTU120 action recognition and early prediction | Top-1 accuracy (%): 89.3->88.0 (action); 53.25->52.05 (20% early prediction) | Hurt |
Closest papers by meaning
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · similarity 0.85)
- UniPose: A Unified Multimodal Framework for Human Pose Comprehension, Generation and Editing (CVPR 2025 · similarity 0.84)
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.84)
- PeVL: Pose-Enhanced Vision-Language Model for Fine-Grained Human Action Recognition (CVPR 2024 · similarity 0.84)
- Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition (CVPR 2024 · similarity 0.84)
- PoseMamba: Monocular 3D Human Pose Estimation with Bidirectional Global-Local Spatio-Temporal State Space Model (AAAI 2025 · 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.