Research map / Point clouds / Point cloud classification
PUMPS: Skeleton-Agnostic Point-based Universal Motion Pre-Training for Synthesis in Human Motion Tasks
Clinton Ansun Mo, Kun Hu, Chengjiang Long, Dong Yuan, Wan-Chi Siu, Zhiyong 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 |
|---|---|---|---|
| use sequence-wise dropout instead of element-wise dropout in the TPC decoder (ablation) | PUMPS auto-encoder · TPC reconstruction | L_rec: 0.0204 -> 0.0029 | Improved |
| use rotary self-attention instead of sinusoidal positions (ablation) | PUMPS auto-encoder · TPC reconstruction | L_rec: 0.0052 -> 0.0029 | Improved |
| use linear assignment (Hungarian) pairing instead of Chamfer distance objective (ablation) | PUMPS auto-encoder · TPC reconstruction | L_rec: 0.0085 -> 0.0029 | Improved |
| use Gaussian-noise point identifiers instead of FoldingNet 2D folding (ablation) | PUMPS auto-encoder · TPC reconstruction | L_rec: 0.0040 -> 0.0029 | Improved |
| use Gaussian-noise point identifiers instead of PointIVAE VAE sampling (ablation) | PUMPS auto-encoder · TPC reconstruction | L_rec: 0.0260 -> 0.0029 | Improved |
| pre-train on TPC motion completion before fine-tuning for 2D-to-3D estimation (ablation) | PUMPS latent motion synthesiser · 2D-to-3D motion estimation | MPJPE: 73.56 -> 44.15 (Aim) | Improved |
| pre-train on TPC motion completion before fine-tuning for denoising (ablation) | PUMPS latent motion synthesiser · Motion denoising (LaFAN1) | MPJPE: 70.20 -> 49.83 (Aim) | Improved |
| pre-train before fine-tuning for denoising, acceleration error (ablation) | PUMPS latent motion synthesiser · Motion denoising (LaFAN1) | Acceleration error: 8.92 -> 10.45 (Aim) | Hurt |
| interpolate in the PUMPS point cloud latent space rather than skeletal space | PUMPS · Keyframe interpolation and motion transition | L2P: 0.137 vs 0.186 (5-frame, CITL); 0.948 vs 0.908 (30-frame) | Mixed |
| PUMPS vs MHFormer on 2D-to-3D estimation | PUMPS (finetune) · 2D-to-3D motion estimation | MPJPE: 44.15 vs 40.38 (Aim) | Mixed |
Closest papers by meaning
- SAMoR: Motion Modelling for Articulated Objects of Any Skeleton and Topology (NeurIPS 2026 · similarity 0.88)
- How to Move Your Dragon: Text-to-Motion Synthesis for Large-Vocabulary Objects (ICML 2025 · similarity 0.86)
- S3O: A Dual-Phase Approach for Reconstructing Dynamic Shape and Skeleton of Articulated Objects from Single Monocular Video (ICML 2024 · similarity 0.85)
- A Unified Masked Autoencoder with Patchified Skeletons for Motion Synthesis (AAAI 2024 · similarity 0.85)
- Learning Implicit Representation for Reconstructing Articulated Objects (ICLR 2024 · similarity 0.85)
- SPEAL: Skeletal Prior Embedded Attention Learning for Cross-Source Point Cloud Registration (AAAI 2024 · similarity 0.85)
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.
