Research map / 3D vision / 3D generation
LART: Neural Correspondence Learning with Latent Regularization Transformer for 3D Motion Transfer
Haoyu Chen, Hao Tang, Radu Timofte, Luc Van Gool, Guoying Zhao
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 latent metric regularization (LMR) to LART (ablation) | LART · DFAUST single subject | PMD (x10^-4): 1.90 -> 0.32 | Improved |
| add latent metric regularization (LMR) to LART (ablation) | LART · DFAUST single subject | training time (h): 20 h -> 8 h | Improved |
| feed multi-frame driving sequence instead of single frame (ablation) | LART · DFAUST single subject | PMD (x10^-4): 2.77 -> 0.32 | Improved |
| feed multi-frame driving sequence instead of single frame (ablation) | LART · DFAUST single subject | training time (h): 25 h -> 8 h | Improved |
| replace NPT with LART (Transformer with multi-frame input and LMR) | NPT · DFAUST single subject | PMD (x10^-4): 2.82 -> 0.32 | Improved |
| single-frame LART versus NPT baseline (ablation) | LART single frame · DFAUST single subject | PMD (x10^-4): 2.82 -> 2.77 | No clear effect |
| replace 3D-CoreNet with LART | 3D-CoreNet · DFAUST seen motions | PMD (x10^-4): 1.25 -> 0.41 | Improved |
| replace 3D-CoreNet with LART | 3D-CoreNet · DFAUST unseen motions | PMD (x10^-4): 3.05 -> 1.80 | Improved |
| replace NPT with LART | NPT · DFAUST unseen motions | PMD (x10^-4): 5.31 -> 1.80 | Improved |
| evaluate on unseen instead of seen driving motions | LART · DFAUST | PMD (x10^-4): 0.41 -> 1.80 | Hurt |
| use multi-frame processing instead of frame-by-frame deformation on unseen motions | LART vs NPT and 3D-CoreNet · AMASS long sequences | qualitative robustness | Improved |
| add NPT motion-prior variant (NPT-MP) over NPT | NPT · DFAUST seen motions | PMD (x10^-4): 4.37 -> 4.51 | Hurt |
Closest papers by meaning
- Temporal Residual Jacobians for Rig-free Motion Transfer (ECCV 2024 · similarity 0.87)
- DIMO: Diverse 3D Motion Generation for Arbitrary Objects (ICCV 2025 · similarity 0.87)
- LEIA: Latent View-invariant Embeddings for Implicit 3D Articulation (ECCV 2024 · similarity 0.87)
- Geometry-Contrastive Transformer for Generalized 3D Pose Transfer (AAAI 2022 · similarity 0.86)
- Video Motion Transfer with Diffusion Transformers (CVPR 2025 · similarity 0.85)
- Motion2VecSets: 4D Latent Vector Set Diffusion for Non-rigid Shape Reconstruction and Tracking (CVPR 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.
