Research map / 3D vision / Depth and stereo
Deep Non-Rigid Structure-From-Motion: A Sequence-to-Sequence Translation Perspective
Hui Deng, Tong Zhang, Yuchao Dai, Jiawei Shi, Yiran Zhong, Hongdong Li
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 whole input sequence instead of single frame (ablation) | sequence-to-sequence NRSfM · Human3.6M (GT 2D) | MPJPE: 90.7 -> 72.5 | Improved |
| add nuclear norm low-rank loss (ablation) | sequence-to-sequence NRSfM · Human3.6M (GT 2D) | MPJPE: 81.1 -> 72.5 | Improved |
| add canonicalization loss (ablation) | sequence-to-sequence NRSfM · Human3.6M (GT 2D) | MPJPE: 98.1 -> 72.5 | Improved |
| add smoothness loss with absolute temporal encoding (ablation) | sequence-to-sequence NRSfM · Human3.6M (GT 2D) | MPJPE: 79.8 -> 72.5 | Improved |
| use four-layer transformer instead of single-attention Context Layer (ablation) | Context Layer · Human3.6M (GT 2D) | MPJPE: 72.5 -> 93.6 | Hurt |
| shuffle frame order at inference (ablation) | trained model · Human3.6M (GT 2D) | MPJPE: 72.5 -> 83.7 | Hurt |
| reverse frame order at inference (ablation) | trained model · Human3.6M (GT 2D) | MPJPE: 72.5 -> 80.9 | Hurt |
| add smoothness loss without temporal encoding (ablation) | Base+nopos · Human3.6M, train length 32, inference length 32 | MPJPE: 78.3 -> 108.9 | Hurt |
| replace absolute encoding with no encoding for longer inference (ablation) | Base+abspos · Human3.6M, inference length 512 | MPJPE: 204.8 -> 77.1 | Improved |
| use ALIBI instead of RoPE encoding (ablation) | Base+rope · Human3.6M, inference length 512 | MPJPE: 90.9 -> 78.7 | Improved |
| add independent re-projection loss to the shape/motion predictor (ablation) | full model · Human3.6M (GT 2D) | MPJPE: 72.5 -> 89.8 | Hurt |
| freeze pre-trained shape/motion predictor (ablation) | full model · Human3.6M (GT 2D) | MPJPE: 72.5 -> 91.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- C3DPO — C3DPO: Canonical 3D Pose Networks for Non-Rigid Structure From Motion (ICCV 2019) · better on 8 of 8 shared results
- PAUL — PAUL: Procrustean Autoencoder for Unsupervised Lifting (CVPR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Deep Non-Rigid Structure-from-Motion Revisited: Canonicalization and Sequence Modeling (AAAI 2025 · similarity 0.94)
- Neural Dense Non-Rigid Structure from Motion with Latent Space Constraints (ECCV 2020 · similarity 0.92)
- Deep Non-Rigid Structure From Motion With Missing Data (TPAMI 2020 · similarity 0.92)
- Non-Rigid Structure-from-Motion: Temporally-Smooth Procrustean Alignment and Spatially-Variant Deformation Modeling (CVPR 2024 · similarity 0.88)
- VGGSfM: Visual Geometry Grounded Deep Structure From Motion (CVPR 2024 · similarity 0.87)
- TFS-NeRF: Template-Free NeRF for Semantic 3D Reconstruction of Dynamic Scene (NeurIPS 2024 · 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.
