Research map / 3D vision / Human pose estimation
SmoothNet: A Plug-and-Play Network for Refining Human Poses in Videos
Ailing Zeng, Lei Yang, Xuan Ju, Jiefeng Li, Jianyi Wang, Qiang Xu
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 acceleration loss to the backbone training (ablation) | VPose In=27 · Human3.6M | Accel / MPJPE: Accel 5.07 -> 4.12; MPJPE 50.13 -> 51.48 | Mixed |
| train SmoothNet end-to-end with the backbone instead of two-stage (ablation) | VPose In=27 · Human3.6M | MPJPE (mm): 50.04 -> 52.65 | Hurt |
| train SmoothNet separately (two-stage) after the backbone (ablation) | VIBE In=16 · 3DPW | MPJPE (mm): 86.56 -> 81.42 | Improved |
| add intermediate L1 supervision and acceleration loss to end-to-end SmoothNet (ablation) | VPose In=27 w/ B · Human3.6M | Accel / MPJPE: Accel 5.46 -> 2.69; MPJPE 51.06 -> 50.94 | Improved |
| add SmoothNet to single-frame backbone | VIBE In=1 · 3DPW | Accel: 32.69 -> 6.12 | Improved |
| use SmoothNet instead of filters with similar Accel | FCN · Human3.6M 3D pose | MPJPE (mm): 143.24 -> 52.72 | Improved |
| use SmoothNet instead of best filter | TPoseNet · MuPoTS-3D | Accel: 8.29 -> 7.23 | Improved |
| add SmoothNet on top of RefineNet | TPoseNet w/ RefineNet · MuPoTS-3D | MPJPE (mm): 93.97 -> 91.78 | Improved |
| add acceleration loss to pose loss (ablation) | SmoothNet · 3DPW (EFT) | Accel: 6.42 -> 6.30 | Improved |
| add pose loss to acceleration-only training (ablation) | SmoothNet · 3DPW (EFT) | MPJPE (mm): 446.54 -> 86.39 | Improved |
| use 3D position instead of axis-angle as training modality (ablation) | SmoothNet · 3DPW (EFT) | Accel: 77.89 -> 6.30 | Improved |
| use 3D position instead of 6D rotation as training modality (ablation) | SmoothNet · 3DPW (EFT) | MPJPE (mm): 86.92 -> 86.39 | Improved |
Closest papers by meaning
- Synthesizing Images of Humans in Unseen Poses (CVPR 2018 · similarity 0.83)
- Robust Pose Transfer With Dynamic Details Using Neural Video Rendering (TPAMI 2022 · similarity 0.83)
- HumanDreamer: Generating Controllable Human-Motion Videos via Decoupled Generation (CVPR 2025 · similarity 0.82)
- ReGenHuman: Re-Generating Human Appearances for Realistic Full-Body Video Anonymization (NeurIPS 2026 · similarity 0.82)
- HumanVid: Demystifying Training Data for Camera-controllable Human Image Animation (NeurIPS 2024 · similarity 0.82)
- OpenHumanVid: A Large-Scale High-Quality Dataset for Enhancing Human-Centric Video Generation (CVPR 2025 · similarity 0.81)
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.
