Research map / 3D vision / Face and shape
Cyclic Test-Time Adaptation on Monocular Video for 3D Human Mesh Reconstruction
Hyeongjin Nam, Daniel Sungho Jung, Yeonguk Oh, Kyoung Mu Lee
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 3D supervision targets from MDNet (L_SMPL) to 2D reprojection loss (ablation) | CycleAdapt HMRNet · 3DPW (H36M source) | MPJPE: 125.5 -> 96.9 | Improved |
| use MDNet outputs as 3D targets for test samples instead of external H36M 3D loss (ablation) | CycleAdapt HMRNet · 3DPW (H36M source) | MPJPE: 115.2 -> 96.9 | Improved |
| add cyclic adaptation (also adapt MDNet iteratively) (ablation) | CycleAdapt · 3DPW (H36M source) | MPJPE: 96.9 -> 87.7 | Improved |
| add self-supervised masked adaptation of MDNet (ablation) | MDNet · 3DPW (H36M source) | MPJPE: 114.2 -> 96.2 | Improved |
| use MDNet instead of Gaussian filter, Motion infiller or SmoothNet for denoising (ablation) | CycleAdapt · 3DPW (H36M source) | MPJPE: 92.0 -> 87.7 | Improved |
| pre-train HMRNet on Human3.6M before adaptation (ablation) | CycleAdapt · 3DPW | MPJPE after adaptation: 140.6 -> 87.0 | Improved |
| pre-train HMRNet on Human3.6M, no adaptation (ablation) | HMRNet · 3DPW | MPJPE before adaptation: 272.0 -> 230.3 | No clear effect |
| CycleAdapt vs prior test-time adaptation with estimated OpenPose keypoints | BOA, DynaBOA, DAPA · 3DPW (H36M source) | MPJPE: 108.0 -> 87.7 (vs DAPA) | Improved |
| online adaptation setting (MDNet cannot see future frames) | CycleAdapt · 3DPW (H36M source) | MPJPE: 87.7 -> 90.3 | Hurt |
| CycleAdapt vs DAPA running time | test-time adaptation · 3DPW | ms per image: 431.0 -> 74.1 | Improved |
| apply CycleAdapt to PyMAF and Pose2Pose HMRNets instead of SPIN | CycleAdapt · 3DPW | MPJPE: SPIN 87.7, PyMAF 98.5, Pose2Pose 108.1 | Mixed |
Closest papers by meaning
- Bilevel Online Adaptation for Out-of-Domain Human Mesh Reconstruction (CVPR 2021 · similarity 0.88)
- Incorporating Test-Time Optimization into Training with Dual Networks for Human Mesh Recovery (NeurIPS 2024 · similarity 0.88)
- Co-Evolution of Pose and Mesh for 3D Human Body Estimation from Video (ICCV 2023 · similarity 0.88)
- Self-Supervised 3D Human Mesh Recovery from a Single Image with Uncertainty-Aware Learning (AAAI 2024 · similarity 0.87)
- FactorizedHMR: A Hybrid Framework for Video Human Mesh Recovery (NeurIPS 2026 · similarity 0.87)
- SAT-HMR: Real-Time Multi-Person 3D Mesh Estimation via Scale-Adaptive Tokens (CVPR 2025 · 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.
