Research map / 3D vision / Human pose estimation
Inference Stage Optimization for Cross-scenario 3D Human Pose Estimation
Jianfeng Zhang, Xuecheng Nie, Jiashi Feng
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 |
|---|---|---|---|
| adapt the pose model on each target sample using online ISO-Cycle | FSL-trained 3D pose model · MPI-INF-3DHP unscaled | PCK: 78.9 -> 83.6 | Improved |
| adapt the pose model on each target sample using online ISO-Cycle | FSL-trained 3D pose model · MPI-INF-3DHP unscaled | MPJPE: 103.8 -> 92.2 | Improved |
| add geometric cycle consistency to adversarial SSL (ablation) | ISO-Adversary on MPI-INF-3DHP · MPI-INF-3DHP unscaled, online setting | PCK: 83.0 -> 83.6 | Improved |
| add geometric cycle consistency to adversarial SSL (ablation) | ISO-Adversary on MPI-INF-3DHP · MPI-INF-3DHP unscaled, joint setting | PCK: 80.9 -> 81.3 | Improved |
| add geometric cycle consistency to adversarial SSL (ablation) | ISO-Adversary on MPI-INF-3DHP · MPI-INF-3DHP unscaled, vanilla setting | PCK: 82.1 -> 82.5 | Improved |
| stream adaptation parameters across target instances instead of resetting per sample (ablation) | ISO-Cycle · MPI-INF-3DHP unscaled | PCK: 82.5 -> 83.6 | Improved |
| increase vanilla ISO iterations from 1 to 10 (ablation) | Vanilla ISO-Cycle · MPI-INF-3DHP | PCK: 81.5% -> 82.5% | Improved |
| increase online ISO iterations above 1 (ablation) | Online ISO-Cycle · MPI-INF-3DHP | PCK | Hurt |
| use a large learning rate such as 2e-3 for online adaptation (ablation) | Online ISO-Cycle · MPI-INF-3DHP | PCK: 70.9% PCK | Hurt |
| perform SSL once per 10 samples instead of on every sample (ablation) | Online ISO · MPI-INF-3DHP unscaled | PCK: 83.6 -> 83.0 | Hurt |
| perform SSL once per 10 samples instead of on every sample (ablation) | Online ISO · MPI-INF-3DHP unscaled | Time[s]: 0.027 -> 0.004 | Improved |
| add Gaussian noise with standard deviation 10 to 2D inputs (ablation) | ISO · MPI-INF-3DHP | PCK: 83.6 -> 79.6 | Hurt |
Closest papers by meaning
- Geometry-Driven Self-Supervised Method for 3D Human Pose Estimation (AAAI 2020 · similarity 0.89)
- AdaptPose: Cross-Dataset Adaptation for 3D Human Pose Estimation by Learnable Motion Generation (CVPR 2022 · similarity 0.89)
- ManiPose: Manifold-Constrained Multi-Hypothesis 3D Human Pose Estimation (NeurIPS 2024 · similarity 0.88)
- Weakly-Supervised Discovery of Geometry-Aware Representation for 3D Human Pose Estimation (CVPR 2019 · similarity 0.88)
- Self-Supervised 3D Human Pose Estimation via Part Guided Novel Image Synthesis (CVPR 2020 · similarity 0.88)
- MV-SSM: Multi-View State Space Modeling for 3D Human Pose Estimation (CVPR 2025 · similarity 0.88)
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.
