Research map / 3D vision / Human pose estimation
Learning Causal Representation for Training Cross-Domain Pose Estimator via Generative Interventions
Xiheng Zhang, Yongkang Wong, Xiaofei Wu, Juwei Lu, Mohan Kankanhalli, Xiangdong Li, Weidong Geng
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 counterfactual feature generator with unlabeled target data (SD + TD) | pose estimator HRNet-W32 · H3.6M->3DPW | MPJPE: 118.7 -> 94.7 | Improved |
| use MMD instead of naive l2 as discrepancy loss (ablation) | SD + TD model · H3.6M->3DPW | MPJPE: 100.3 -> 94.7 | Improved |
| use MMD instead of KL divergence as discrepancy loss (ablation) | SD + TD model · H3.6M->3DPW | MPJPE: 96.5 -> 94.7 | Improved |
| apply method on ResNet-18 backbone | SD + TD model · H3.6M->3DPW | MPJPE: 122.4 -> 98.3 | Improved |
| apply method on ResNet-50 backbone | SD + TD model · H3.6M->3DPW | MPJPE: 120.2 -> 96.3 | Improved |
| add 3DHP, HumanEva and SURREAL as extra source datasets (ablation) | SD + TD model · H3.6M->3DPW | MPJPE: 94.7 -> 94.1 | No clear effect |
| add H3.6M, HumanEva and SURREAL as extra source datasets (ablation) | SD + TD model · 3DPW->3DHP | MPJPE: 109.7 -> 102.7 | Improved |
| use multiple unconstrained datasets (UCF101, HMDB, Kinetics) instead of Kinetics only (ablation) | SD + UD model · H3.6M->3DPW | MPJPE: 97.5 -> 94.9 | Improved |
| use all three general image datasets instead of ImageNet only as interventions (ablation) | SD + UD model · H3.6M->3DPW | MPJPE: 102.3 -> 99.2 | Improved |
| use unconstrained data (Kinetics) instead of target data | domain generalization · H3.6M->3DPW | MPJPE: 94.7 -> 97.5 | Hurt |
| add counterfactual features with unlabeled target data on hand pose | FreiHAND model · FreiHAND->STB | EPE: 36.1 -> 22.4 | Improved |
| single unconstrained intervention dataset (UCF101 vs ImageNet) (ablation) | SD + UD model · H3.6M->3DPW | MPJPE: 98.6 vs 102.3 | Mixed |
Closest papers by meaning
- A Dual-Augmentor Framework for Domain Generalization in 3D Human Pose Estimation (CVPR 2024 · similarity 0.88)
- AdaptPose: Cross-Dataset Adaptation for 3D Human Pose Estimation by Learnable Motion Generation (CVPR 2022 · similarity 0.87)
- Causal-Inspired Multitask Learning for Video-Based Human Pose Estimation (AAAI 2025 · similarity 0.86)
- Source-Free and Image-Only Unsupervised Domain Adaptation for Category Level Object Pose Estimation (ICLR 2024 · similarity 0.86)
- Sparse-view Pose Estimation and Reconstruction via Analysis by Generative Synthesis (NeurIPS 2024 · similarity 0.86)
- Can Generative Video Models Help Pose Estimation? (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.
