Research map / 3D vision / Human pose estimation
View-Invariant, Occlusion-Robust Probabilistic Embedding for Human Pose
Ting Liu, Jennifer J. Sun, Long Zhao, Jiaping Zhao, Liangzhe Yuan, Yuxiao Wang, Liang-Chieh Chen, Florian Schroff, Hartwig Adam
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 camera augmentation (random 3D rotations) (ablation) | Pr-VIPE · 3DHP (all cameras) retrieval | Hit@1: 19.8 -> 27.3 | Improved |
| add camera augmentation (ablation) | Pr-VIPE · H3.6M hold-out retrieval | Hit@1: 74.6 -> 72.9 | Hurt |
| use probabilistic embedding instead of L2-normalized point embedding (ablation) | Pr-VIPE · H3.6M retrieval | Hit@1: 73.2 -> 74.6 | Improved |
| use Pr-VIPE instead of 3D lifting with Procrustes alignment | Pr-VIPE · H3.6M retrieval | Hit@1: 68.8 -> 74.6 | Improved |
| raise Pr-VIPE embedding dimension from 16D to 32D (ablation) | Pr-VIPE · H3.6M hold-out subset | Hit@1: 78.3 -> 81.6 | Improved |
| add keypoint dropout (independent) during training (ablation) | Pr-VIPE · H3.6M targeted occlusion | Hit@1: 2.52 -> 66.2 | Improved |
| add structured instead of independent keypoint dropout (ablation) | Pr-VIPE · H3.6M real-distribution occlusion | Hit@1: 70.3 -> 71.0 | Improved |
| add keypoint dropout to training (ablation) | Pr-VIPE · H3.6M no occlusion | Hit@1: 72.9 -> 71.7 | Hurt |
| use independent keypoint dropout instead of none (ablation) | Pr-VIPE · 3DPW realistic occlusion | Hit@1: 43.8 -> 64.6 | Improved |
| use temporal Pr-VIPE instead of stacked frame embeddings (ablation) | Pr-VIPE · Penn Action video alignment | Kendall's Tau: 0.7405 -> 0.7810 | Improved |
| use groundtruth 2D keypoints instead of detected (ablation) | Pr-VIPE · H3.6M retrieval | Hit@1: 74.6 -> 97.2 | Improved |
| increase number of samples K from 10 to 20 (ablation) | Pr-VIPE · H3.6M retrieval | Hit@1: 72.6 -> 72.9 | No clear effect |
Closest papers by meaning
- View-Invariant Probabilistic Embedding for Human Pose (ECCV 2020 · similarity 0.94)
- Occlusion-Aware Networks for 3D Human Pose Estimation in Video (ICCV 2019 · similarity 0.90)
- 3D Human Pose Estimation Using Spatio-Temporal Networks with Explicit Occlusion Training (AAAI 2020 · similarity 0.90)
- Dual Pose-invariant Embeddings: Learning Category and Object-specific Discriminative Representations for Recognition and Retrieval (CVPR 2024 · similarity 0.88)
- ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos (CVPR 2025 · similarity 0.87)
- $\text{Di}^2\text{Pose}$: Discrete Diffusion Model for Occluded 3D Human Pose Estimation (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.
