Research map / 3D vision / Human pose estimation
Occlusion Handling in 3D Human Pose Estimation with Perturbed Positional Encoding
Niloofar Azizi, Mohsen Fayyaz, Horst Bischof
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 PerturbPE positional encoding to MobiusGCN (0.66M) | MobiusGCN · Human3.6M Protocol #1 (2D GT) | MPJPE (mm): 34.1 -> 32.7 | Improved |
| widen MobiusGCN channels from 128 to 192 (ablation) | MobiusGCN · Human3.6M Protocol #1 (2D GT) | MPJPE (mm): 36.2 -> 34.1 | Improved |
| use perturbed eigenvectors (0-edge, resolving multiplicity) instead of plain Laplacian eigenvectors (ablation) | PerturbPE with one edge missing · Human3.6M, one edge missing | MPJPE (mm): 55.0 -> 51.4 | Improved |
| average over 1-edge perturbations instead of 0-edge perturbation (ablation) | PerturbPE with one edge missing · Human3.6M, one edge missing | MPJPE (mm): 51.4 -> 49.0 | Improved |
| perturb with two removed edges instead of one (ablation) | PerturbPE with one edge missing · Human3.6M, one edge missing | MPJPE (mm): 49.0 -> 48.0 | Improved |
| add PerturbPE to MobiusGCN when two edges are missing | MobiusGCN · Human3.6M, two edges missing | MPJPE (mm): 60.0 -> 54.0 | Improved |
| add PerturbPE when trained with few subjects (S1 only) | MobiusGCN · Human3.6M reduced training data | MPJPE (mm): 44.7 -> 42.9 | Improved |
| add PerturbPE to MobiusGCN on cross-dataset test | MobiusGCN · MPI-INF-3DHP | PCK (All): 80.0 -> 82.0 | Improved |
| add PerturbPE to SemGCN | SemGCN · Human3.6M (2D GT) | MPJPE (mm): 43.1 -> 42.0 | Improved |
| use one network with PerturbPE vs GFPose per-case networks for occluded limbs | PerturbPE · Human3.6M, both arms occluded | MPJPE (mm): 60.0 -> 58.6 | Improved |
| add PerturbPE (adds RSPT computation) to MobiusGCN inference | MobiusGCN · inference time per sample | seconds: 0.009 -> 0.010 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GFPose — GFPose: Learning 3D Human Pose Prior With Gradient Fields (CVPR 2023) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Occlusion-Aware Networks for 3D Human Pose Estimation in Video (ICCV 2019 · similarity 0.88)
- 3D Human Pose Estimation Using Spatio-Temporal Networks with Explicit Occlusion Training (AAAI 2020 · similarity 0.88)
- $\text{Di}^2\text{Pose}$: Discrete Diffusion Model for Occluded 3D Human Pose Estimation (NeurIPS 2024 · similarity 0.88)
- Modulated Graph Convolutional Network for 3D Human Pose Estimation (ICCV 2021 · similarity 0.87)
- GRPose: Learning Graph Relations for Human Image Generation with Pose Priors (AAAI 2025 · similarity 0.86)
- Deep Semantic Graph Transformer for Multi-View 3D Human Pose Estimation (AAAI 2024 · similarity 0.85)
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.
