Research map / 3D vision / Human pose estimation
Learning 3D Human Shape and Pose from Dense Body Parts
Hongwen Zhang, Jie Cao, Guo Lu, Wanli Ouyang, Zhenan Sun
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 |
|---|---|---|---|
| use IUV maps instead of conv features as intermediate representation (ablation) | baseline network · Human3.6M | PVE (mm): 98.9 -> 87.8 | Improved |
| use IUV maps instead of part segmentation (ablation) | baseline network · Human3.6M | PVE (mm): 90.4 -> 87.8 | Improved |
| add decomposed multi-stream perception (D-Net) (ablation) | baseline network · Human3.6M | PVE (mm): 87.8 -> 74.3 | Improved |
| add decomposed perception (ablation) | baseline network · Human3.6M | PVE-S (mm): 38.0 -> 36.3 | No clear effect |
| use scale ratios alpha_k instead of equal scales 0.5 (ablation) | D-Net · Human3.6M | PVE (mm): 76.1 -> 74.3 | Improved |
| use partial IUV maps instead of all body parts (ablation) | D-Net · Human3.6M | PVE (mm): 76.8 -> 74.3 | Improved |
| add aggregated refinement (position-aided) (ablation) | DaNet · Human3.6M | PVE (mm): 71.7 -> 66.5 | Improved |
| use position-aided instead of direct refinement (ablation) | DaNet · Human3.6M | MPJPE-PA (mm): 45.5 -> 42.9 | Improved |
| replace LSTM aggregation with graph-based aggregation (ablation) | DaNet · COCO val | AP50: 58.7 -> 65.5 | Improved |
| add PartDrop to DaNet-GCN (ablation) | DaNet · COCO val | AP: 31.9 -> 33.8 | Improved |
| add Dropout instead of PartDrop (ablation) | DaNet-GCN · COCO val | AP: 31.9 -> 30.6 | Hurt |
| add DropBlock instead of PartDrop (ablation) | DaNet-GCN · COCO val | AP: 31.9 -> 32.0 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VIBE — VIBE: Video Inference for Human Body Pose and Shape Estimation (CVPR 2020) · better on 3 of 3 shared results
- HoloPose — HoloPose: Holistic 3D Human Reconstruction In-The-Wild (CVPR 2019) · better on 2 of 2 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
- End-to-End Recovery of Human Shape and Pose (CVPR 2018 · similarity 0.90)
- BodyNet: Volumetric Inference of 3D Human Body Shapes (ECCV 2018 · similarity 0.89)
- Accurate 3D Body Shape Regression Using Metric and Semantic Attributes (CVPR 2022 · similarity 0.89)
- Learning to Estimate 3D Human Pose and Shape From a Single Color Image (CVPR 2018 · similarity 0.89)
- Hierarchical Kinematic Probability Distributions for 3D Human Shape and Pose Estimation From Images in the Wild (ICCV 2021 · similarity 0.89)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · 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.
