Research map / 3D vision / Human pose estimation
Harvesting Multiple Views for Marker-Less 3D Human Pose Annotations
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Combine multi-view 2D heatmaps with 3D camera geometry and skeleton constraints | 3D pose estimation on KTH Multiview Football II · KTH Multiview Football II, Sequence 1, two cameras | 3D PCP (Average): 87.5 -> 96.5 | Helped |
| Combine multi-view 2D heatmaps with 3D camera geometry and skeleton constraints | 3D pose estimation on KTH Multiview Football II · KTH Multiview Football II, Sequence 1, three cameras | 3D PCP (Average): 90.3 -> 100 | Helped |
| Use multi-view optimization rather than the single-view Zhou et al. estimate | Human3.6M multi-view pose estimation · Human3.6M, 15 actions | average 3D joint error (mm): 107.26 -> 56.89 | Helped |
| Fine-tune the generic 2D ConvNet using harvested annotations | 2D pose prediction on KTH Multiview Football II · Player 2, Sequence 1 | 2D PCP (Average): 95.3 -> 98.0 | Helped |
| Fine-tune the generic 2D ConvNet using harvested annotations | 2D pose prediction on KTH Multiview Football II · Player 2, Sequence 2 | 2D PCP (Average): 77.0 -> 84.5 | Helped |
| Replace generic predictions with heatmap-maximum annotations for fine-tuning (ablation) | Human3.6M multi-view pose estimation after fine-tuning · Human3.6M, 15 actions | average 3D joint error (mm): 56.89 -> 55.13 | Helped |
| Select the top 70% of heatmap-maximum annotations for fine-tuning (ablation) | Human3.6M multi-view pose estimation after fine-tuning · Human3.6M, 15 actions | average 3D joint error (mm): 55.13 -> 55.62 | No clear effect |
| Replace heatmap-derived annotations with all multi-view pictorial-structure annotations (ablation) | Human3.6M multi-view pose estimation after fine-tuning · Human3.6M, 15 actions | average 3D joint error (mm): 56.89 -> 51.18 | Helped |
| Select reliable joints from the multi-view pictorial-structure annotations (ablation) | Human3.6M multi-view pose estimation after fine-tuning · Human3.6M, 15 actions | average 3D joint error (mm): 51.18 -> 47.83 | Helped |
| Train a single-view 3D ConvNet from scratch using harvested annotations instead of dataset 3D groundtruth | Single-view 3D pose prediction · Human3.6M, 15 actions | average 3D joint error (mm): 107.26 -> 118.41 | Mixed |
Closest papers by meaning
- Coarse-To-Fine Volumetric Prediction for Single-Image 3D Human Pose (CVPR 2017 · similarity 0.90)
- Learning Monocular 3D Human Pose Estimation From Multi-View Images (CVPR 2018 · similarity 0.89)
- Learning to Estimate 3D Human Pose and Shape From a Single Color Image (CVPR 2018 · similarity 0.88)
- In the Wild Human Pose Estimation Using Explicit 2D Features and Intermediate 3D Representations (CVPR 2019 · similarity 0.88)
- Multiview-Consistent Semi-Supervised Learning for 3D Human Pose Estimation (CVPR 2020 · similarity 0.88)
- ManiPose: Manifold-Constrained Multi-Hypothesis 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.