Research map / 3D vision / Human pose estimation
Regularizing Vector Embedding in Bottom-Up Human Pose Estimation
Haixin Wang, Lu Zhou, Yingying Chen, Ming Tang, Jinqiao Wang
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 embedding regularization (scale loss) (ablation) | AE (HRNet-W32) · CrowdPose Test | AP: 64.8 -> 66.0 | Improved |
| add adjusted heatmap (ablation) | AE + embedding regularization · CrowdPose Test | AP: 66.0 -> 67.6 | Improved |
| add weighted L2-loss on top of regularization and adjusted heatmap (ablation) | Coupled Embedding · CrowdPose Test | AP: 67.6 -> 68.9 | Improved |
| add all strategies (regularization, adjusted heatmap, weighted loss) (ablation) | AE (HRNet-W32) · CrowdPose Test | AP_hard: 57.5 -> 60.8 | Improved |
| add scale loss to AE for grouping (ablation) | AE · CrowdPose Test subset3 (>6 instances, top 20% occlusion) | AP: 49.8 -> 52.0 | Improved |
| add scale loss to AE for grouping (ablation) | AE · CrowdPose Test subset4 (>6 instances, last 20% occlusion) | AP: 60.3 -> 62.2 | Improved |
| add scale loss to AE for grouping (ablation) | AE · CrowdPose Test subset1 (2-5 instances, top 20% occlusion) | AP: 64.4 -> 65.9 | Improved |
| add scale loss to AE for grouping (ablation) | AE · CrowdPose Test subset2 (2-5 instances, last 20% occlusion) | AP: 78.8 -> 78.7 | No clear effect |
| use instance scale directly as Gaussian scale factor instead of fixed sigma (ablation) | AE · CrowdPose Test | AP: 64.8 -> 63.9 | Hurt |
| use learned embedding-derived scale factor for heatmap sigma (ablation) | Coupled Embedding · CrowdPose Test | AP: 67.6 -> 67.6 | Improved |
| increase embedding dimensionality M from 4 to 8 (ablation) | Coupled Embedding · CrowdPose Test | AP: 67.1 -> 67.6 | Improved |
| increase embedding dimensionality M from 8 to 16 (ablation) | Coupled Embedding · CrowdPose Test | AP: 67.6 -> 67.3 | No clear effect |
Closest papers by meaning
- Associative Embedding: End-to-End Learning for Joint Detection and Grouping (NeurIPS 2017 · similarity 0.88)
- QueryPose: Sparse Multi-Person Pose Regression via Spatial-Aware Part-Level Query (NeurIPS 2022 · similarity 0.88)
- Bottom-Up Human Pose Estimation via Disentangled Keypoint Regression (CVPR 2021 · similarity 0.87)
- View-Invariant, Occlusion-Robust Probabilistic Embedding for Human Pose (IJCV 2021 · similarity 0.87)
- HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation (CVPR 2020 · similarity 0.87)
- Dual Pose-invariant Embeddings: Learning Category and Object-specific Discriminative Representations for Recognition and Retrieval (CVPR 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.
