Research map / Detection and segmentation / CNN features
Cascaded Pyramid Network for Multi-Person Pose Estimation
Read the paperCheck 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 |
|---|---|---|---|
| Add RefineNet to GlobalNet (ablation) | CPN · COCO minival | AP (OKS): 66.6 -> 68.6 | Helped |
| Add online hard keypoint mining to RefineNet (ablation) | CPN · COCO minival | AP (OKS): 68.6 -> 69.4 | Helped |
| Apply online hard keypoint mining to GlobalNet too (ablation) | CPN · COCO minival | AP (OKS): 69.4 -> 69.1 | Hurt |
| Remove GlobalNet intermediate supervision (ablation) | CPN · COCO minival | AP (OKS): 69.1 -> 68.2 | Hurt |
| Use all pyramid levels C2-C5 rather than only C2 (ablation) | RefineNet · COCO minival | AP (OKS): 68.3 -> 69.4 | Helped |
| Select the top 8 keypoint losses for mining (ablation) | RefineNet · COCO minival | AP (OKS): 68.8 (M=6); 69.4 (M=8); 69.0 (M=10); 69.0 (M=12); 69.0 (M=14); 68.6 (M=17) | Helped |
| Increase crop input size to 384×288 (ablation) | CPN · COCO minival | AP (OKS): 69.4 -> 71.6 | Helped |
| Increase hourglass input size to 256×256 (ablation) | 8-stage hourglass · COCO minival | AP (OKS): 66.9 -> 67.1 | No clear effect |
| Replace hard NMS with Soft-NMS (ablation) | Top-down pose pipeline · COCO minival | AP(OKS): 69.2 -> 69.4 | Helped |
| Increase hard NMS threshold from 0.3 to 0.5 (ablation) | Top-down pose pipeline · COCO minival | AP(OKS): 68.2 -> 69.2 | Helped |
| Increase detector accuracy from FPN-3 to ensemble-1 (ablation) | Top-down pose pipeline · COCO minival | AP(OKS): 69.7 -> 69.8 | No clear effect |
| Ensemble CPN models (ablation) | CPN · COCO test-dev | AP: 72.1 -> 73.0 | Helped |
Closest papers by meaning
- Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation (AAAI 2020 · similarity 0.90)
- Towards Accurate Multi-Person Pose Estimation in the Wild (CVPR 2017 · similarity 0.90)
- DGCN: Dynamic Graph Convolutional Network for Efficient Multi-Person Pose Estimation (AAAI 2020 · similarity 0.89)
- Graph and Temporal Convolutional Networks for 3D Multi-person Pose Estimation in Monocular Videos (AAAI 2021 · similarity 0.87)
- Deep Dual Consecutive Network for Human Pose Estimation (CVPR 2021 · similarity 0.87)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · 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.