Research map / 3D vision / Human pose estimation
PersonLab: Person Pose Estimation and Instance Segmentation with a Bottom-Up, Part-Based, Geometric Embedding Model
George Papandreou, Tyler Zhu, Liang-Chieh Chen, Spyros Gidaris, Jonathan Tompson, Kevin Murphy
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 recurrent mid- and long-range offset refinement (ablation) | PersonLab ResNet-101 · COCO keypoints val | AP: 0.632 -> 0.665 | Improved |
| add recurrent offset refinement (ablation) | PersonLab ResNet-101 · COCO person segmentation val | mask AP: 0.355 -> 0.382 | Improved |
| add offset refinement, effect on large instances (ablation) | PersonLab ResNet-101 · COCO person segmentation val | large-instance mask AP: 0.501 -> 0.592 | Improved |
| use Expected-OKS keypoint scoring instead of Hough scores (hard NMS) (ablation) | PersonLab ResNet-101 · COCO keypoints val | AP: 0.632 -> 0.647 | Improved |
| use soft-NMS instead of hard NMS (Expected-OKS scoring) (ablation) | PersonLab ResNet-101 · COCO keypoints val | AP: 0.647 -> 0.665 | Improved |
| impute keypoints of small person instances for training (ablation) | PersonLab ResNet-101 · COCO keypoints val | AP: 0.665 -> 0.657 | Hurt |
| impute keypoints of small person instances for training (ablation) | PersonLab ResNet-101 · COCO person segmentation val | mask AP: 0.338 -> 0.382 | Improved |
| impute small-instance keypoints, effect on small instances (ablation) | PersonLab ResNet-101 · COCO person segmentation val | small-instance mask AP: 0.076 -> 0.164 | Improved |
| reduce output stride from 32 to 16 with atrous convolution (ablation) | PersonLab ResNet-101 · COCO keypoints val, input 1401 | AP: 0.641 -> 0.656 | Improved |
| reduce output stride from 16 to 8 (ablation) | PersonLab ResNet-101 · COCO keypoints val, input 1401 | AP: 0.656 -> 0.665 | Improved |
| increase input size beyond 1401 to 1801 (stride 8) (ablation) | PersonLab ResNet-101 · COCO keypoints val | AP: 0.665 -> 0.652 | Hurt |
| use ResNet-152 instead of ResNet-101 backbone (single-scale) | PersonLab · COCO keypoints test-dev | AP: 0.655 -> 0.665 | Improved |
Closest papers by meaning
- Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation (AAAI 2020 · similarity 0.89)
- DeepCut: Joint Subset Partition and Labeling for Multi Person Pose Estimation (CVPR 2016 · similarity 0.88)
- Joint Multi-Person Pose Estimation and Semantic Part Segmentation (CVPR 2017 · similarity 0.88)
- Instance-level Human Parsing via Part Grouping Network (ECCV 2018 · similarity 0.88)
- Detangling People: Individuating Multiple Close People and Their Body Parts via Region Assembly (CVPR 2017 · similarity 0.88)
- Joint Human Pose Estimation and Instance Segmentation with PosePlusSeg (AAAI 2022 · similarity 0.87)
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.
