AI Paper Atlas

Research map / 3D vision / Human pose estimation

Shape-Aware Multi-Person Pose Estimation From Multi-View Images

Zijian Dong, Jie Song, Xu Chen, Chen Guo, Otmar Hilliges

ICCV 2021 · 3D vision › Human pose estimation · approach: Multi-person pose estimation · 58 citations (OpenAlex)

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 changedWhereMeasuredEffect
add shape-aware refinement (multi-view constraints plus SMPL regularization) on top of the initial 3D proposals (ablation)Ours · Association DatasetF1-score (%): 83.4 -> 89.2Improved
add shape-aware refinement (ablation)Ours · Association DatasetPrecision (%): 83.7 -> 90.1Improved
add shape-aware refinement (ablation)Ours · Association DatasetRecall (%): 82.8 -> 89.0Improved
use the proposed fitting (joint 3D pose optimization with body prior) instead of SMPL-only fitting (ablation)Ours · Association DatasetF1-score (%): 77.8 -> 89.2Improved
use the proposed fitting instead of SMPL-only fitting (ablation)Ours · Association Datasetprecision (%): 78.3% -> 90.1%Improved
use learned gradient descent for optimization instead of standard optimization (ablation)Ours · Association Datasetruntime: 2s -> 0.1s; 100 -> 14 iterations (20x speed-up)Improved
add SMPL body-model regularization for joints with low confidence (ablation)Ours · Association Dataset (MPJPE, low-confidence joints)MPJPEImproved
add SMPL body-model regularization for joints with confidence above 0.25 (ablation)Ours · Association Dataset (MPJPE, high-confidence joints)MPJPENo clear effect
use the proposed 3D proposal generation instead of VoxelPose (ablation)proposal generation · Association Dataset (human proposal)F1-score (%): 72.8 -> 96.4Improved
use the proposed 3D proposal generation instead of Zhang et al. (ablation)proposal generation · Association Dataset (human proposal)Recall (%): 51.2 -> 94.2Improved
use the proposed method instead of a 3D-supervised learned method (VoxelPose) on unseen datamulti-view pose estimation · Association DatasetF1-score (%): 60.3 -> 89.2Improved
use the proposed method instead of 3D-supervised VoxelPose trained on the Shelf setmulti-view pose estimation · ShelfPCP average (%): 97.0 -> 96.9No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy