AI Paper Atlas

Research map / 3D vision / Human pose estimation

Mining Multi-View Information: A Strong Self-Supervised Framework for Depth-Based 3D Hand Pose and Mesh Estimation

Pengfei Ren, Haifeng Sun, Jiachang Hao, Jingyu Wang, Qi Qi, Jianxin Liao

CVPR 2022 · 3D vision › Human pose estimation · approach: Depth-based hand pose regression · 26 citations (OpenAlex)

Check 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 UGCN-based cross-view fusion with joint coordinates instead of no fusion (ablation)NYU multi-view hand-pose and mesh estimation · NYU multi-view fusionmean joint error: 23.71 -> 13.05 (HME); 14.14 -> 11.16 (HPE)Improved
Add visual features to UGCN node representations (ablation)NYU multi-view hand-pose and mesh estimation · NYU multi-view fusionmean joint error: 13.05 -> 12.62 (HME); 11.16 -> 10.46 (HPE)Improved
Add node confidence to UGCN fusion (ablation)NYU multi-view hand-pose and mesh estimation · NYU multi-view fusionmean joint error: 12.62 -> 11.83 (HME); 10.46 -> 10.32 (HPE)Improved
Use group-wise confidence encoding (ablation)NYU multi-view hand-pose and mesh estimation · NYU multi-view fusionmean joint error: 11.83 -> 11.64 (HME); 10.32 -> 10.29 (HPE)Improved
Add cross-view model-fitting loss (ablation)Self-supervised fine-tuning · NYUmean joint error: 17.49 -> 13.49 (HME); 12.47 -> 12.32 (HPE)Improved
Add multi-view self-distillation loss (ablation)Self-supervised fine-tuning · NYUmean joint error: 17.49 -> 12.52 (HME); 12.47 -> 10.71 (HPE)Improved
Add prior loss (ablation)Self-supervised fine-tuning · NYUmean joint error: 12.39 -> 12.43 (HME); 10.66 -> 10.50 (HPE)No clear effect
Use synthetic data translated with CycleGAN for pre-training (ablation)Single-view estimation network · NYUmean joint error: 24.32 -> 23.71 (HME); 16.21 -> 14.14 (HPE)Improved
Fine-tune with both training and testing real samples rather than testing samples only (ablation)Self-supervised method · NYU multi-viewmean joint error: 12.43 -> 10.49 (HME); 10.50 -> 10.23 (HPE)Improved
Use pseudo multi-view training with augmented samples of one inputSelf-supervised hand-pose estimation · NYU single-viewmean joint error: 16.96 -> 11.82Improved
Use cross-view fusion in the multi-view settingSelf-supervised hand-pose estimation · NYU multi-viewmean joint error: 12.26 -> 10.11Improved

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