AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

Synthetic-to-Real Pose Estimation with Geometric Reconstruction

Qiuxia Lin, Kerui Gu, Linlin Yang, Angela Yao

NeurIPS 2023 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Geometric alignment and synthetic-to-real adaptation

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
replace UniFrame pseudo-labelling with integrated 2D+3D-kinematic selection (ablation)Ours · RHD->H3DPCK@0.05: 80.2 -> 81.4Improved
replace UniFrame pseudo-labelling with integrated selection (ablation)Ours · SURREAL->Human3.6MPCK@0.05: 79.8 -> 80.8Improved
add geometric reconstruction loss L_rec (ablation)Ours · RHD->H3DPCK@0.05: 81.6 -> 84.1Improved
add geometric reconstruction loss L_rec (ablation)Ours · SURREAL->Human3.6MPCK@0.05: 81.2 -> 83.7Improved
use warping estimation (bone maps) instead of heatmap-only reconstruction (ablation)Ours · RHD->H3DPCK@0.05: 81.6 -> 84.1Improved
heatmap-only reconstruction on top of pseudo-labelling (ablation)Ours · RHD->H3DPCK@0.05: 81.4 -> 81.6No clear effect
add temporal consistency loss on consecutive frames for fair comparison (ablation)RegDA · RHD->MVHandPCK@0.05: +0.6Improved
add temporal consistency loss on consecutive frames (ablation)UniFrame · SURREAL->Human3.6MPCK@0.05: +0.3Improved
frame stride [5, 15] versus [0, 5] (ablation)Ours · SURREAL->Human3.6MPCK@0.05: 81.8 -> 83.7Improved
frame stride [30, 50] versus [5, 15] (ablation)Ours · SURREAL->Human3.6MPCK@0.05: 83.7 -> 82.5Hurt
integrated pseudo-labelling selection vs UniFrame selection (ablation)Ours · MVHand pseudo-label accuracypercentage of correct pseudo-labels: 59% -> 72%Improved
add reconstruction loss for confident-but-inaccurate keypoints (ablation)Ours · Human3.6M train keypointspseudo-label PCKImproved

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