AI Paper Atlas

Research map / 3D vision / Human pose estimation

Progressive Guessing to Fixed Point: Rethinking Human Motion Prediction with Deep Equilibrium Models

Dong Wei, Huaijiang Sun, Fan Liu, Yuhui Zheng

CVPR 2026 · 3D vision › Human pose estimation · approach: Human motion prediction

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
use last observed frame as initial guess instead of observed motion (ablation)DEQ motion predictor · Human3.6M 400msMPJPE: 104.6 -> 55.3Improved
add equivariant geometric injection (ablation)MotionDEQ · Human3.6M 400msMPJPE: 64.0 -> 55.3Improved
add invariant pattern injection (ablation)MotionDEQ · Human3.6M 400msMPJPE: 61.3 -> 55.3Improved
use sparse fixed-point supervision instead of smoothed GT targets (ablation)MotionDEQ · 3DPW 1000msMPJPE: 120.59 -> 107.80Improved
use sparse fixed-point supervision instead of dense corrections (ablation)MotionDEQ · 3DPW 1000msMPJPE: 114.09 -> 107.80Improved
use truncated (phantom) gradient instead of inexact JFB gradient (ablation)MotionDEQ · Human3.6MMPJPEImproved
increase DEQ forward iterations up to about 10 (ablation)MotionDEQ · Human3.6MMPJPEImproved
share parameters across refinement stages (RNN-origin) without last-frame init (ablation)EqMotion · 3DPWAvg MPJPE: 67.07 -> 69.22Hurt
replace multi-stage EqMotion with MotionDEQEqMotion · Human3.6M 400msMPJPE: 55.9 -> 55.3Improved
replace multi-stage EqMotion with MotionDEQEqMotion · 3DPW averageMPJPE: 67.07 -> 65.95Improved
reuse previous fixed point as warm initial guessMotionDEQ · streaming Human3.6MDEQ iterations to convergeImproved
add equivariant correction adapter (ECA)MotionDEQ · streaming Human3.6MMPJPEMixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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