Research map / 3D vision / Human pose estimation
AnyLift: Scaling Motion Reconstruction from Internet Videos via 2D Diffusion
Hongjie Li, Heng Yu, Jiaman Li, Hong-Xing Yu, Ehsan Adeli, C. Karen Liu, Jiajun Wu
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 hybrid data-source training (local reprojected poses from GVHMR) to AnyLift (ablation) | AnyLift · Gymnastics Internet videos | J2D: 36.1 -> 21.6 | Improved |
| add hybrid data-source training (ablation) | AnyLift · Gymnastics Internet videos | FID: 11.2 -> 10.9 | Improved |
| add hybrid data-source training (ablation) | AnyLift · Martial Arts Internet videos | J2D: 25.2 -> 15.1 | Improved |
| add hybrid data-source training (ablation) | AnyLift · Martial Arts Internet videos | FID: 4.3 -> 3.6 | Improved |
| human study: prefer full AnyLift over version without hybrid training (ablation) | AnyLift · Human study (Internet videos) | preference for ground contact / motion quality: 65.6% ground contact, 66.3% motion quality | Improved |
| camera-conditioned diffusion with hybrid training instead of MVLift | AnyLift vs MVLift · Gymnastics Internet videos | J2D: 33.1 -> 21.6 | Improved |
| AnyLift instead of MVLift | AnyLift vs MVLift · Martial Arts Internet videos | J2D: 24.6 -> 15.1 | Improved |
| AnyLift instead of MVLift under synthetic dynamic cameras | AnyLift vs MVLift · AIST++ (dynamic camera) | MPJPE: 122.1 -> 109.3 | Improved |
| AnyLift instead of MVLift under static cameras | AnyLift vs MVLift · AIST++ (static camera) | PA-MPJPE: 79.2 -> 82.3 | Hurt |
| AnyLift instead of MVLift under static cameras | AnyLift vs MVLift · AIST++ (static camera) | root translation error Troot: 67.6 -> 64.9 | Improved |
| AnyLift (trained on 2D only) instead of GVHMR (trained on AMASS 3D) | AnyLift vs GVHMR · AIST++ (static camera) | MPJPE: 97.6 -> 108.0 | Hurt |
| AnyLift instead of GVHMR | AnyLift vs GVHMR · Gymnastics Internet videos | foot sliding FS: 0.061 -> 0.152 | Hurt |
Closest papers by meaning
- Real2Sim in HOI: Toward Physically Plausible HOI Reconstruction from Monocular Videos (NeurIPS 2026 · similarity 0.90)
- Lifting Motion to the 3D World via 2D Diffusion (CVPR 2025 · similarity 0.90)
- InterAct: Advancing Large-Scale Versatile 3D Human-Object Interaction Generation (CVPR 2025 · similarity 0.90)
- BEDLAM2.0: Synthetic humans and cameras in motion (NeurIPS 2025 · similarity 0.88)
- VideoMDM: Towards 3D Human Motion Generation From 2D Supervision (NeurIPS 2026 · similarity 0.88)
- Motions as Queries: One-Stage Multi-Person Holistic Human Motion Capture (CVPR 2025 · similarity 0.88)
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.
