Research map / 3D vision / Human pose estimation
Delving Deep Into Hybrid Annotations for 3D Human Recovery in the Wild
Yu Rong, Ziwei Liu, Cheng Li, Kaidi Cao, Chen Change Loy
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 dense correspondence supervision to full 3D (ablation) | image-only model · UP-3D | PVE (mm): 127.4 -> 123.7 | Improved |
| replace paired in-the-wild 3D with dense correspondence (both with sparse 2D) (ablation) | image-only model · UP-3D | PVE (mm): 203.2 -> 137.7 | Improved |
| add dense correspondence to 20% 3D (ablation) | image-only model · UP-3D | PVE (mm): 127.5 vs 127.4 (full 3D) | No clear effect |
| add IUV map as auxiliary input (sparse 2D supervision only) (ablation) | image-only model · UP-3D | PVE (mm): 203.2 -> 197.3 | Improved |
| add body segmentation as auxiliary input (sparse 2D supervision only) (ablation) | image-only model · UP-3D | PVE (mm): 203.2 -> 201.2 | Improved |
| add IUV map as input with 3D & dense & sparse 2D supervision (ablation) | image-only model · UP-3D | PVE (mm): 123.7 -> 122.4 | Improved |
| refine IUV maps by removing noisy predictions (ablation) | dense keypoint supervision · UP-3D | PVE (mm): 20.1% performance drop without refinement | Improved |
| keep paired in-the-wild 3D annotations (20% vs 0%) (ablation) | image-only model · UP-3D | PVE (mm): 203.2 -> 136.0 | Improved |
| add constrained 3D (Human3.6M) with no in-the-wild 3D (ablation) | sparse 2D only model · UP-3D | PVE (mm): 230.9 -> 203.2 | Improved |
| add constrained 3D with dense & sparse 2D supervision (ablation) | image-only model · UP-3D | PVE (mm): 173.4 -> 137.3 | Improved |
| add constrained 3D when full in-the-wild 3D exists (ablation) | image-only model · UP-3D | PVE (mm): 128.9 -> 127.4 | No clear effect |
| train with dense correspondence instead of HMR on COCO | Ours-DC vs HMR · COCO-DensePose | DKD (mm): 102.7 -> 51.8 | Improved |
Closest papers by meaning
- End-to-End Recovery of Human Shape and Pose (CVPR 2018 · similarity 0.88)
- DensePose: Dense Human Pose Estimation in the Wild (CVPR 2018 · similarity 0.88)
- Unite the People: Closing the Loop Between 3D and 2D Human Representations (CVPR 2017 · similarity 0.88)
- MultiPly: Reconstruction of Multiple People from Monocular Video in the Wild (CVPR 2024 · similarity 0.88)
- GeneMAN: Generalizable Single-Image 3D Human Reconstruction from Multi-Source Human Data (NeurIPS 2025 · similarity 0.86)
- PICO: Reconstructing 3D People In Contact with Objects (CVPR 2025 · similarity 0.86)
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.
