Research map / 3D vision / Human pose estimation
SimPose: Effectively Learning DensePose and Surface Normals of People from Simulated Data
Tyler Zhu, Per Karlsson, Christoph Bregler
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 |
|---|---|---|---|
| train on simulated Renderpeople humans instead of SMPL-based simulated humans | SimPose · DensePose COCO minival | AP: 56.2 -> 57.3 | Improved |
| train with only simulated UV labels instead of real DensePose labels (vs DP-RCNN-cascade + keypoints) | SimPose (Renderpeople) · DensePose COCO minival | AP: 55.8 -> 57.3 | Improved |
| add simulated data to the training batch mixture (0% to 100%) (ablation) | SimPose · Renderpeople val | Normal ADD (lower better): 74.6 -> 19.3 | Improved |
| add simulated data to the training batch mixture (0% to 100%) (ablation) | SimPose · Renderpeople val | UV L2 (lower better): 0.386 -> 0.075 | Improved |
| increase simulated data share from 75% to 100% (ablation) | SimPose · Renderpeople val | UV L2: 0.075 -> 0.075 | No clear effect |
| increase simulated data share from 0% to 100% (ablation) | SimPose · MSCOCO 2D tasks | Keypoint OKS: 0.77 -> 0.32 | Hurt |
| increase simulated data share from 0% to 100% (ablation) | SimPose · MSCOCO 2D tasks | Segmentation IOU: 0.70 -> 0.26 | Hurt |
| add UV task loss (weight 0.00 to 0.25) (ablation) | SimPose · Renderpeople val | UV L2 (lower better): 0.384 -> 0.077 | Improved |
| increase UV task weight from 1.00 to 2.00 (ablation) | SimPose · Renderpeople val | UV L2 (lower better): 0.074 -> 0.071 | Improved |
| raise UV task weight from 0.00 to 2.00 (ablation) | SimPose · MSCOCO 2D tasks | Segmentation IOU: 0.698 -> 0.697 | No clear effect |
| add 3D normal task loss (weight 0.00 to 0.25) (ablation) | SimPose · Renderpeople val | Normal ADD (lower better): 76.1 -> 22.1 | Improved |
| increase 3D normal task weight from 1.00 to 2.00 (ablation) | SimPose · Renderpeople val | Normal ADD (lower better): 19.9 -> 18.6 | Improved |
Closest papers by meaning
- DensePose: Dense Human Pose Estimation in the Wild (CVPR 2018 · similarity 0.93)
- UltraPose: Synthesizing Dense Pose With 1 Billion Points by Human-Body Decoupling 3D Model (ICCV 2021 · similarity 0.91)
- DenseRaC: Joint 3D Pose and Shape Estimation by Dense Render-and-Compare (ICCV 2019 · similarity 0.90)
- Learning 3D Human Shape and Pose from Dense Body Parts (TPAMI 2020 · similarity 0.89)
- Learning Visibility for Robust Dense Human Body Estimation (ECCV 2022 · similarity 0.89)
- MeshPose: Unifying DensePose and 3D Body Mesh Reconstruction (CVPR 2024 · similarity 0.87)
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.
