Based on in-body 2D pose point detection of MVSPIN, real 3D pose points are obtained through clustering algorithm and real camera matrix. With reference to Vae, a SMPL-VAE human regression model combining step training and joint training is proposed. The model corrects Smpl modeling results according to Vae training results. While maintaining the overall proportion, it is more in line with the local modeling of human motion structure.
Ubuntu 18.04
Python 3.7.4
cuda 11.1
cudnn 8.0.5
Pytorch 1.10.0 + cu111
Download dataset gt from: https://pan.baidu.com/s/1MSDj2DHkTy669yOGLmKUEg. code: 6666.
Save them in gt
Download basicModel_neutral_lbs_10_207_0_v1.0.0.pkl from: https://pan.baidu.com/s/1MSDj2DHkTy669yOGLmKUEg. code: 6666.
Save basicModel_neutral_lbs_10_207_0_v1.0.0.pkl in smplpytorch/native/models
run example main.py
Our work has achieved satisfactory results in pedestrian detection and modeling
MPJPE - Mean Per Joint Position Error PCK - Percentage of Correct Keypoints
| method | MPJPE | PCK |
|---|---|---|
| SMPLify | 109.30 | 88.76 |
| HMR | 114.62 | 83.26 |
| Shape-aware | 105.62 | 85.35 |
| Ours | 101.74 | 92.26 |