README.md

CLIFF

Introduction

We provide the config files for CLIFF: CLIFF: Carrying Location Information in Full Frames into Human Pose and Shape Estimation.

@Inproceedings{li2022cliff,
  author    = {Li, Zhihao and
               Liu, Jianzhuang and
               Zhang, Zhensong and
               Xu, Songcen and
               Yan, Youliang},
  title     = {CLIFF: Carrying Location Information in Full Frames into Human Pose and Shape Estimation},
  booktitle = {ECCV},
  year      = {2022}
}

Notes

• SMPL v1.0 is used in our experiments.
• J_regressor_extra.npy
• J_regressor_h36m.npy
• pascal_occluders.npy
• resnet50_a1h2_176-001a1197.pth
• resnet50_a1h2_176-001a1197.pth(alternative download link)

Download the above resources and arrange them in the following file structure:

mmhuman3d
├── mmhuman3d
├── docs
├── tests
├── tools
├── configs
└── data
    ├── checkpoints
    │   ├── resnet50_a1h2_176-001a1197.pth
    ├── body_models
    │   ├── J_regressor_extra.npy
    │   ├── J_regressor_h36m.npy
    │   ├── smpl_mean_params.npz
    │   └── smpl
    │       ├── SMPL_FEMALE.pkl
    │       ├── SMPL_MALE.pkl
    │       └── SMPL_NEUTRAL.pkl
    ├── preprocessed_datasets
    │   ├── cliff_coco_train.npz
    │   ├── cliff_mpii_train.npz
    │   ├── h36m_mosh_train.npz
    │   ├── muco3dhp_train.npz
    │   ├── mpi_inf_3dhp_train.npz
    │   └── pw3d_test.npz
    ├── occluders
    │   ├── pascal_occluders.npy
    └── datasets
        ├── coco
        ├── h36m
        ├── muco
        ├── mpi_inf_3dhp
        ├── mpii
        └── pw3d

Training

Stage 1: First use resnet50_pw3d_cache.py to train.

Stage 2: After around 150 epoches, switch to resume.py by using "--resume-from" optional argument.

Results and Models

We evaluate HMR on 3DPW. Values are MPJPE/PA-MPJPE.

Config	3DPW	Download
Stage 1: resnet50_pw3d_cache.py	48.65 / 76.49	model | log
Stage 2: resnet50_pw3d_cache.py	47.38 / 75.08	model | log