Harmonious Feature Learning for Interactive Hand-Object Pose Estimation

Directory

${ROOT}  
|-- data  
|   |-- HO3D
|   |   |-- data
|   |   |   |-- train
|   |   |   |   |-- ABF10
|   |   |   |   |-- ......
|   |   |   |-- evaluation
|   |   |   |-- train_segLable
|   |   |   |-- ho3d_train_data.json
|   |-- DEX_YCB
|   |   |-- data
|   |   |   |-- 20200709-subject-01
|   |   |   |-- ......
|   |   |   |-- object_render
|   |   |   |-- dex_ycb_s0_train_data.json
|   |   |   |-- dex_ycb_s0_test_data.json

Data

You need to follow directory structure of the data as below.

Download HO3D(version 2) data data
Download DexYCB data data
Download the process data data

Pytorch MANO layer

For the MANO layer, I used manopth. The repo is already included in manopth.
Download MANO_RIGHT.pkl and MANO_LEFT.pkl from here and place at assets/mano_models.

Train

HO3d

sh sh/train_ho3d.sh

Dex-ycb

sh sh/train_dex-ycb.sh

Test

HO3d

sh sh/train_ho3d_test.sh

Dex-ycb

sh sh/train_dex-ycb_test.sh

Acknowledgments

We thank:

About

cvpr23 : Harmonious Feature Learning for Interactive Hand-Object Pose Estimation

Languages

Language:Python 99.1%Language:Shell 0.9%