This repository includes hand(mano) version of VPoser.

2021.12.02 first version updated. However, the code is not well trimmed yet.

Requirements

• Python 3.8
• Pytorch 1.11.0
• opencv-python
• pyrender
• chumpy
• pytorch-lightning
• dotmap

Clone this repo and run the following from the root folder:

python install -r requirements.txt
python setup.py develop

Revise main/config.yaml file

Essential

• general.dataset_dir : the path of dataset directory that includes FreiHand and Interhand
• general.mano_dir : the path of dataset directory that includes mano pkl file

Optional

• train_params.*
• val_params.*
• model_params.*

$data_repository
        |-InterHand
        |   |-annots
        |       |-train
        |       |   |-InterHand2.6M_train_MANO_NeuralAnnot.json
        |       |   |- ...                
        |       |-test
        |       |   |-InterHand2.6M_test_MANO_NeuralAnnot.json
        |       |   |- ...
        |       |-val
        |           |-InterHand2.6M_val_MANO_NeuralAnnot.json
        |           |- ...             
        |-FreiHand   
            |-training_mano.json
            |- ...

import human_hand_prior as HPoser

hposer = HPoser.create()
latent_hand_pose = ... # torch [1, 32] tensor following gaussian distribution
hand_pose = hposer.decode(latent_hand_pose)['hand_pose']