First please create an appropriate environment using conda:
conda env create -f environment.yml
conda activate vhoi
Download Data (cite from 2G-GCN:https://github.com/tanqiu98/2G-GCN)
Please download the necessary data from the link below, and put the
downloaded data folder in this current directory (i.e. ./data/...).
Link: data.
To train the model from scratch, first sh tt_[dataset].sh for stage1, then
edit the conf/config_[dataset]_[number].yaml file for stage2, and run sh tt_[dataset].sh.
Examples on MPHOI-72: when you get pre-trained models for all subject groups, you can get the cross-validation result by python -W ignore predict.py --pretrained_model_dir ./outputs_hiergat/mphoi/2G-GCN/hs512_e40_bs8_lr0.0001_0.3_Subject14 --cross_validate.