Constructing ground-truth for custom dataset

[dqj5182]

I want to build ground-truth for my own custom meshes that are not ShapeNet meshes (e.g., YCB object models).
I need to extract Xct, Xbd, and Xtg, to train ShapeFormer model pretrained only on VQDIF.

First, I tried to use make_imnet_dataset function.
But, the function was too complex and some data such as generate_dataitem was ambiguous for me to get (could not understand how to get data such as "datasets/hsp_shapenet/modelBlockedVoxels256/04530566/cc3957e0605cd684bb48c7922d71f3d0.mat").

Then, I found out generate_gt_samples function that was used to extract ground-truth data for DFAUST dataset.

My questions are as follows:

1. Would it be okay to directly use the generate_gt_samples function to extract Xbd, Xtg, and Ytg of my custom mesh?
2. Should there be any part of the code that I should modify to process my custom mesh? (It seems that the code do not contain scaling of mesh so that mesh can be fitted into unit cube before constructing Xbd. Why does the code do not scale DFAUST mesh to unit cube & should I scale my custom mesh to unit cube before processing with generate_gt_samples function?
3. Is it okay to train ShapeFormer with ground-truth data generated in the format of DFAUST dataset (I am referring to utilizing generate_gt_samples function for my custom dataset) while VQDIF is trained in IMNet dataset format (I am referring to ShapeFormer model that loads pre-trained VQDIF model provided by this repo)?
4. If generate_gt_samples function is not applicable for this situation, how can I modify generate_gt_samples function to be applicable for this situation?