Regarding volume of training dataset

[Zou-2004]

Dear authors,

Thank you very much for your great work and for releasing the code.

I am currently using your pipeline to process the PartNet-Mobility and GRScene datasets. I follow the same train/test split for PartNet-Mobility and apply the same part-count filtering for GRScene (keeping assets with fewer than 16 parts).

After this filtering, I obtain around 4,000 assets in total. However, in the paper, the reported number is approximately 3,800 assets.

Could you please clarify whether there are any additional filtering steps beyond the PartNet-Mobility split and the part-number constraint on GRScene (e.g., removing invalid meshes, missing joint annotations, articulation failures, etc.)? This would help me verify whether I am missing any preprocessing steps in my implementation.

Thank you very much for your time.