Details about the consist loss

[trshen925]

Dear open-source community and contributors:

Thanks to the contributors for open-sourcing this amazing repository! I find there is a loss called $L_{consist}$ in Supp A.2, which is describe as "penalizes the variance of attributes associated with the same voxel vertex". How it implement? If you haven't prepared code, you could briefly describe in nature language. Thank you!