Train my own gausssians

[starlordlxt]

Hi!
I am using Panohead to generate my own avatar and want to use the decoder you provided to convert it into a 3DGS model. I have generated a fintuned_generator.pkl file and a projected_w.npz file using Panohead. I trained the decoder with fintuned_generator.pkl using the code you provided, but how should I use the projected_w.npz to generate the desired portrait? I noticed that in your work, this aspect wasn't covered, and the "z" is mostly randomly generated. Thank you very much, and kudos to your work!