Further training from previous result (i.e. from PLY)

[nerpa87]

Hello, gsplat is great and fast, however it's not clean if some kind of loading from previous state and training continuation implemented.
Could you advice how to do further training, having PLY file with gaussians and colmap model?

[soumya997]

Im still experimenting, but I think you can use the method mentioned here

[nerpa87]

Thanks for reply, currently I can load and save PLY with gaussians but it seems it's not enough to resume training from saved state. Result gets worse, I suggest it's because of optimizer params should be saved as well to correctly resume training