Hi, thanks for the open-sourced project!
I try to use my own mesh (without texture) as the initialization of stage 3 and hope to optimize the corresponding SuGaR using default vsd loss, but the texture of final result seems bad
For the loss term, I only use the "2d" guidance.
out_2d = self.training_substep(batch, batch_idx, guidance="2d")
total_loss += out_2d["loss"]
# if self.guidance_control is not None and self.C(self.cfg.loss.lambda_sds_control) > 0:
# out_mvcontrol = self.training_substep(batch, batch_idx, guidance="mvcontrol")
# total_loss += out_mvcontrol["loss"]
# out_ref = self.training_substep(batch, batch_idx, guidance="ref")
# total_loss += out_ref["loss"]
I will highly appreciate it if you have any suggestion for improving the result.
Hi, thanks for the open-sourced project! I try to use my own mesh (without texture) as the initialization of stage 3 and hope to optimize the corresponding SuGaR using default vsd loss, but the texture of final result seems bad
For the loss term, I only use the "2d" guidance.
I will highly appreciate it if you have any suggestion for improving the result.