Hello author,
I have been testing restoration effects on my own 3DGS artifact data, comparing the results between the original Difix model and the Fixer model. However, I found that the restoration quality of Fixer did not meet my expectations; specifically, its performance on distant trees is inferior to that of Difix3D.
I noticed a parameter in the list for enabling vae-skip-connection, but the inference failed when I attempted to use it. I would like to clarify whether the currently open-sourced weights actually include the VAE skip connection architecture.
Looking forward to your clarification. Thank you!
Hello author,
I have been testing restoration effects on my own 3DGS artifact data, comparing the results between the original Difix model and the Fixer model. However, I found that the restoration quality of Fixer did not meet my expectations; specifically, its performance on distant trees is inferior to that of Difix3D.
I noticed a parameter in the list for enabling vae-skip-connection, but the inference failed when I attempted to use it. I would like to clarify whether the currently open-sourced weights actually include the VAE skip connection architecture.
Looking forward to your clarification. Thank you!