Open yanmie-1122 opened 1 month ago
I use the 2D CNN decoder to output the feature tri-planes.
Thank you very much for your timely reply. So in this experimental setup, what you ultimately need to store is a 2D CNN decoder, latent vectors, hash tables, and a standard NERF-like MLP network?
Yes.
I think I understand. Thank you for your answer.
In the experimental part of your paper, you mentioned that the experimental configuration of feature tri-planes + hash tables + single NeRF was used, which is the fifth line of Table 2. I'd like to know how you get the feature tri-planes because using single-shared-MLP seems to mean that you don't need to use the 2D CNN decoder, so how do you build the feature tri-planes? Thank you very much for your answer.