Hello, thank you for sharing the code for this amazing research!
I have two questions:
The initial parameter values for the layer that embeds 3D points (UniDirsEmbed) in the code are fixed.
During the forward pass of the NERF model, the embedded input is split into two parts (emb_size1, emb_size2) and placed in different locations.
Could you explain about these two things? Would changing them according to the iMap settings yield similar performance?
The UniDirsEmbed is adopted from MipNeRF-360's embedding, where the fixed matrix is the basis Eq. 18 from the paper. Any other positional embedding could work as well.
The emb_size2 is for inputting a slightly higher frequency only for the color head to get more details. It gives a small improvement from our experiments but should perform similarly.
Hello, thank you for sharing the code for this amazing research!
I have two questions:
Thank you!