joeylitalien
commented
2 years ago Hi @dhuwzj,
It's been quite a while so we can't exactly recall how this was computed. It is most likely just (# of active feature vectors per LOD) × (feature vector size) × (4 bytes for 64-bit floats) / 1024 (to convert to KB). AFAIK we didn't use half-precision floats for the table to be fair to other techniques. Note that the octree structure incurs an extra storage cost because we need to track of active voxels (i.e., the occupancy bits in Morton order at each level). I doubt this was included in the numbers of Table 1 back then, but this is quite minimal compared to the feature grid so the numbers should roughly stand.
If you want the "true" storage on disk, you can also save each octree level without the decoder network.
wzjwzj00
What is included in the storage volume(KB) of different LODs in Table 1 in your paper？Just the octree related data structures saved by the weight file？ Looking forward to your reply！