search

steineggerlab / foldcomp

Compressing protein structures effectively with torsion angles
GNU General Public License v3.0
145 stars 14 forks source link

[Question] Why use NeRF over faster alternatives #26

Open a-r-j opened 1 year ago

a-r-j commented 1 year ago

Hi again,

I was wondering why you use the classical NeRF (as far as I can tell from the notes in the code) instead of faster alternatives like pNeRF and mp NeRF?

Thanks :grinning:

khb7840 commented 1 year ago

Hi there,

Thanks for your question! We actually checked those papers during development and currently our implementation of classical NeRF against pNeRF and mp-NeRF is just as fast as mp-NeRF (ours: 0.01s for avg 480 AAs including I/O on HDD, mp-nerf: 0.0091s for 500 AAs). We chose to apply file-level parallelization, which helps handling mis-formatted files. We know that we can get speed-up of single-file decompression and side-chain reconstruction from mp-NeRF. We're exploring a few different options.

Let us know if you have any further questions or concerns!