build_nerf_to_ours and number of basis

[derrick-xwp]

Dear author,

Why do you need the nerf_to_ours function

nex-code/utils/sfm_utils.py

Line 320 in eeff38c

def nerf_pose_to_ours(cam):

?

Do you save the basis for all viewing directions for rendering? or only the reference view?

Thanks

[pureexe]

For the explanation of nerf_pose_to_ours,
please see #13

For saving,
the coefficient (k_n) is predicted at the reference view only.
However, the basis (H_n) is predicting from the viewing direction [-0.7,0.7] in the x-axis and [-0.6,0.6] in the y-axis.

[derrick-xwp]

Could you give a reference to the webgl renderer about how the basis is used?

[pureexe]

I'm not sure what are you looking for?

maybe this fragment shader.

https://github.com/nex-mpi/nex-mpi.github.io/blob/1714c80808eda528ff49f4e037f4dc7d06875470/viewer/js/scene_learned_basis8.js#L301-390