conversion to surface mesh

[jordandekraker]

Is there a lossless way to go from normals or other model outputs, for example here

DeepFit/tutorial/compute_normals.py

Line 66 in f19b83c

n_est, beta, weights, trans, trans2, neighbors_n_est = model.forward(points)

to a triangulated mesh?

[sitzikbs]

Sure. You are looking for a Surface Reconstruction algorithm (which is outside the scope of this repo).
Several methods I recommend looking into:
Poisson surface reconstruction (and screen poison surface reconstruction).
DeepSDF, Occupancy networks, Neural splines, SAL, SALD, IGR.

I personally like SIREN, and their implementation.

I have a current paper on the topic but it is not published yet so the code will be available only after publication.

[jordandekraker]

Great, thanks for the speedy reply! I'll check these out.