Remove inner mesh without refiner

[jjjkkyz]

Hi, I notice that you use a refiner network to remove the inner mesh introduing by sparse to dense conversion.
I think a better way to sovle it is using the mask attribute in scikit-image with a little modification in both repos.

Here I show a result of my current version with remove_interior=False.

In your repo (ss_vae.py line 122):

  mask[sparse_index_i[..., 0], sparse_index_i[..., 1], sparse_index_i[..., 2]] = True
  vertices, faces, _, _ = measure.marching_cubes(
      sdf.numpy(),
      mc_threshold,
      method="lewiner",
      mask = mask.numpy()
  )

In scikit-image repo (skimage/measure/_marching_cubes_lewiner_cy.pyx line 983)

if no_mask or (mask[z_st, y_st, x_st] and mask[z, y_st, x_st] and mask[z_st, y, x_st] and mask[z_st, y_st, x] and mask[z, y, x_st] and mask[z, y_st, x] and mask[z_st, y, x] and mask[z, y, x]):