Issue about function: get_plane_mask(depth_mask):

[326037785]

Potential Division-by-Zero Risk in get_plane_mask Function

The provided get_plane_mask function, specifically in the calculation of p1, p2, p3, and p4, lacks robust division-by-zero protection.

The calculation for each point (p1 through p4) involves dividing a sum of pixel values by the count of non-zero values in the third channel (depth_image[..., 2]).

# Example for p1:
p1 = np.sum(depth_image[:3, :3, :], axis=(0, 1)) / np.sum(
    depth_image[:3, :3, 2] != 0
)

[MaticFuc]

I will look into this. This did not cause problems during training and inference.

[326037785]

I will look into this. This did not cause problems during training and inference.

I discovered this issue during debugging: when the selected area consists entirely of zero-valued pixels, it leads to a division by zero in the normalization process, resulting in NaN values. For an alternative, I fill in zero to p_i， it works.