Questions on Pre-processing ShapeNet for UDF and SDF Training #8

Hello,

I have a few questions regarding the pre-processing of the ShapeNet dataset, particularly in relation to handling non-watertight meshes in ShapeNet datset and the extraction of SDF data:

1. UDF Training with Non-Watertight Meshes
   I noticed that the pre-processing code directly samples near points for UDF training. Since UDF focuses on the distance, even if the mesh is not watertight (affecting mainly the normal direction rather than the actual distance), is it correct to assume that ShapeNet dataset will not perform watertight pre-processing in formal training?

2. SDF Data Extraction for Surf-D
   In the original paper (Appendix, Fig. 11), SDF data is used for training Surf-D, and the ground truth data appears somewhat blurred. Did you pre-process the ShapeNet cars into a watertight formulation before extracting the SDF data for the experiment? In other words, if I want to use SDF data from ShapeNet for training, must I first convert the meshes into a watertight form?

3. Recommended Libraries/Tools
   If a watertight formulation is necessary for SDF extraction, do you have any recommendations for libraries or tools to convert ShapeNet meshes into watertight versions and extract the SDF?

Thank you! Thanks for your answer and the awesome work!