Evaluation and visualization

[zgojcic]

Hi @pengsongyou,

thank you for open-sourcing the code. I would have two questions regarding the evaluation:

• In the code I have observed that you have a function refine_mesh, which performs an optimization of the mesh after MC. Is this function used for the evaluation or visualizations? In all config files refine is set to False.
• What exactly does you mean with "To ensure a fair comparison to SPSR, we compare all methods with only a single surface for walls/ground planes when calculating Chamfer-L1 and F-Score.". Do you remove all but one floor/wall? Is this also reflected in the computation of the IoU?

Thanks in advance,

Zan

[pengsongyou]

Hi Zan @zgojcic ,

Thanks so much for your interest in our work.

• We modified the generation code directly from Occupancy Networks' code, and forgot to delete much stuff, e.g. refine_mesh. Therefore, refine_mesh is not used at all.
• Because SPSR with trimming gives reconstruction with only single surfaces for walls/ground planes, we cannot evaluate IoU for SPSR. In contrast, our method always reconstructs walls/ground planes with some thickness (double sides) and it is always watertight (so we can evaluate IoU). To fairly compare with SPSR on Chamfer and F-Score, we also keep only the inner surface and remove the outer surface for walls/ground planes. This is not reflected in the computation of IoU.

Best, Songyou

[zgojcic]

Hi Songyou,

this makes a lot of sense, thanks for the fast response.

Best Zan