On Mono Depth Testing, DA3MONO-LARGE Outperforms DA3NESTED-GIANT-LARGE?

Hi, @haotongl ,
I tested mono depth on the IBims1 and Sintel datasets and observed that DA3MONO-LARGE outperforms DA3NESTED-GIANT-LARGE. Specifically:
Depth accuracy: DA3NESTED-GIANT-LARGE generally shows lower depth and edge precision compared to DA3MONO-LARGE:

Edge blurring: DA3NESTED-GIANT-LARGE occasionally produces blurry edges :

Additionally, in the tested metrics, DA3MONO-LARGE consistently performs better:

Relative accuracy on the DA2K dataset: DA3NESTED-GIANT-LARGE = 94.30%, DA3MONO-LARGE = 97.86%,

SI_boundary_f1 edge metric on the Sintel dataset: DA3NESTED-GIANT-LARGE = 0.331655, DA3MONO-LARGE = 0.388513,

Could you provide any insights on why DA3NESTED-GIANT-LARGE underperforms in these mono depth tests compared to DA3MONO-LARGE?