Discrepancy between predicted $f_x/f_y$ ratio and GT intrinsics when inferring metric depth with DA3NESTED-GIANT-LARGE

Hi,

Thx for your great work!

I am using `DA3NESTED-GIANT-LARGE` to infer metric depth. I have the **Ground Truth (GT) intrinsics**, but I noticed the model's predicted intrinsics (especially the $f_x / f_y$ ratio) differ significantly from the GT.

**My question:**
To get the correct metric depth aligned with my GT intrinsics, which approach is correct?

1. **Directly replace intrinsics:** Ignore the predicted intrinsics and pair the raw predicted depth with my GT intrinsics.
2. **Scale the depth:** Apply a scale factor to the depth map based on the focal length difference ($Depth_{new} = Depth_{pred} \times \frac{f_{gt}}{f_{pred}}$).

Thanks!

Start with the DA3NESTED-GIANT-LARGE inference entry point or documentation and verify how predicted intrinsics and raw depth are defined. Reproduce the reported fx/fy comparison against the GT intrinsics, then document which alignment approach the project supports; the issue names no specific file or test.

Written by the indexing model from the issue text.