Questions on the reconstruction of the space website demo v.s. self-made reconstruction.

Hi, great work and fancy demo!
Tried the **same picutres** both on the [website demo from huggingface](https://huggingface.co/spaces/depth-anything/depth-anything-3) and self-made reconstruction. I see the web demo result is much more smoother and spatially better. (by the way, confidence threshold here is >=2 and the web demo is set to be >=10.)

top-down:

Anyways, the demo in the website is amazing. Would the author kindly share some code/thought of the post-process of reconstructing? or refer to some useful tools to help this process? THX!!

Start by comparing the self-made reconstruction with the Hugging Face Depth Anything 3 demo using the same pictures and the reported confidence thresholds (>=2 versus >=10). The issue does not name files, tests, or an entry point; completion would require a maintainer-defined explanation of the demo's post-processing or pointers to relevant reconstruction tools.

Written by the indexing model from the issue text.