Question about the first demo on the project page (implementation details / hardware / method)

Hi,

First of all, I’m genuinely impressed by how good the first demo on your project page looks:
https://depth-anything-3.github.io/
(I mean the very first demo at the top of the page.)


To better reproduce it on my side, could you please share some concise technical details about this specific demo?

1. Implementation
- Is it running in real time or precomputed?
- What input resolution and approximate FPS?
- Any extra processing (e.g., temporal smoothing, post-processing)?

2. Hardware
- Which GPU model(s) are you using for this demo?
- Approximate GPU memory usage?
- Any important config details (batch size, FP16/BF16, etc.)?

3. Method / Code
- Is this demo using the open-sourced `da3-streaming` method as-is, or a modified/internal version?
- If it’s based on the released code, which script/config is used?
- Any notable differences compared to the public repo?

If you could share these details, I would really appreciate it. It would help a lot in reproducing a similar demo. Thank you again for the impressive work!

Start with the first demo on the project page and inspect the public da3-streaming method and its released scripts or configuration. The issue is complete when the implementation mode, resolution, FPS, processing, hardware, memory, configuration, and any differences from the public code are documented.

Written by the indexing model from the issue text.