About DA3-Long

Hi, thanks for your great work on Depth Anything 3!

I have two questions:

1. When will the code for DA3-Long be released?

I saw the mention of the “long” version in the paper and demos, and I’m wondering if there is an expected timeline for releasing the code or pretrained models for DA3-Long.

2. How to obtain metric-scaled point clouds when using DA3-Long without the pose-conditioned mode?

If I don’t use the pose-conditioned mode, but I do have prior poses or trajectories that already contain metric scale (e.g., from GPS/INS/visual odometry), is there a way to enforce or recover metric scale in the depth predictions produced by DA3-Long?

In other words, is there a recommended way to combine DA3-Long with externally scaled poses so that the resulting point clouds also maintain real-world scale?

Thanks!

Review the paper and demos referenced in the issue alongside the current project documentation and release information. A useful outcome would be a maintainer-confirmed DA3-Long release status and clear guidance on whether externally scaled poses can produce metric-scaled point clouds without pose-conditioned mode.

Written by the indexing model from the issue text.