Hi, thanks for the great work!
I noticed that Track4World uses monocular depth estimators (DepthAnythingV3 / MoGe / Pi3) as its geometry backbone. Since monocular depth estimation is inherently scale-ambiguous, I'm wondering:
Does the output flow3d carry metric (physically meaningful) scale, or is it up-to-scale only?
Specifically:
- Is the unit of the 3D coordinates (e.g., meters) meaningful across different videos?
Hi, thanks for the great work!
I noticed that Track4World uses monocular depth estimators (DepthAnythingV3 / MoGe / Pi3) as its geometry backbone. Since monocular depth estimation is inherently scale-ambiguous, I'm wondering:
Does the output
flow3dcarry metric (physically meaningful) scale, or is it up-to-scale only?Specifically: