You ask AI: How far is that tree? And your model says... what exactly?

In 3D, “depth” is often just an impression - relative numbers without metric meaning. But in practice, we often need metric output.

We ran 8 leading monocular models on real iPhone footage, grounded by actual LiDAR measurements. Each model processed the same scene, same calibration, same target distances. We kept everything constant except the model:

• Some gave cinematic heatmaps but couldn’t guess distance
• Others nailed relative depth but broke down at scale
• But UniK3D predicted metric depth directly! Yes, it’s a research-grade tool, not easy to adopt commercially due to its license but it's a clear signal of what’s possible.

If you're deciding which depth model to trust or avoid this comparison might save you a few misfires.

Want to see it in action on your own data? Contact us or get full report - here

Results