UniDepth

Why we included this project

If your pipeline needs real distances in meters rather than relative depth, UniDepth is worth a look. It predicts metric depth from a single monocular image without requiring known camera intrinsics, so you are not locked into a fixed focal length assumption the way you often are with other depth models. The project has done well on standard benchmarks such as KITTI and NYU-Depth-V2, and both the V1 and V2 releases ship with inference code and pretrained weights, so getting predictions up and running takes little effort. There is also tooling to visualize depth maps and point clouds, which helps a lot when you want to sanity-check outputs before feeding them into downstream reconstruction or planning. One thing to weigh before adopting it: the code is under a non-commercial Creative Commons license.