Depth Anything v2

A representative model in monocular depth estimation that relies on large-scale depth-annotated datasets for supervised training. It implicitly learns scene-scale priors through data distribution, and is widely used as a benchmark in the field.