This work presents Depth Anything, a highly practical solution for robust monocular depth estimation by training on a combination of 1.5M labeled images and 62M+ unlabeled images.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results