monocular depth estimation

Estimating per-pixel depth from a single image, with no second camera view to triangulate from.

Monocular depth estimation predicts per-pixel depth from a single image, with no stereo pair or known baseline to triangulate from -- relying instead on learned monocular depth cues (relative object size, occlusion, texture gradients) a trained model has learned to exploit.

Examples

From a single street photo with no stereo pair, a monocular depth model still correctly ranks a nearby parked car as closer than a distant building, using cues like relative size and occlusion.