What a depth map actually is
A depth map is a grayscale representation of a scene where pixel brightness encodes distance from the camera — near objects appear bright, distant objects appear dark. It's the data layer underneath most modern 3D conversion.
How AI learned to estimate depth from one image
Monocular depth-estimation models are trained on millions of image pairs where true depth is known, learning to recognize cues like relative object size, occlusion, texture gradient and perspective lines — the same cues human vision uses unconsciously.
Where depth maps are used beyond stereo photos
- 2D-to-3D photo and video conversion
- Portrait mode background blur on smartphones
- Autonomous vehicle and robotics obstacle detection
- AR object placement and occlusion
- 3D scene reconstruction from ordinary video
The current limits
AI depth estimation still struggles with reflective and transparent surfaces, fine detail like hair, and scenes with unusual or ambiguous lighting — the same edge cases that make human depth perception occasionally get fooled, too.
See it in action at 3dstreaming.org