Monocular Depth & Geometry Estimation
A depth map is a single-channel image where each pixel is a distance from the camera. Predicting it from one RGB frame used to be impossible without stereo or LiDAR. In 2026 a frozen ViT encoder plus a lightweight head gets within a few percent of ground truth. Distinguish relative and metric depth and state which one each production model (MiDaS, Marigold, Depth Anything V3, ZoeDepth) solves. Use Depth Anything V3 (DINOv2 backbone) to predict depth for arbitrary single images with no calibration. Explain why monocular depth works at all from a single image (perspective cues, texture gradients, learned priors) and what it cannot recover (absolute scale, occluded geometry). Lift 2D detections to 3D points using a depth map and pinhole camera intrinsics. Depth is the missing axis in 2D computer vision. Given RGB, you know where things appear in the image plane; you do not know how far they are. Depth sensors (stereo rigs, LiDAR, time-of-flight) solve this directly but are expensive, fragile, and limited in range. Monocular depth estimation — predicting depth from a single RGB frame — used to produce blurry, unreliable output. By 2026 large pretrained encoders changed that: Depth Anything V3 uses a frozen DINOv2 backbone and produces depth maps that generalise across indoor, outdoor, medical, and satellite domains. Marigold reframes depth as a conditional…
Monocular Depth & Geometry Estimation: A depth map is a single-channel image where each pixel is a distance from the camera. Predicting it from one RGB frame…
This free lesson is part of the AI Engineering from Scratch curriculum. Read the full explanation, run the lesson code, and verify the result in the interactive reader or from the repository source.
Browse the complete course catalog or open this lesson on GitHub.