Early-Signal Detector

method · first seen 4 Oct 2026

monocular depth estimation

Also: MDE, MiDaS, Marigold, monocular depth, monocular depth model, depth-derived structural priors, monocular depth foundation model

Predicting per-pixel depth from a single image using learned models.

Monocular depth estimation predicts a dense depth map from a single RGB image. It has been an active computer vision task for over a decade, with widely used models such as MiDaS, Marigold and the Depth Anything series. The sources are mostly arXiv papers that use or adapt existing depth foundation models. Examples include a frozen-CLIP depth decoder, scene-level semantic priors for metric scale, uncertainty-aware 'safe depth' for robot navigation, metric monocular SLAM built on Depth Anything 3, and navigation that tolerates prediction errors. Two hobby repos apply Depth Anything to retail analytics and 2D-to-3D video conversion. The volume reflects steady downstream use rather than a new idea.

The recurring thread is replacing depth sensors with off-the-shelf depth foundation models in robotics and SLAM. Several papers show this needs explicit handling of scale and errors to be safe, so that work is worth watching more than the umbrella concept itself.

Analyst note (established), 17 Nov 2026.

Mentions per week

01019 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 114 on 14 Nov 2026 · 31 mentions in 28 days · 31 independent groups · papers → code · first seen 6 weeks ago · acceleration +0.44

Earliest sources

Latest sources

Mentioned together with

segment anything modelGaussian splattingsimultaneous localization and mappingcontrastive language-image pretrainingfrozen-backbone feature decodingYOLOzero-shot segmentationreference-based object compositingvideo diffusion for robot navigationneural architecture search