Monocular and Stereo Depth Estimation
Estimating the distance of every pixel from one image, a video or a stereo pair.
Tier Grades
Tier A: Frontier AI labs
feasible
Data engines over about 62M images are big-tech scale.
Tier B: Indonesian AI labs
stretch
Fine-tuning released depth models on local scenes is feasible; the data engine is a stretch for an Indonesian lab.
Tier C: Institution units such as ITB
feasible
Fine-tuning on NYUv2 and KITTI, as in the anchor, fits a university unit.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Zero-shot depth from one camera helps measurement and volume estimation in proofs of concept.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTraining compute is not in the sources read.
- DataDepth Anything scales training to about 62M automatically annotated unlabelled images, then fine-tunes on NYUv2 and KITTI for metric depth.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 6 authors on the Depth Anything paper (arXiv 2401.10891).
- Time horizonMonths to a year.
- Shelf-life signalsFeed-forward 3D models predict depth and camera pose together, Metric depth works zero-shot on new scenes
- Pipeline roleTier A releases depth foundation models; Tier C fine-tunes for local metric depth; Tier D uses them for stockpile, tree height and distance estimates in pilots.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
Depth Provisionexemplar
Depth Anythingvisionexemplar
ZoeDepthvisionexemplar
FoundationStereovisionexemplar
Sapiens2visionproduced byDepth Anything 3visionproduced by
DepthCraftervisionproduced bySapiensvisionproduced byDepth Anything V2 Smallvisionproduced byDepth Anything V2 Largevisionproduced by
Marigold Depthvisionproduced byLDM3D3Dproduced byVPDvisionproduced by
NeWCRFsvisionproduced byDPT (Dense Prediction Transformer)visionproduced by
AdaBinsvisionproduced by
vid2depthvisionproduced byMiDaSvisionproduced by
FastDepthvisionproduced by
FCRN depthvisionproduced by
AR Portrait Depthvisionproduced by
CREStereovisionproduced byStereoNetvisionproduced byLinks
R&D TopicMonocular and Stereo Depth Estimation
ModelsDepth AnythingThe University of Hong Kong, visionZoeDepthIntel, visionDepth ProApple, visionFoundationStereoNVIDIA, vision+19 more
Depth AnythingThe University of Hong Kong, visionZoeDepthIntel, visionDepth ProApple, visionFoundationStereoNVIDIA, vision+19 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 23
Depth AnythingThe University of Hong Kong, ByteDance, visionexemplarDepth Anything V2 LargeThe University of Hong Kong, ByteDance, visionproduced bySapiensMeta, visionproduced bySapiens2Meta, visionproduced byDepth Anything 3visionproduced byDepth Anything V2 SmallThe University of Hong Kong, ByteDance, visionproduced byvid2depthUniversity of Texas at Austin, Google, visionproduced byDPT (Dense Prediction Transformer)Intel, visionproduced byFastDepthMassachusetts Institute of Technology, visionproduced byLDM3DIntel, 3Dproduced byMiDaSIntel, ETH Zurich, visionproduced byAdaBinsKing Abdullah University of Science and Technology, visionproduced by+11
AR Portrait DepthGoogle, visionproduced byCREStereoMegvii, visionproduced byDepth ProApple, visionexemplarDepthCrafterTencent, visionproduced byFCRN depthTechnical University of Munich, visionproduced byFoundationStereoNVIDIA, visionexemplarMarigold DepthETH Zurich, visionproduced byNeWCRFsAlibaba Group, visionproduced byStereoNetGoogle, visionproduced byVPDvisionproduced byZoeDepthIntel, visionexemplarPapers
- Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset TransferRené Ranftl et al., arXiv preprint, 2019 founding
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled DataLihe Yang et al., CVPR 2024 (per arXiv comment), 2024 state of the art
- Depth Anything V2Lihe Yang et al., arXiv preprint, 2024 state of the art
- Metric3Dv2: A Versatile Monocular Geometric Foundation Model for Zero-shot Metric Depth and Surface Normal EstimationMu Hu et al., IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 12,, 2024 state of the art
- Repurposing Diffusion-Based Image Generators for Monocular Depth EstimationBingxin Ke et al., CVPR 2024 (per arXiv comment), 2023 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2401.10891 https://arxiv.org/abs/2401.10891
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Monocular and Stereo Depth Estimation", https://aurion.semai.id/topics/monocular-and-stereo-depth-estimation/, records of October 2026.