Monocular depth models (one image)

18 depth models. Each section says what the model is, whether it may be used commercially and how fast it is; licences and checks are under Details.

Input: one RGB image. Output: a depth or inverse-depth map. kind (a card field, result.kind) says what the numbers mean: metric = metres; scale_invariant, affine_invariant and affine_invariant_inverse are relative (unknown scale or shift, not metres). Metric models that use the focal length take model(image, intrinsics=K) (K [3, 3], pixels of the input): metric3d-vit-large and da3metric-large need it for correct metres (da3metric-large: metres = output x focal / 300; without intrinsics= a 300 px focal length is assumed, wrong for a real camera). unidepth-v2-vitl14, unik3d-vitl, depth-pro and MoGe estimate the camera themselves (given intrinsics= replace fx); zoedepth-nk and the relative models (da3mono-large too) take none.

Measured on RTX 5060 Ti, torch 2.14.1+cu130; fastforks bench MODEL --size HxW measures yours. The summary uses one input, 994x1482: kernels time with its speed-up over the faster original run, and kernels peak reserved memory with its ratio to the smaller original one. Where measured (one variant per family), a variant's table also lists portable under torch.compile and the fp16 TensorRT engine of portable (export; memory: the engine's device memory). – = not measured. The accuracy column is lower-is-better (NYUv2 AbsRel (scale+shift aligned)).

Other devices: Apple M1 Pro 16 GB, NVIDIA L4, NVIDIA L40S, NVIDIA A10, NVIDIA A100-SXM4-40GB, NVIDIA RTX PRO 6000; across devices.

All models

Modes, licences, training data and outputs per model

Commercial status per variant: yes if the code and the weights licence both allow commercial use (the training data is listed but does not change it; weights without a licence of their own are assumed to carry the code licence); no if either forbids it; unclear for custom licence terms (treat as no).

model	modes	code licence	weights licence	training data	commercial	outputs
da3-giant-1.1	portable, kernels	Apache-2.0	CC-BY-NC-4.0	unstated	no (weights: CC-BY-NC-4.0)	depth, confidence, intrinsics, mask, sky
da3-large-1.1	portable, kernels	Apache-2.0	CC-BY-NC-4.0	unstated	no (weights: CC-BY-NC-4.0)	depth, confidence, intrinsics, mask, sky
da3metric-large	portable, kernels	Apache-2.0	Apache-2.0	unstated	yes	depth, confidence, intrinsics, mask, sky
da3mono-large	portable, kernels	Apache-2.0	Apache-2.0	unstated	yes	depth, confidence, intrinsics, mask, sky
da3nested-giant-large-1.1	portable, kernels	Apache-2.0	CC-BY-NC-4.0	unstated	no (weights: CC-BY-NC-4.0)	depth, confidence, intrinsics, mask, sky
depth-anything-v2-b	portable, kernels	Apache-2.0	CC-BY-NC-4.0	unstated	no (weights: CC-BY-NC-4.0)	inverse_depth
depth-anything-v2-l	portable, kernels	Apache-2.0	CC-BY-NC-4.0	unstated	no (weights: CC-BY-NC-4.0)	inverse_depth
depth-anything-v2-s	portable, kernels	Apache-2.0	Apache-2.0	unstated	yes	inverse_depth
depth-pro	portable, kernels	LicenseRef-Apple-Sample-Code	LicenseRef-research-only	unstated	no (weights: LicenseRef-research-only)	depth, intrinsics
marigold-depth-v1-1	portable, kernels	Apache-2.0	LicenseRef-CreativeML-OpenRAIL-PlusPlus-M	unstated	yes	depth
metric3d-vit-large	portable, kernels	BSD-2-Clause	BSD-2-Clause (assumed from code)	unstated	yes	depth, normals, confidence
midas-dpt-beit-large-512	portable, kernels	MIT	MIT (assumed from code)	unstated	yes	inverse_depth
moge-2-vitl-normal	portable, kernels	MIT	MIT	unstated	yes	depth, points, intrinsics, mask, normals
moge-2-vits-normal	portable, kernels	MIT	MIT	unstated	yes	depth, points, intrinsics, mask, normals
moge-3-vitl	portable, kernels	MIT	MIT	unstated	yes	depth, points, intrinsics, mask, normals
unidepth-v2-vitl14	portable, kernels	CC-BY-NC-4.0	CC-BY-NC-4.0 (assumed from code)	unstated	no (code: CC-BY-NC-4.0 (noncommercial); weights: CC-BY-NC-4.0 (assumed from code))	depth, points, intrinsics, confidence
unik3d-vitl	portable, kernels	CC-BY-NC-SA-4.0	CC-BY-NC-SA-4.0 (assumed from code)	unstated	no (code: CC-BY-NC-SA-4.0 (noncommercial); weights: CC-BY-NC-SA-4.0 (assumed from code))	depth, points, confidence
zoedepth-nk	portable, kernels	MIT	MIT (assumed from code)	unstated	yes	depth

Training datasets