YvanYin/Metric3D

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

2302 stars · 171 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Metric3D is the official PyTorch implementation of Metric3Dv1 and Metric3Dv2, monocular geometric foundation models that predict metric depth and surface normals from a single image with zero-shot transfer. It supports canonical camera space transformation, depth-normal joint estimation, and applications like 3D reconstruction and SLAM.

Use cases

• estimate metric depth from a single image
• predict surface normals from photos
• zero-shot depth estimation on new scenes
• reconstruct 3D scenes from one image
• feed depth maps into SLAM pipelines
• benchmark monocular depth estimation on NYU and KITTI

When to choose

• you need metric (absolute-scale) depth rather than relative depth from a single image
• you want state-of-the-art zero-shot depth and normal estimation
• you need a pretrained geometric foundation model for downstream 3D tasks

When to avoid

• you need real-time depth on edge devices with limited compute
• you only need relative/monocular depth without metric scale
• you need multi-view or stereo depth estimation

Facets

library · maturity active

machine-learning deep-learning computer-vision image-processing computer-vision machine-learning artificial-intelligence python cross-platform monocular-depth-estimation metric-depth surface-normal-estimation zero-shot 3d-reconstruction pytorch foundation-model single-image gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="YvanYin/Metric3D")