WorldFoundry
English中文

Depth Anything 3 Metric Prior

Method overview: a vanilla DINO transformer with within-view and cross-view attention, then a dual-DPT head for depth, rays, and points.

Depth Anything 3 Metric Prior from ByteDance (Seed): monocular-depth-estimation and metric-depth-estimation; Verified.

Edit this pageGitHubMarkdown

Depth Anything 3 Metric Prior is the metric-depth head of ByteDance Seed's Depth Anything 3, integrated as an independent geometry prior with verified runner status. The recorded checkpoint is depth-anything/DA3METRIC-LARGE, and it is the depth prior the Prior-Depth-Anything refiner was verified against in this repository.

Method overview: a vanilla DINO transformer with within-view and cross-view attention, then a dual-DPT head for depth, rays, and points.
Method overview: a vanilla DINO transformer with within-view and cross-view attention, then a dual-DPT head for depth, rays, and points.

The recorded paper is arXiv 2511.10647.

Project page: depth-anything-3.github.io. Code: ByteDance-Seed/Depth-Anything-3.

Same paper as the export card: arXiv 2511.10647. Project: depth-anything-3.github.io. Runner is verified on the shared prior contract. Runtime configs live under worldfoundry/data/models/runtime/configs/depth_anything_v3, separate from ViPE.

This is not the SOH export depth-anything-v3.

Build your run

Pipeline
DepthAnythingV3PriorPipelineworldfoundry.pipelines.geometry_priors.pipeline_geometry_priors
TaskDefault default
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  depth-anything-v3-prior \
  --pipeline.task-profile default \
  --json

What it is for

Pick depth-anything-v3-prior for feed-forward geometry: views in, reconstruction / trajectory / point cloud out. Use it when you need the catalog 3D route rather than a depth-only prior.

DA3 scales the Depth Anything recipe and adds metric and geometry heads. This prior entry wires the metric-large checkpoint; runtime configs live under runtime/configs/depth_anything_v3, separate from VIPE configs. It is published by ByteDance (Seed).

Catalog tasks include monocular-depth-estimation, metric-depth-estimation, and geometry-prior. Catalog aliases include dav3-prior and da3metric-large-prior. Documented modalities: image in, depth out.

Runtime contract

Optional input_path, focal_length, execute, and max_points; artifacts depth-anything-v3-prior.npz.

CLI task profiles: default. WorldFoundry binds DepthAnythingV3PriorPipeline.

Family status: Verified.

It is integrated as an independent geometry prior under worldfoundry/base_models and exposed through the shared geometry-prior pipeline/operator contract. Higher-level pipelines (e.g. ViPE) import it from this canonical owner instead of keeping private copies. Use the sibling depth-anything-v3 entry for the official SOH export route.

Integrated as an independent depth prior at worldfoundry/base_models/three_dimensions/depth/depth_anything/depth_anything_v3.

Running in WorldFoundry

The catalog launch is worldfoundry-eval run depth-anything-v3-prior --pipeline.task-profile default. The command builder above switches --pipeline.task-profile among recorded tasks and still emits worldfoundry-eval run.

It runs in the unified environment worldfoundry-unified-cu128 (Python 3.11, CUDA 12.8).

Recorded weight repositories include depth-anything/DA3METRIC-LARGE. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh depth-anything-v3-prior. Add --download to fetch missing public checkpoints.

Status: Verified.