WorldFoundry
English中文

GeoCalib

Method overview: a network predicts a Perspective Field, then Levenberg–Marquardt fits camera parameters.

GeoCalib from ETH Zurich (CVG): camera-calibration and gravity-estimation; Verified.

Edit this pageGitHubMarkdown

GeoCalib estimates camera calibration and the gravity direction from a single image by combining geometric optimization with deep learning. In this catalog it is a verified camera-calibration prior consumed by higher-level pipelines (notably VIPE) through the shared geometry-prior contract.

Official paper teaser: GeoCalib estimates camera calibration from a single image by combining learning with geometry.
Official paper teaser: GeoCalib estimates camera calibration from a single image by combining learning with geometry.
Method overview: a network predicts a Perspective Field, then Levenberg–Marquardt fits camera parameters.
Method overview: a network predicts a Perspective Field, then Levenberg–Marquardt fits camera parameters.

The recorded paper is arXiv 2409.06704.

Project page: cvg.ethz.ch/research/geocalib. Code: cvg/GeoCalib.

Paper: arXiv 2409.06704. Project: cvg.ethz.ch/research/geocalib. Code: cvg/GeoCalib (Apache-2.0). Official release tarball: geocalib-pinhole.tar from the v1.0 GitHub release. Publisher: ETH Zurich (CVG). Aliases: geocalib, cvg-geocalib. Runner: verified. Per-run evidence stays outside the catalog.

Build your run

Pipeline
GeoCalibPriorPipelineworldfoundry.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 \
  geocalib-prior \
  --pipeline.task-profile default \
  --json

What it is for

geocalib-prior is the recorded reconstruction line. Start here to emit geometry from the documented views.

A neural network predicts geometric cues which a differentiable optimization layer turns into focal length and gravity estimates (per the upstream ECCV 2024 release). Registered at base_models/three_dimensions/general_3d/geocalib; apache-2.0. It is published by ETH Zurich (CVG).

Catalog tasks include camera-calibration, gravity-estimation, and geometry-prior. Catalog aliases include geocalib and cvg-geocalib.

Runtime contract

Optional input_path, camera_model, execute, max_points, and focal_length; artifacts geocalib-prior.json.

CLI task profiles: default. WorldFoundry binds GeoCalibPriorPipeline.

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. Runner parity is verified through the shared contract.

Integrated as an independent camera-calibration prior at worldfoundry/base_models/three_dimensions/general_3d/geocalib.

Running in WorldFoundry

The catalog launch is worldfoundry-eval run geocalib-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).

No public checkpoint id is recorded on this card; follow the usage notes for staging. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh geocalib-prior. Add --download to fetch missing public checkpoints.

Status: Verified.