AI Paper Atlas

Research map / 3D vision / Depth and stereo

GeoCalib: Learning Single-image Calibration with Geometric Optimization

Alexander Veicht, Paul-Edouard Sarlin, Philipp Lindenberger, Marc Pollefeys

ECCV 2024 · 3D vision › Depth and stereo · approach: Scene coordinate regression for localization · 39 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train the network end-to-end by supervising the LM optimization output (ablation)GeoCalib · OpenPano testGravity median error (deg): 3.03 -> 2.65Improved
add learned uncertainties (confidences) to the perspective field (ablation)GeoCalib · OpenPano testGravity median error (deg): 2.65 -> 2.24Improved
replace SegFormer backbone with SegNeXt (ablation)GeoCalib · OpenPano testGravity median error (deg): 2.24 -> 1.88Improved
replace ParamNet regression head with second-order LM optimization (no end-to-end training) (ablation)PerspectiveNet+ParamNet · OpenPano testFoV median error (deg): 4.23 -> 4.58 in-domain; MegaDepth 11.23 -> 8.70Mixed
replace Adam/SGD optimization with LM optimization out of the box (ablation)GeoCalib ablation · OpenPano testGravity median error (deg): 3.03 -> 3.03No clear effect
train on OpenPano (about 5x smaller) instead of 360 Cities (ablation)PerspectiveNet+ParamNet · MegaDepthGravity median error (deg): OpenPano 4.99 -> 3.08; MegaDepth 4.06 -> 4.51Mixed
use heuristic or solver initialization instead of trivial initialization (ablation)GeoCalib · OpenPano testGravity median error (deg): 1.88 -> 1.90 (heuristic), 2.03 (solver)No clear effect
fix known FoV in the optimization (partial calibration) (ablation)GeoCalib · Stanford2D3DGravity median error (deg): 1.12 -> 0.86Improved
fix known gravity in the optimization (partial calibration) (ablation)GeoCalib · Stanford2D3DFoV median error (deg): 3.21 -> 2.30Improved
jointly optimize intrinsics over multiple images from the same camera (ablation)GeoCalib · TartanAirAUC of FoV and gravityImproved
train GeoCalib on distorted images rather than pinhole only (ablation)GeoCalib · MegaDepth distortionpixel distortion recall @5px: 72.01 -> 75.57Improved
add estimated gravity prior in RANSAC and refinementSP+SG localization · InLoc DUC1pose recall @0.25m,10deg: 43.4 -> 44.9Improved

Later papers that compared with this method

3 later accepted papers report GeoCalib as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy