AI Paper Atlas

Research map / 3D vision / Depth and stereo

CoLoR: The Devil is in Scene Coordinate Regression for Large-Scale Visual Localization

Xindong Mao, Hang Li, Yuchen Wu, Jiahe Li, Xiao Bai, Jin Zheng

CVPR 2026 · 3D vision › Depth and stereo · approach: Scene coordinate regression for localization

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
partition points and train only on multi-view points (pixel partition) (ablation)R-SCoRe baseline · Hyundai Dept. 4F testaccuracy at (0.1 m, 1 deg) (%): 60.2 -> 70.2Improved
replace single-view reprojection loss with multi-view reprojection supervision (ablation)CoLoR stage one · Hyundai Dept. 4F testaccuracy at (0.1 m, 1 deg) (%): 70.2 -> 75.8Improved
add pseudo-depth supervision for single-view points in stage two (ablation)CoLoR · Hyundai Dept. 4F testaccuracy at (0.1 m, 1 deg) (%): 75.8 -> 76.6Improved
add pseudo-depth supervision (coarser thresholds) (ablation)CoLoR · Hyundai Dept. 4F testaccuracy at (0.25 m, 2 deg) (%): 86.7 -> 86.7No clear effect
replace matching-oriented local features with pixel-level retrieval features (ablation)CoLoR · Hyundai Dept. 4F testaccuracy at (0.1 m, 1 deg) (%): 76.6 -> 80.9Improved
concatenate matching-oriented local features with global features (ablation)global retrieval feature · Aachen Day-Night PR curveaverage precision: AP 0.924 -> 0.816Hurt
retrain local features with pixel-level retrieval (MoCo contrastive) objective (ablation)concatenated global+local feature · Aachen Day-Night PR curveaverage precision: AP 0.816 -> 0.926Improved
use CoLoR instead of R-SCoReSCR large-scale localization · Hyundai Dept. 4F testaccuracy at (0.1 m, 1 deg) (%): 60.2 -> 80.9Improved
use CoLoR instead of R-SCoRe at medium threshold on Aachen nightSCR localization · Aachen Nightaccuracy at (0.5 m, 5 deg) (%): 89.8 -> 88.8Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

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