AI Paper Atlas

Research map / 3D vision / Depth and stereo

ACE-G: Improving Generalization of Scene Coordinate Regression Through Query Pre-Training

Leonard Bruns, Axel Barroso-Laguna, Tommaso Cavallari, Aron Monszpart, Sowmya Munukutla, Victor Adrian Prisacariu, Eric Brachmann

ICCV 2025 · 3D vision › Depth and stereo · approach: Scene coordinate regression for localization · 10 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
replace ACE encoder with DINOv2 features (ablation)ACE · Indoor-6% localized: 70 -> 91Improved
add query pre-training of scene-agnostic regressor (ACE-G vs DINO-ACE) (ablation)DINO-ACE · Indoor-6% localized: 91 -> 95Improved
add query pre-training (ACE-G vs DINO-ACE) (ablation)DINO-ACE · RIO10median translation error (cm): 83.8 -> 41.1Improved
add query pre-training (ACE-G vs DINO-ACE) (ablation)DINO-ACE · Cambridge Landmarksmedian translation error (cm): 43.9 -> 35.1Improved
add query pre-training with ACE encoder (ablation)ACE · RIO10median position error (cm): 358.4 -> 144.5Improved
add query pre-training with ACE encoder (ablation)ACE · 7Scenesmedian position error (cm): 1.1 -> 1.3No clear effect
replace ACE encoder with DINOv2 encoder (ablation)ACE-G · 7Scenes (static)median position error (cm): 1.3 -> 4.6Hurt
add query iterations during pre-training (ablation)ACE-G · validation splitsaccuracy (20deg,20cm)Improved
add uncertainty-based quantile prefiltering of scene coordinates (ablation)ACE-G · registrationpose accuracyImproved
use ACE-G instead of ACE at fine thresholdACE · Cambridge (10deg/10cm)% localized: 26.3 -> 17.9Hurt
use ACE-G (5 min) instead of ACEACE · Indoor-6 (5deg/5cm)% localized: 33.5 -> 54.7Improved
fine-tune on indoor datasets only (ablation)ACE-G · SN, SN++, ARK vs MF, BMVG, WRaccuracy (20deg,20cm): +1.9 / +2.6 / +1.2 on SN, SN++, ARK; -3.4 / -7.5 / -9.2 on MF, BMVG, WRMixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 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