AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

GECO: Geometrically Consistent Embedding with Lightspeed Inference

Regine Hartwig, Dominik Muhle, Riccardo Marin, Daniel Cremers

ICCV 2025 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning · 2 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 argmax-based supervision with optimal transport soft assignment loss with dustbinDINOv2-B + LoRA · PFPascalPCK@0.1: 86.0 -> 92.1 (DINOv2-B vs GECO)Improved
use lightweight DINOv2 plus LoRA instead of diffusion-feature based GeoGECO · APK inference timetime (ms): 2127 -> 40Improved
use GECO rather than Geo on SPairGECO · SPair testPCK@0.1: 90.1 -> 85.2 (Geo vs GECO)Hurt
use GECO on CUB versus the DINOv2-B backboneGECO · CUB (unseen in training)PCK@0.1: 92.8 -> 92.5No clear effect
scale backbone from DINOv2-S to DINOv2-B (ablation)GECO · SPair testPCK@0.1: 78.0 -> 85.2Improved
use mask-based marginals instead of standard SuperGlue marginals (ablation)GECO · CUB (unseen in training)PCK@0.1: 89.5 -> 92.5Improved
use KL-regularized soft marginals instead of hard marginal constraints (ablation)GECO · PascalParts segmentationmean mIoU: 36.4 -> 37.9Improved
sample fewer image pairs per category (K 100-400) when jointly training (ablation)GECO · PFPascal, SPair, CUBPCKHurt
apply GECO to preserve pretrained features versus GeoGECO · CUB (unseen in training)PCK@0.1: 88.4 -> 92.5Improved
use GECO features for part segmentation versus DINOv2-B overallGECO · PascalParts segmentationmean mIoU: 39.8 -> 37.9Hurt

Later papers that compared with this method

1 later accepted paper reports GECO as a baseline in their results tables; 1 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