GECO: Geometrically Consistent Embedding with Lightspeed Inference

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hartwig, Regine, Muhle, Dominik, Marin, Riccardo, Cremers, Daniel
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912515197239296
author Hartwig, Regine
Muhle, Dominik
Marin, Riccardo
Cremers, Daniel
author_facet Hartwig, Regine
Muhle, Dominik
Marin, Riccardo
Cremers, Daniel
contents Recent advances in feature learning have shown that self-supervised vision foundation models can capture semantic correspondences but often lack awareness of underlying 3D geometry. GECO addresses this gap by producing geometrically coherent features that semantically distinguish parts based on geometry (e.g., left/right eyes, front/back legs). We propose a training framework based on optimal transport, enabling supervision beyond keypoints, even under occlusions and disocclusions. With a lightweight architecture, GECO runs at 30 fps, 98.2% faster than prior methods, while achieving state-of-the-art performance on PFPascal, APK, and CUB, improving PCK by 6.0%, 6.2%, and 4.1%, respectively. Finally, we show that PCK alone is insufficient to capture geometric quality and introduce new metrics and insights for more geometry-aware feature learning. Link to project page: https://reginehartwig.github.io/publications/geco/
format Preprint
id arxiv_https___arxiv_org_abs_2508_00746
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle GECO: Geometrically Consistent Embedding with Lightspeed Inference
Hartwig, Regine
Muhle, Dominik
Marin, Riccardo
Cremers, Daniel
Computer Vision and Pattern Recognition
Recent advances in feature learning have shown that self-supervised vision foundation models can capture semantic correspondences but often lack awareness of underlying 3D geometry. GECO addresses this gap by producing geometrically coherent features that semantically distinguish parts based on geometry (e.g., left/right eyes, front/back legs). We propose a training framework based on optimal transport, enabling supervision beyond keypoints, even under occlusions and disocclusions. With a lightweight architecture, GECO runs at 30 fps, 98.2% faster than prior methods, while achieving state-of-the-art performance on PFPascal, APK, and CUB, improving PCK by 6.0%, 6.2%, and 4.1%, respectively. Finally, we show that PCK alone is insufficient to capture geometric quality and introduce new metrics and insights for more geometry-aware feature learning. Link to project page: https://reginehartwig.github.io/publications/geco/
title GECO: Geometrically Consistent Embedding with Lightspeed Inference
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.00746