Saved in:
Bibliographic Details
Main Authors: Pogorelyuk, Leonid, Bracher, Niels, Verkleeren, Aaron, Kühmichel, Lars, Radev, Stefan T.
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.04970
Tags: Add Tag
No Tags, Be the first to tag this record!
Table of Contents:
  • We pilot a family of stable contrastive losses for learning pixel-level representations that jointly capture semantic and geometric information. Our approach maps each pixel of an image to an overcomplete descriptor that is both view-invariant and semantically meaningful. It enables precise point-correspondence across images without requiring momentum-based teacher-student training. Two experiments in synthetic 2D and 3D environments demonstrate the properties of our loss and the resulting overcomplete representations.