Stable Single-Pixel Contrastive Learning for Semantic and Geometric Tasks

Leonid Pogorelyuk, Niels Leif Bracher, Aaron Verkleeren, Lars Kühmichel, Stefan T. Radev

Track: Extended Abstract Track

Keywords: contrastive learning, overcomplete representation, self-supervised learning, image processing, 3d representations

TL;DR: We introduce a family of stable contrastive losses that learn overcomplete, pixel-level descriptors combining semantic and geometric information without momentum-based training.

Abstract: 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 images 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.

Submission Number: 34