Stable Single-Pixel Contrastive Learning for Semantic and Geometric Tasks
By: Leonid Pogorelyuk , Niels Bracher , Aaron Verkleeren and more
Potential Business Impact:
Helps computers understand images by matching points.
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.
Similar Papers
Compression Beyond Pixels: Semantic Compression with Multimodal Foundation Models
CV and Pattern Recognition
Makes pictures smaller, keeping their meaning.
Sample-Centric Multi-Task Learning for Detection and Segmentation of Industrial Surface Defects
CV and Pattern Recognition
Finds tiny flaws on factory products.
Context-Enriched Contrastive Loss: Enhancing Presentation of Inherent Sample Connections in Contrastive Learning Framework
CV and Pattern Recognition
Teaches computers to learn better from pictures.