RaysUp Framework for Geometry-Aware Feature Upsampling
June 21, 2026
RaysUp is an ultra-lightweight, task-agnostic framework that reconstructs high-resolution feature maps from low-resolution Vision Foundation Model outputs. It uses geometry-aware ray representation to lift feature reconstruction to arbitrary resolutions without requiring VFM-specific retraining.
HOW THIS AFFECTS YOU
●
builderYou can achieve pixel-level reasoning from low-resolution vision models without the overhead of heavy attention-based upsamplers.
●
researcherThis offers a way to upgrade VFM outputs to higher resolutions while maintaining semantic fidelity.