Announcement Anyup

New pre-print: “AnyUp: Universal Feature Upsampling” is now available on arXiv! Super excited to share this work, where we propose a first-of-its-kind feature-agnostic upsampling architecture that can upsample features from any vision model at any resolution, without requiring any encoder-specific training. New state-of-the-art results on multiple downstream benchmarks, while being the first upsampler that naturally generalizes to different feature types at inference time.