AnyUp: Universal Feature Upsampling

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wimmer, Thomas, Truong, Prune, Rakotosaona, Marie-Julie, Oechsle, Michael, Tombari, Federico, Schiele, Bernt, Lenssen, Jan Eric
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911450498334720
author Wimmer, Thomas
Truong, Prune
Rakotosaona, Marie-Julie
Oechsle, Michael
Tombari, Federico
Schiele, Bernt
Lenssen, Jan Eric
author_facet Wimmer, Thomas
Truong, Prune
Rakotosaona, Marie-Julie
Oechsle, Michael
Tombari, Federico
Schiele, Bernt
Lenssen, Jan Eric
contents We introduce AnyUp, a method for feature upsampling that can be applied to any vision feature at any resolution, without encoder-specific training. Existing learning-based upsamplers for features like DINO or CLIP need to be re-trained for every feature extractor and thus do not generalize to different feature types at inference time. In this work, we propose an inference-time feature-agnostic upsampling architecture to alleviate this limitation and improve upsampling quality. In our experiments, AnyUp sets a new state of the art for upsampled features, generalizes to different feature types, and preserves feature semantics while being efficient and easy to apply to a wide range of downstream tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2510_12764
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AnyUp: Universal Feature Upsampling
Wimmer, Thomas
Truong, Prune
Rakotosaona, Marie-Julie
Oechsle, Michael
Tombari, Federico
Schiele, Bernt
Lenssen, Jan Eric
Computer Vision and Pattern Recognition
Machine Learning
We introduce AnyUp, a method for feature upsampling that can be applied to any vision feature at any resolution, without encoder-specific training. Existing learning-based upsamplers for features like DINO or CLIP need to be re-trained for every feature extractor and thus do not generalize to different feature types at inference time. In this work, we propose an inference-time feature-agnostic upsampling architecture to alleviate this limitation and improve upsampling quality. In our experiments, AnyUp sets a new state of the art for upsampled features, generalizes to different feature types, and preserves feature semantics while being efficient and easy to apply to a wide range of downstream tasks.
title AnyUp: Universal Feature Upsampling
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2510.12764