JAFAR: Jack up Any Feature at Any Resolution

TL;DR

JAFAR is a lightweight, attention-based feature upsampler that enhances the resolution of visual features from foundation vision encoders, improving detail recovery and outperforming existing methods across various tasks.

Contribution

Introduces JAFAR, a flexible and efficient upsampling module using attention and SFT modulation, capable of generalizing to higher resolutions without high-res supervision.

Findings

JAFAR outperforms existing upsampling methods in multiple downstream tasks.

It effectively recovers fine-grained spatial details.

JAFAR generalizes well to higher output scales without high-resolution training.

Abstract

Foundation Vision Encoders have become essential for a wide range of dense vision tasks. However, their low-resolution spatial feature outputs necessitate feature upsampling to produce the high-resolution modalities required for downstream tasks. In this work, we introduce JAFAR, a lightweight and flexible feature upsampler that enhances the spatial resolution of visual features from any Foundation Vision Encoder to an arbitrary target resolution. JAFAR employs an attention-based module designed to promote semantic alignment between high-resolution queries, derived from low-level image features, and semantically enriched low-resolution keys, using Spatial Feature Transform (SFT) modulation. Notably, despite the absence of high-resolution supervision, we demonstrate that learning at low upsampling ratios and resolutions generalizes remarkably well to significantly higher output scales. Extensive experiments show that JAFAR effectively recovers fine-grained spatial details and consistently outperforms existing feature upsampling methods across a diverse set of downstream tasks. Project page at https://jafar-upsampler.github.io