SPAR: Single-Pass Any-Resolution ViT for Open-vocabulary Segmentation

TL;DR

SPAR is a novel ViT-based dense feature extractor that enables efficient high-resolution inference for open-vocabulary segmentation, surpassing existing methods in accuracy and computational efficiency.

Contribution

Introduces SPAR, a resolution-agnostic ViT that distills fine-grained spatial reasoning into a single pass without architectural changes or pixel supervision.

Findings

SPAR improves single-pass segmentation accuracy by up to 10.5 mIoU.

SPAR surpasses the teacher model in open-vocabulary segmentation tasks.

Efficient high-resolution reasoning achieved without architectural modifications.

Abstract

Foundational Vision Transformers (ViTs) have limited effectiveness in tasks requiring fine-grained spatial understanding, due to their fixed pre-training resolution and inherently coarse patch-level representations. These challenges are especially pronounced in dense prediction scenarios, such as open-vocabulary segmentation with ViT-based vision-language models, where high-resolution inputs are essential for accurate pixel-level reasoning. Existing approaches typically process large-resolution images using a sliding-window strategy at the pre-training resolution. While this improves accuracy through finer strides, it comes at a significant computational cost. We introduce SPAR: Single-Pass Any-Resolution ViT, a resolution-agnostic dense feature extractor designed for efficient high-resolution inference. We distill the spatial reasoning capabilities of a finely-strided, sliding-window teacher into a single-pass student using a feature regression loss, without requiring architectural changes or pixel-level supervision. Applied to open-vocabulary segmentation, SPAR improves single-pass baselines by up to 10.5 mIoU and even surpasses the teacher, demonstrating effectiveness in efficient, high-resolution reasoning. Code: https://github.com/naomikombol/SPAR