SkyNative: A Native Multimodal Framework for Remote Sensing Visual Evidence Reasoning

2026-05-18

AI Summary

This work addresses the limitations of existing remote sensing vision-language models, which rely on pretrained visual encoders and are prone to language priors that obscure fine-grained visual evidence. To overcome this, the authors propose a native multimodal framework that eschews conventional visual backbones and instead tokenizes remote sensing images directly into raw image patches. Within a unified autoregressive architecture, they introduce a modality-aware disentanglement mechanism to enable deep fusion of visual and textual representations. The approach facilitates native patch-level modeling and is accompanied by a newly constructed benchmark for evaluating visual grounding capabilities. Experiments demonstrate that the model significantly enhances image grounding performance on both standard remote sensing understanding tasks and large-scale spatial reasoning scenarios, while exhibiting greater robustness against misleading textual prompts.

Abstract

Remote sensing vision-language models commonly rely on pretrained visual encoders to convert images into semantic features before language-model reasoning. While effective for scene-level understanding, this pipeline may prematurely compress local visual evidence, making fine-grained spatial reasoning vulnerable to language priors, especially in ultra-high-resolution remote sensing imagery. We present SkyNative, a native multimodal framework for remote sensing that adopts an encoder-free architecture, removing the pretrained visual backbone to directly represent images as raw patch tokens in the language-model token space. To reconcile low-level visual patches with textual tokens, SkyNative introduces a modality-aware decoupling mechanism that uses modality-specific parameters within a unified autoregressive backbone. We further introduce a visual reliance benchmark that diagnoses whether models ground their answers in image evidence through progressive visual degradation and misleading textual prompts. Across standard remote sensing understanding tasks and large-format spatial reasoning evaluations, SkyNative shows stronger image-grounded perception and improved robustness against prompt-induced language priors. These results suggest that native patch-level multimodal modeling is a promising direction for reliable remote sensing vision-language reasoning.

Problem

• remote sensing
• vision-language models
• visual evidence reasoning
• language priors
• spatial reasoning

Innovation

• native multimodal modeling
• encoder-free architecture
• modality-aware decoupling
• visual reliance benchmark
• remote sensing vision-language reasoning