RemoteVAR: Autoregressive Visual Modeling for Remote Sensing Change Detection

TL;DR

RemoteVAR introduces a novel autoregressive visual model for remote sensing change detection, leveraging multi-resolution features and a specialized training strategy to outperform existing diffusion and transformer baselines.

Contribution

It presents a new VAR-based framework that improves pixel-level change detection by conditioning on fused features and employing autoregressive training tailored for this task.

Findings

Outperforms diffusion and transformer-based methods on benchmarks.

Provides a competitive autoregressive alternative for change detection.

Demonstrates significant improvements in dense prediction accuracy.

Abstract

Remote sensing change detection aims to localize and characterize scene changes between two time points and is central to applications such as environmental monitoring and disaster assessment. Meanwhile, visual autoregressive models (VARs) have recently shown impressive image generation capability, but their adoption for pixel-level discriminative tasks remains limited due to weak controllability, suboptimal dense prediction performance and exposure bias. We introduce RemoteVAR, a new VAR-based change detection framework that addresses these limitations by conditioning autoregressive prediction on multi-resolution fused bi-temporal features via cross-attention, and by employing an autoregressive training strategy designed specifically for change map prediction. Extensive experiments on standard change detection benchmarks show that RemoteVAR delivers consistent and significant improvements over strong diffusion-based and transformer-based baselines, establishing a competitive autoregressive alternative for remote sensing change detection. Code will be available \href{https://github.com/yilmazkorkmaz1/RemoteVAR}{\underline{here}}.