GeoAlignCLIP: Enhancing Fine-Grained Vision-Language Alignment in Remote Sensing via Multi-Granular Consistency Learning

TL;DR

GeoAlignCLIP introduces a multi-granular semantic alignment framework for remote sensing, significantly improving fine-grained vision-language understanding and outperforming existing methods on various benchmarks.

Contribution

It presents a novel unified framework for fine-grained alignment in remote sensing, incorporating multi-granular semantic learning and intra-modal consistency, along with a new hierarchical dataset.

Findings

Outperforms existing RS-specific methods across multiple benchmarks.

Achieves more precise visual-semantic alignment at region and concept levels.

Demonstrates robustness and accuracy in complex, fine-grained tasks.

Abstract

Vision-language pretraining models have made significant progress in bridging remote sensing imagery with natural language. However, existing approaches often fail to effectively integrate multi-granular visual and textual information, relying primarily on global image-text alignment. This limitation hinders the model's ability to accurately capture fine-grained details in images, thus restricting its performance in complex, fine-grained tasks. To address this, we propose GeoAlignCLIP, a unified framework that achieves fine-grained alignment in remote sensing tasks by learning multi-granular semantic alignments and incorporating intra-modal consistency, enabling more precise visual-semantic alignment between image regions and text concepts. Additionally, we construct RSFG-100k, a fine-granular remote sensing dataset containing scene descriptions, region-level annotations, and challenging hard-negative samples, providing hierarchical supervision for model training. Extensive experiments conducted on multiple public remote-sensing benchmarks demonstrate that GeoAlignCLIP consistently outperforms existing RS-specific methods across diverse tasks, exhibiting more robust and accurate fine-grained vision-language alignment.