New framework enables remote sensing models to adapt to scale variations

Researchers have developed ScaleEarth, a novel framework for remote sensing vision-language models (RS-VLMs) that addresses the challenge of varying ground sampling distances (GSDs). Unlike previous methods that treat GSD as a discrete token, ScaleEarth uses a continuous conditioning variable to dynamically adjust the model's computation path based on physical scale. This approach, implemented with CS-HLoRA and SSE-U for GSD prediction, achieves state-of-the-art results on remote sensing benchmarks. AI

IMPACT Introduces a new method for handling scale variations in remote sensing data, potentially improving performance on Earth-system tasks.

RANK_REASON The cluster contains an academic paper detailing a new method and framework for improving remote sensing vision-language models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• CS-HLoRA
• GeoScale-VQA
• OmniEarth-Bench
• QLoRA
• Qwen3-VL
• ScaleEarth
• XLRS-Bench

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →