RingMOSS: A Comprehensive Multi-Modal Pre- Training Dataset

Citation Author(s):: hanbo
bi
Aerospace Information Research Institute
Submitted by:: Hanbo Bi
Last updated:: Fri, 03/28/2025 - 08:54
DOI:: 10.21227/2pht-hq54
Data Format:: *.avi; *.csv; *.txt; *.zip; *.xls;
License:: Creative Commons Attribution

11 Views

Categories:: Artificial Intelligence
Keywords:: remote sensing; multi-modal; pre-training

0 ratings - Please login to submit your rating.

ACCESS DATASET CITE

Abstract

To establish a versatile RSFM adaptable to diverse tasks, RingMoE requires a comprehensive and diverse pre-training dataset that accounts for significant variations in imaging modalities, spatial resolutions, temporal dynamics, geographic regions, and scene complexities. To meet this challenge, we curate RingMOSS, a large-scale multi-modal RS dataset comprising 400 million images from nine satellite platforms, covering a broad spectrum of Earth observation scenarios.

Instructions:

To establish a versatile RSFM adaptable to diverse tasks, RingMoE requires a comprehensive and diverse pre-training dataset that accounts for significant variations in imaging modalities, spatial resolutions, temporal dynamics, geographic regions, and scene complexities. To meet this challenge, we curate RingMOSS, a large-scale multi-modal RS dataset comprising 400 million images from nine satellite platforms, covering a broad spectrum of Earth observation scenarios.