MMEarth: Exploring Multi-Modal Pretext Tasks For Geospatial Representation Learning

arXiv:2405.02771v1 Announce Type: cross
Abstract: The volume of unlabelled Earth observation (EO) data is huge, but many important applications lack labelled training data. However, EO data offers the unique opportunity to pair data from different modalities and sensors automatically based on geographic location and time, at virtually no human labor cost. We seize this opportunity to create a diverse multi-modal pretraining dataset at global scale. Using this new corpus of 1.2 million locations, we propose a Multi-Pretext Masked Autoencoder (MP-MAE) …

abstract applications arxiv cost cs.ai cs.cv cs.lg data earth earth observation geospatial however human labor location modal multi-modal observation representation representation learning sensors tasks training training data type unique