Is Self-Supervised Pre-training on Satellite Imagery Better than ImageNet? A Systematic Study with Sentinel-2
This addresses the problem of optimizing pre-training strategies for remote sensing applications, indicating incremental insights for researchers and practitioners in the field.
The study investigated whether self-supervised pre-training on satellite imagery (Sentinel-2) outperforms ImageNet pre-training for remote sensing tasks, finding that it offers only modest improvements and remains competitive, suggesting the benefits may be overstated.
Self-supervised learning (SSL) has demonstrated significant potential in pre-training robust models with limited labeled data, making it particularly valuable for remote sensing (RS) tasks. A common assumption is that pre-training on domain-aligned data provides maximal benefits on downstream tasks, particularly when compared to ImageNet-pretraining (INP). In this work, we investigate this assumption by collecting GeoNet, a large and diverse dataset of global optical Sentinel-2 imagery, and pre-training SwAV and MAE on both GeoNet and ImageNet. Evaluating these models on six downstream tasks in the few-shot setting reveals that SSL pre-training on RS data offers modest performance improvements over INP, and that it remains competitive in multiple scenarios. This indicates that the presumed benefits of SSL pre-training on RS data may be overstated, and the additional costs of data curation and pre-training could be unjustified.